Three-dimensional (3D) head pose prediction for automotive systems and applications

Abstract

The disclosure relates to three-dimensional (3D) head pose prediction for automotive systems and applications. In various examples, head pose prediction for automobile occupant sensing systems and applications is presented. Systems and methods described herein provide a machine learning model trained using a dataset that includes truth-valued head pose data calculated using a registered head model of a training subject. While operating the vehicle, one or more cameras and depth sensors capture synchronized images of the training subject. To calculate a true-valued 3D head pose, an angular deviation between the 3D point cloud and the registered head model may be calculated to obtain a 3D true-valued head pose measurement. Head attitude measurements may be mapped into a sensor coordinate system using external reference calibration transformations. A training sample may be generated for training a machine learning model, the training sample including an optical image frame and a head pose measurement transposed into a frame of reference of the optical image frame.

Description

Background Art

[0001] Human posture monitoring is a component of advanced driver assistance systems (ADAS), which aim to improve the safety and comfort of the driver and other vehicle occupants. By detecting the driver's body posture and orientation based on sensor data, ADAS can infer the driver's attention, intentions, and emotions and provide appropriate feedback or intervention. A specific aspect of human posture monitoring is head pose recognition. Head pose recognition can be used by ADAS to warn the driver of potential hazards, adjust vehicle settings based on the driver's preferences, enhance the driver's experience through interactive features, and / or be used in conjunction with estimating the occupant's gaze direction. Summary of the Invention

[0002] Embodiments of the present disclosure relate to head pose prediction for vehicle occupant sensing systems and applications. The disclosed systems and methods provide occupant head pose measurement data of a vehicle occupant that can be inferred from optical image data using a machine learning model.

[0003] Compared to existing head pose inference techniques, the systems and methods described herein can provide a machine learning model trained using a dataset that includes ground truth head pose data calculated based on a personalized optimized head model of a vehicle occupant training subject. A registered head model can be generated based on depth data of the training subject's head and face collected while the training subject is seated in a vehicle from which the training dataset for training the machine learning model is captured. In some embodiments, the registration process can obtain a three-dimensional (3D) point cloud representation of the generic head model and a 3D point cloud representation of the training subject. The depth data captured during the registration process can be used to optimize the generic head model to personalize the model to reflect the characteristics of the training subject to generate a registered head model.

[0004] Once a registered head model has been established for a training subject, a dataset for training a head pose inference machine learning model can be collected while the training subject is operating a vehicle (e.g., by driving the vehicle on a public road for a period of time). During this driving phase, one or more cameras and depth sensors capture synchronized images of the training subject's head and face. A pose dataset collection system obtains a 3D point cloud of the training subject's head and one or more synchronized captured camera images. To calculate the true 3D head pose, the pose dataset collection system can compare and calculate the angular deviation between the head pose represented by the 3D point cloud and the registered head model. The 3D point cloud can then be registered with the 3D point cloud of the registered head model using a geometric registration matching algorithm to calculate a transformation between the two 3D point clouds, which represents a measurement of the head pose appearing in the depth image. The calculated head pose measurement indicates the head pose measurement of the head pose appearing in the time-synchronized camera image frames. The pose dataset collection system can then transform the 3D ground-truth head pose measurements to the sensor coordinate reference frame of the image frames captured by each camera using an extrinsic calibration transform between the sensors. A training sample for training a head pose inference machine learning model can be generated, comprising an optical image frame and the 3D ground-truth head pose measurements transposed into the reference frame of the optical image frame. The 3D ground-truth head pose measurements are transformed to the viewpoint of the camera that captured the images in the training sample so that the machine learning model can be trained based on training subjects trained to accurately predict (infer) the 3D head pose relative to the viewpoint represented by the captured image. Thus, in a production vehicle, the trained head pose inference machine learning model can be fed an image from an occupant monitoring system (OMS) and / or driver monitoring system (DMS) camera and predict the 3D head pose represented by the image without being programmed with information about the precise position and / or orientation of the OMS / DMS camera relative to the vehicle interior and / or vehicle occupants being monitored. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] The present system and method for head posture prediction in automotive occupant sensing systems and applications are described in detail below with reference to the accompanying drawings, wherein:

[0006] Figure 1 is a data flow diagram for an example head pose dataset collection system according to some embodiments of the present disclosure;

[0007] Figure 2A is a schematic diagram illustrating an example placement of data set collection sensors within an example vehicle according to some embodiments of the present disclosure;

[0008] Figure 2B is a schematic diagram illustrating an example field of view of an optical image sensor located inside an example vehicle according to some embodiments of the present disclosure;

[0009] Figure 2C is a schematic diagram illustrating example calibration parameters corresponding to extrinsic calibration between an optical image sensor and a depth sensor inside an example vehicle according to some embodiments of the present disclosure;

[0010] Figure 3 is a data flow diagram illustrating an example registration process for generating a registered head model of a training subject according to some embodiments of the present disclosure;

[0011] Figure 4 An example head pose prediction training system for training a machine learning model to infer head pose predictions according to some embodiments of the present disclosure is shown;

[0012] Figure 5 is a data flow diagram illustrating an example head pose prediction system including a head pose prediction model according to some embodiments of the present disclosure;

[0013] Figure 6 is a flowchart illustrating a method for training a head pose prediction model according to some embodiments of the present disclosure;

[0014] Figure 7 is a flowchart illustrating a method for head pose prediction according to some embodiments of the present disclosure;

[0015] Figure 8A is an illustration of an example autonomous vehicle according to some embodiments of the present disclosure;

[0016] Figure 8B According to some embodiments of the present disclosure Figure 8A Examples of camera positions and fields of view for autonomous vehicles;

[0017] Figure 8C According to some embodiments of the present disclosure Figure 8A a block diagram of an example system architecture for an example autonomous vehicle;

[0018] Figure 8D According to some embodiments of the present disclosure, a method for Figure 8A System diagram of an example of communication between autonomous vehicles;

[0019] Figure 9 is a block diagram of an example computing device suitable for implementing some embodiments of the present disclosure; and

[0020] Figure 10is a block diagram of an example data center suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0021] Systems and methods are disclosed relating to head pose prediction for automotive occupant sensing systems and applications. Although the present disclosure may relate to an example autonomous or semi-autonomous vehicle or machine 800 (referred to herein alternatively as "vehicle 800" or "ego-machine 800"), the examples relate to Figures 8A-8D For example, the systems and methods described herein may be used with, but are not limited to, non-autonomous vehicles or machines, semi-autonomous vehicles or machines (e.g., in one or more advanced driver assistance systems (ADAS)), autonomous vehicles or machines, manned and unmanned robots or robotic platforms, warehouse vehicles, off-road vehicles, vehicles connected to one or more trailers, aircraft, ships, space shuttles, emergency response vehicles, motorcycles, electric or motorized bicycles, airplanes, construction vehicles, trains, underwater vehicles, remotely controlled vehicles (e.g., drones), and / or other vehicle types. Furthermore, while the present disclosure may be described with respect to vehicle occupant monitoring for advanced driver assistance systems, this is not intended to be limiting, and the systems and methods described herein may be used in augmented reality, virtual reality, mixed reality, robotics, security and surveillance, autonomous or semi-autonomous machine applications, and / or any other technology space where human posture monitoring may be used.

[0022] Head pose recognition is typically implemented using deep neural networks that are trained using large-scale datasets. However, collecting accurate three-dimensional (3D) ground-truth pose data for training datasets is physically challenging and can be prohibitively expensive.

[0023] One example of current head pose inference technology involves generating a dataset of training data for a machine learning model using a training subject seated in a vehicle and wearing a set of inertial measurement sensors on their head (e.g., using inertial sensors attached to a hat worn by the test subject). Images of the training subject are captured while the inertial measurement sensors obtain three-dimensional rotational measurements of the training subject's head pose (e.g., rotation angles for yaw, pitch, and roll) while the training subject operates the vehicle. The inertial measurement sensor data can be used as ground truth when training a machine learning model to infer head pose based on the captured image data. While the inertial measurement sensors themselves provide highly accurate rotational data, several factors can introduce errors that affect the accuracy of the sensor data as ground truth. For example, errors can be introduced by linear offsets between the rotational axes of the inertial measurement sensors and the distribution of rotational axes within the human neck associated with human head movement. Furthermore, by training the machine learning model using images of the training subject seated in a vehicle and wearing inertial measurement sensors, the machine learning model is trained on images of occupants who inherently have different appearances than typical vehicle occupants who would not typically be expected to wear sensors on their heads. Furthermore, the additional weight of the inertial measurement sensors worn by the test subjects may cause the test subjects to orient their heads differently than they would in natural driving conditions, thereby introducing another source of bias when using the inertial measurement sensor measurements as ground truth.

[0024] Another example of current head pose inference technology involves generating a dataset of machine learning model training data using a training subject sitting in a vehicle while attempting to capture a 3D image of the training subject's head pose and estimate the three-dimensional deviation of the head pose rotation from a predefined original head pose orientation based on the change in depth. Common technologies used to capture depth perception data include stereo vision and time-of-flight cameras. Stereo vision methods have challenges in terms of field of view and have limitations because they are affected by occlusions. Time-of-flight sensors have the advantage of providing point cloud data from a single viewpoint, but typically have lower resolution and are limited in operation under strong light.

[0025] Compared to these existing head pose inference techniques, the systems and methods described herein can provide a machine learning model trained using a dataset that includes ground truth head pose data calculated based on a personalized optimized head model of a vehicle occupant training subject (referred to herein as a registered head model), which is generated using depth data captured from the training subject during a registration process. The registered head model can be generated based on depth data of the training subject's head and face collected while the training subject is seated in a vehicle from which the training dataset for training the head pose inference machine learning model is captured. The depth sensor can include, for example, a stereo camera depth sensor, a time-of-flight camera depth sensor, a millimeter-wave radar sensor, a Kinect sensor, or other depth sensor.

[0026] In some embodiments, during the registration process, the training subject may occupy a car seat of a vehicle and be instructed to perform predefined head maneuvers to cover multiple perspectives of their head while capturing depth data using a depth sensor. In some embodiments, feedback may be provided to the training subject (e.g., visual guidance and / or through auditory / voice instructions) instructing them on how to rotate their head in order to capture a specified range of perspectives. The predefined head maneuvers may include rotations of the head in three directions, including rotations in roll, pitch, and / or yaw. In some embodiments, the depth data captured during the registration process may be used to optimize a generic head model to personalize the model to reflect the characteristics of the training subject. This optimization results in the generation of a registered head model.

[0027] In some embodiments, the registration process can begin with a 3D point cloud representation of a generic head model and personalize the generic head model using a 3D point cloud representation of a depth image of a training subject obtained from a depth sensor. For example, a rigid model transformation between the point clouds can be obtained by finding closest point correspondences using an iterative closest point (ICP) registration algorithm, which can be used to align the point clouds in 3D space by iteratively minimizing the distance between corresponding points in the clouds. In some embodiments, a least-squares optimization iteratively estimates updates to minimize the error between the point clouds from the generic head model and the depth image. The optimization results in a set of coefficients that represent a personalized head model and are unique to the training subject performing the registration process. Furthermore, during the registration process, head model transformations can be applied in real time to determine the head pose of the training subject in the current depth image frame. The head pose can be displayed in a rotatable overlay of the head model that tracks the pose of the training subject's head. As the training subject adopts various prescribed head poses, the registration process results in real-time iterative optimization of the training subject's facial features. This real-time display of the transformation of the generic head model into the registered head model makes the registration process interactive, wherein the training subject can visually observe that the displayed head model fits better and better and adapts to their facial characteristics. Thus, the generic head model adapts to the facial features of the training subject based on the depth data from the depth images captured during the registration process.

[0028] With a registered head model established for a training subject, a dataset for training a head pose inference machine learning model can be collected while the training subject operates a vehicle (e.g., by driving the vehicle on public roads for a period of time). More specifically, while the training subject operates the vehicle, one or more cameras of the vehicle's occupant monitoring system (OMS) and / or driver monitoring system (DMS) can be used to capture images of the training subject's head and face, while a depth sensor captures depth images of the training subject. The cameras can include, for example, red, green, and blue (RGB) cameras, infrared (IR) cameras, RGB-IR cameras, or other types of cameras that can capture image frames of the training subject's head and face. While driving the vehicle, the training subject's head will naturally turn and rotate as they perform various driving tasks (e.g., monitoring traffic and directing their gaze toward various vehicle controls and / or displays).

[0029] During this driving phase, one or more cameras and depth sensors capture synchronized images of the training subject's head and face. A pose dataset collection system (which can be implemented as an application executed using onboard processing resources) obtains a 3D point cloud of the training subject's head (e.g., from 3D data captured by the depth sensor) and one or more synchronized captured camera images (e.g., from one or more OMS and / or DMS cameras) such that both the 3D point cloud and the camera images capture the same head pose position at the same time—both captured from the training subject's specific perspective. To compute the ground truth 3D head pose, the pose dataset collection system can compare and calculate the angular deviation between the head pose represented by the 3D point cloud and a registered (e.g., personalized) head model. Because the baseline provided by the registered head model has been individually optimized for the training subject's head and facial features, the angular deviation can be calculated with higher accuracy than comparing the 3D point cloud to a generic head model—thereby providing a more accurate 3D head pose calculation.

[0030] In some embodiments, once synchronized frames of camera and depth images are obtained, a 3D point cloud is generated and outliers are removed. For example, the pose dataset collection system can implement a feature detection function that generates a tight bounding box around the face and head and excludes points of the 3D point cloud that fall outside the bounding box. The 3D point cloud can then be registered against the 3D point cloud of the registered head model using a geometric registration matching algorithm (e.g., an ICP registration algorithm) to calculate a transformation between the two 3D point clouds, which represents the measurement results of the head pose appearing in the depth image (e.g., with respect to roll, pitch, and / or yaw angular displacement). By using external parameters to calibrate the transformation between the sensors, the pose dataset collection system can then convert the 3D head pose measurements into sensor coordinates of the reference frame of the image frames captured by each camera. The calculated head pose measurements indicate the head pose measurements of the head pose appearing in the time-synchronized camera image frames. In some embodiments, the pose dataset collection system can annotate one or more camera image frames with head pose measurements derived from the deviations between the corresponding 3D point clouds.

[0031] As described above, in order for the head pose measurements to accurately represent the head pose measurements that appear in the camera image frames, the depth sensor and one or more camera sensors operate in a synchronized manner to obtain temporally synchronized image frames and point cloud data. This synchronization can present challenges, for example, where IR pulses from the depth sensor can introduce noise and / or illumination from the pulses can interfere with the images captured by one or more cameras. In some embodiments, the pose dataset collection system can use an onboard synchronization signal provided by a vehicle system to adjust image frame synchronization. For example, the OMS and / or DMS functions can be synchronized using a vehicle synchronization signal generated by common hardware or software. The pose dataset collection system can use the synchronization signal to trigger the OMS and / or DMS cameras while simultaneously triggering the depth sensor to generate infrared pulses (e.g., for generating time-of-flight depth data) in a manner synchronized with the camera offset, where the IR pulse emission is closely, but not exactly, aligned with the camera's timing. In some embodiments, the camera and depth sensor can operate in a master / proxy configuration, where the depth sensor is triggered to follow the operation of one or more cameras.

[0032] A dataset of camera-captured image frames, each annotated with transformed 3D ground-truth head pose measurements, can be used to train a head pose inference machine learning model to infer the head pose of a vehicle occupant from input camera images—without including depth data or depth images in the input. In some embodiments, a calibration transformation can be used to translate a registered head model (as oriented by the 3D pose measurements) to the camera image frames, and the translated registered head model can be used as an additional input channel along with the camera images to train the head pose inference machine learning model.

[0033] In some embodiments, the head pose inference machine learning model may include, for example, a deep neural network (DNN), a convolutional neural network (CNN), and / or other neural network architectures. To train the head pose inference machine learning model, in some embodiments, training samples from a dataset of image frames captured by a camera (annotated with converted 3D ground-truth head pose measurements) may be input into the machine learning model, and the resulting predicted head pose output from the machine learning model is compared with the converted 3D ground-truth head pose measurements. To generate the training samples, the pose dataset collection system may obtain 3D ground-truth head pose measurements (based on a 3D point cloud calculation as described above) and convert the 3D ground-truth head pose measurements into converted 3D ground-truth head pose measurements for each of one or more synchronously captured camera images generated by one or more cameras.

[0034] For example, for one embodiment, in which the depth sensor synchronously captures a depth image of the training subject while two cameras each capture an optical image frame of the training subject, this configuration can be used to generate two training samples simultaneously. The 3D point cloud from the depth image can then be evaluated against the 3D point cloud of the registered head model of the training subject, and the transformation between the two 3D point clouds obtained is used to calculate the 3D true head pose measurement result of the training subject at that moment. A first extrinsic calibration between the depth sensor and the first camera can be used to derive a first RT transformation between the depth sensor and the first camera. Based on the first RT transformation, the 3D true head pose measurement result can be converted into the frame of the first camera to produce a first converted 3D true head pose measurement result, and the first converted 3D true head pose measurement result can be used to annotate the first image of the training subject captured by the first camera. Similarly, a second extrinsic calibration between the depth sensor and the second camera can be used to derive a second RT transformation between the depth sensor and the second camera. Based on this second RT transform, the 3D ground-truth head pose measurements can be transformed into frames of a second camera to produce second transformed 3D ground-truth head pose measurements, which can be used to annotate a second image of the training subject captured by the second camera. In this way, the ground-truth head pose measurements can be extended to multiple cameras. The resulting annotated images each represent an individual training example that can be used in a training dataset to train a head pose inference machine learning model.

[0035] Because each training sample includes an image of a subject labeled with converted 3D ground-truth head pose measurements (converted to the viewpoint of the camera that captured the image), the training of the head pose inference machine learning model is invariant relative to the position of the camera sensor that captured the image. That is, the training of the head pose inference machine learning model can be performed without knowledge of the physical relationship between one or more cameras and the depth sensor. The 3D ground-truth head pose measurements are converted to the viewpoint of the camera that captured the images in the training sample so that the machine learning model can be trained to accurately predict (infer) the 3D head pose of the training subject relative to the viewpoint represented by the captured image. Therefore, in a deployed vehicle, the trained head pose inference machine learning model can input images from an OMS / DMS camera (e.g., without the need for a dedicated depth sensor) and predict the 3D head pose represented by the image without being programmed with information about the precise location and / or orientation of the OMS / DMS camera relative to the vehicle interior and / or vehicle occupants being monitored.

[0036] In some embodiments, during the development of a training system, 3D head pose predictions and 3D ground-truth head pose measurement labels can be used as inputs to a loss function to measure the difference between the predicted 3D head pose and the converted 3D ground-truth head pose measurements, and a loss range can be calculated based on the difference in accuracy representing the 3D head pose prediction. The loss score can be used to generate loss feedback to iteratively adjust the head pose inference machine learning model during training to iteratively reduce the loss score and converge to a configuration of the machine learning model that accurately produces 3D head pose predictions that match the 3D ground-truth head pose measurement labels of the training samples. In some embodiments, the converted registered head model of the training subject (e.g., converted using extrinsic calibration parameters) can be further used as an additional input channel along with the camera image to train the head pose inference machine learning model. That is, the converted registered head model can serve as a training aid by helping the machine learning model identify features of the training subject that can be used to more accurately infer head pose.

[0037] As described above, head pose data of the driver and / or occupants of a vehicle can be used as input to ADAS for OMS and / or DMS features, such as, but not limited to, driver attention monitoring, drowsiness detection, gaze tracking, augmented reality displays, and / or enhanced hands-free interfaces, which can enhance safety and / or improve the overall driving experience. In some embodiments, in a production vehicle, image data from an onboard camera can be input to a head pose inference machine learning model, which can then output a 3D head pose prediction of the driver or other occupant. As described above, the 3D head pose prediction can provide a 3D pose estimate (e.g., including yaw, pitch, and / or roll) relative to the local reference frame of the camera that captured the image data. A vehicle system that receives the 3D head pose prediction can be programmed with extrinsic calibration parameters (e.g., an RT transform) that maps the 3D head pose prediction to another reference frame used by the vehicle system, such as a global reference frame corresponding to the physical structure of the vehicle. Thus, vehicle systems operating with 3D head pose data generated by a head pose inference machine learning model can obtain such data transformed to a global reference frame so that it can be more directly used in conjunction with other OMS / DMS sensor data captured by other sensors or generated by other onboard systems.

[0038] For example, in some embodiments, 3D head pose prediction data can be used to enhance an occupant gaze detection system. In some such embodiments, occupant gaze detection can be determined based on a combination of head pose and eye rotation. That is, the angle of rotation of the occupant's eyes (e.g., looking up, down, left, or right) combined with the head pose provides a more accurate set of data to determine what the occupant is looking at than simply considering eye rotation. As described above, the 3D head pose prediction generated by the head pose inference machine learning model based on the image of the occupant can be converted to a global reference frame based on the extrinsic calibration parameters of the camera that captured the image data. Where the eye detection algorithm is calibrated to the same global reference frame, the head pose data can be more easily applied to the detected eye rotation to calculate occupant gaze detection.

[0039] In other embodiments, a registered head model of a training subject can be used in the process of annotating facial key points in 3D point cloud depth data captured from the training subject's head. Since capturing 3D point cloud data allows for mesh representation, a mesh can be generated from the registered head model and superimposed (e.g., overlaid) on the 3D point cloud, allowing a human annotator to more easily identify the locations of facial features (key points) in the 3D point cloud and annotate them more accurately. For example, the corners of the eyes, the corners of the mouth, the corners of the nose, the endpoints of the eyebrows, and the jawline are examples of various key points that can be annotated for samples of the 3D point cloud from the training subject, so that the resulting annotated 3D point cloud data can be used as training data to train a machine learning model (e.g., to recognize emotions and / or expressions), or for other purposes.

[0040] In some embodiments, for example, during annotation, the registered head model can be aligned with the subject's 3D point cloud data (e.g., using an iterative closest point (ICP) registration algorithm) so that the mesh of the registered head model dynamically follows the 3D point cloud as the subject moves their head. Thus, as the subject moves their head to a position where at least a portion of their face is not visible from the 3D point cloud to the annotator, the annotator can annotate hidden keypoints based on the rotated mesh, referencing hidden facial features. In some embodiments, the annotator can control the rotation and / or scaling of the mesh and 3D point cloud so that they can annotate portions of the mesh that otherwise correspond to occluded areas of the face. In this way, relatively sparse 3D point cloud training samples can be augmented with annotations of occluded facial keypoints to improve the training value of the training samples.

[0041] It should be understood that the embodiments described herein may be used in the context of head pose detection in vehicles (e.g., cars, ego vehicles, trucks, trains, airplanes, spacecraft, and / or watercraft) and may be extended to other machinery, such as remotely operated and / or autonomous equipment (e.g., robots and drones) and other industrial and / or engineering machinery.

[0042] refer to Figure 1 , Figure 1 is an example data flow diagram of a posture dataset collection system 100 for training a head posture prediction machine learning model according to some embodiments of the present disclosure. It should be understood that this and other arrangements described herein are presented as examples only. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, functional groupings, etc.) may be used in addition to or in place of the arrangements and elements shown, and certain elements may be omitted entirely. In addition, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in combination with other components, and may be implemented in any suitable combination and location. The various functions performed by the entities described herein may be performed by hardware, firmware, and / or software. For example, the various functions may be performed by a processor executing instructions stored in a memory. In some embodiments, the systems, methods, and processes described herein may be implemented using hardware, firmware, and / or software. Figures 8A-8D Example of autonomous vehicle 800, Figure 9 The example computing device 900 and / or Figure 10 The example data center 1000 may be implemented with similar components, features, and / or functions.

[0043] like Figure 1As shown, system 100 includes a pose dataset collection system 120 that inputs synchronized sensor data 108 and registered head model point cloud data 114 (each corresponding to a training subject occupying vehicle 800) and generates training samples 132 of a head pose training dataset 140 from the data. Synchronized sensor data 108 may include 3D point cloud data 110 representing the training subject's head, such as captured by a depth sensor 104. Depth sensor 104 may include, for example, a stereo camera depth sensor, a time-of-flight camera depth sensor, a millimeter-wave radar sensor, a Kinect sensor, or other depth sensor, where 3D point cloud data 110 includes or is derived from the output of depth sensor 104. Synchronized sensor data 108 may include optical image data 112 (e.g., one or more image frames) representing images of the training subject's head, such as images captured by one or more optical image sensors 106. Optical image sensor 106 may include, for example, an RGB camera, an IR camera, an RGB-IR camera, or other types of cameras that can capture image frames of the training subject's head and face. In some embodiments, the depth sensor 104 and / or the one or more optical image sensors 106 may include sensors of an occupant monitoring system (OMS) and / or a driver monitoring system (DMS) of the vehicle 800, such as, but not limited to, sensors regarding Figures 8A-8D Describe the sensor and system.

[0044] like Figure 1As indicated, the depth sensor 104 and one or more optical image sensors 106 can define a set of dataset collection sensors 102, meaning the sensors operate in a synchronized manner to capture point cloud data 110 and optical image data 112 of a training subject that is synchronized in time. The pose dataset collection system 120 can use an onboard synchronization signal provided by the vehicle 800 to coordinate the synchronization of the depth sensor 104 with the one or more optical image sensors 106. For example, the OMS and / or DMS functions of the vehicle 800 can be synchronized via a vehicle synchronization signal generated by common hardware or software. The pose dataset collection system 120 can use the synchronization signal to trigger the optical image sensor 106 and simultaneously trigger the depth sensor 104 to generate infrared (IR) pulses (e.g., for generating time-of-flight depth data) based on an offset synchronization with the optical image sensor 106, wherein the emission of the IR pulses is close to, but not completely aligned with, the timing of the optical image sensor 106. In some embodiments, the optical image sensor 106 and the depth sensor 104 can operate in a master / proxy configuration, wherein the depth sensor 104 is triggered to follow the operation of the one or more optical image sensors 106. In addition, the depth sensor 104 and the one or more optical image sensors 106 of the data set collecting sensor 102 are calibrated together at least with respect to their extrinsic calibration parameters. Figure 2C As further discussed, by using an extrinsic calibration transformation (shown as calibration parameters 128) between the depth sensor 104 and each individual optical image sensor 106, the pose dataset collection system 120 can convert the 3D ground truth head pose measurements to the sensor coordinate reference frame of the image frames captured by the individual cameras.

[0045] refer to Figure 2A-2C , which illustrate a set of example data set collecting sensors 102 according to an embodiment of the present disclosure. For example, referring to Figure 2A , Figure 2A An example vehicle interior 205 (e.g., the cabin of vehicle 800) is shown at 200, and further depicted are example locations where the depth sensor 104 and the optical image sensor 106 may be placed. In this example, the optical image sensor 106 includes an optical image sensor mounted to an overhead console (shown at 210-1), an optical image sensor mounted to a rearview mirror (shown at 210-2), an optical image sensor mounted to a right structural pillar (shown at 210-3), and an optical image sensor mounted to a left structural pillar (shown at 210-4). Figure 2A Also shown is an example location of the depth sensor 104, which may be located within the vehicle interior 205, as shown, positioned to view the driver and / or one or more passengers, as shown at 220. It should be understood that as Figure 2AThe number and positions of the optical image sensors and depth sensors shown are for example purposes only, and other embodiments may use a different number of optical image sensors located in different positions. Additionally, in some embodiments, the optical image sensors 106 and depth sensors 104 may be positioned to cover viewpoints (e.g., fields of view) of other seating positions (e.g., second row seats, third row seats, etc.).

[0046] For example purposes, Figure 2B A view of the vehicle interior 205 is shown, as can be seen from Figure 2A 1 to 210-4. For example, image 215-1 depicts an image frame of a viewpoint captured by the optical image sensor 210-1 mounted to the overhead console, image 215-2 depicts an image frame of a viewpoint captured by the optical image sensor 210-2 mounted to the rearview mirror, image 215-3 depicts an image frame of a viewpoint captured by the optical image sensor 210-4 mounted to the right front pillar of the vehicle cabin, and image 215-4 depicts an image frame of a viewpoint captured by the optical image sensor 210-5 mounted to the left front pillar of the vehicle cabin.

[0047] The sensor 104 and the one or more optical image sensors 106 can define a set of data sets for capturing synchronized sensor data 108, thereby synchronously capturing representations of the head pose of the vehicle occupant from respective different viewpoints. Each of the plurality of optical image sensors 106 can be calibrated to the extrinsic parameters of the depth sensor 108 and / or a common coordinate reference system by considering extrinsic parameters representing how the respective sensors are mounted and oriented within the space of the vehicle interior 205.

[0048] like Figure 2CAs shown, for each optical image sensor 210-1, 210-2, 210-3, and 210-4, the calibration parameters 128 may include sensor-specific calibration parameters (shown as calibration parameters 230-1, 230-2, 230-3, and 230-4) that calibrate the corresponding image sensor to the depth sensor 104. By using an extrinsic calibration transformation between the sensors, the pose dataset collection system 120 can convert the 3D ground truth pose measurements to the sensor coordinate reference system of the 2D image frames captured by each camera. Parameters that affect how a 3D volume inside the vehicle looks when projected into the 2D coordinate space of a two-dimensional image frame include both extrinsic parameters and intrinsic parameters. Intrinsic parameters can refer to factors that describe the optics of the optical image sensor device, such as the optical center (also known as the principal point), focal length, tilt coefficient, field of view, and / or other parameters. The intrinsic parameters of the OMS / DMS sensor can be established during the manufacturing process. Extrinsic parameters may refer to factors that describe the physical orientation of the OMS / DMS optical image sensor, such as rotation and translation (also known as roll and pitch) and / or other parameters, and depend largely on how the sensor is mounted and oriented within the space of the vehicle cabin. Thus, a rotation-translation (RT) transform may be calculated as a function of extrinsic calibration parameters representing the translation and rotation of the depth sensor and extrinsic calibration parameters representing the translation and rotation of the image sensor. Thus, the RT transform between the depth sensor and the image camera represents an extrinsic calibration transform that maps between the local coordinate systems of the depth sensor and the image sensor so that points of the 3D point cloud can be mapped to pixel locations of the camera image frame. In some embodiments, individual extrinsic calibration parameters may be calculated between the depth sensor and each of one or more image sensors that are synchronized together to capture images of the training subject's head pose. For example, 3D pose measurements (which represent the 3D offset between the registered head model obtained during the registration process and the 3D point cloud obtained while training the subject to operate the vehicle) can be transformed from the coordinate system of the depth sensor to the coordinate system of the camera-captured image frame, and the camera-captured image frame annotated with the ground-truth pose measurements can be transformed (e.g., mapped) to the camera-captured image frame using extrinsic calibration parameters between the depth sensor and the camera. Thus, data from the 3D point cloud 110 captured by the depth sensor 104 can be mapped to the reference system of the camera image frame from the corresponding image sensor. In some embodiments, the calibration parameters 128 can be calculated in the form of one or more RT transforms that take into account the respective rotations and translations of the optical image sensor 106 and the depth sensor 104.

[0049] Back to Figure 1In some embodiments, the pose dataset collection system 120 can receive the point cloud data 110 and apply an outlier filter 122. For example, the outlier filter 122 can implement a feature detection function that generates a tight bounding box around the face and head of the training subject's image and can exclude points in the 3D point cloud 110 that fall outside the bounding box. In some embodiments, the pose dataset collection system 120 can include a geometric registration matching algorithm 124 that generates head pose measurements 126 based on the point cloud data 110. For example, the geometric registration matching algorithm 124 can receive the point cloud data 110 (e.g., filtered by the outlier filter 122) and calculate a transformation between the point cloud data 110 and the registered head model point cloud data 114, the transformation representing a measurement of the head pose present in the point cloud data 110. For example, the geometric registration matching algorithm 124 can apply an ICP registration algorithm to calculate a transformation that accounts for deviations (e.g., with respect to yaw, pitch, and roll) of the point cloud data 110 from the registered head model point cloud data 114. A transform representing this deviation may be used to calculate a head pose measurement 126 output from the geometric registration matching algorithm 124 , which indicates a measurement of the head pose present in the synchronized sensor data 108 .

[0050] To associate the head pose measurements 126 with the head pose observed from the viewpoint of the individual image sensors 106, the pose dataset collection system 120 may include a head pose measurement mapping 130 that applies calibration parameters 128 to map the head pose measurements 126 into a reference frame associated with each of the one or more optical image sensors 106. Thus, for each image frame of the optical image data 112, a transformation of the head pose measurements 126 may be calculated that maps the head pose measurements 126 into a reference frame for the field of view of the image frame—thereby defining a ground-truth head pose measurement for the head pose appearing in the particular image frame. Thus, the output of the head pose measurement mapping 130 may include a set of training samples 132, wherein each individual training sample may include an image frame sample 134 from the optical image data 112, and corresponding ground-truth (GT) pose data 136, the ground-truth pose data including the head pose measurements 126 mapped to the viewpoint and / or reference frame of the image frame sample 134. In some embodiments, the set of training samples 132 may be included in a head pose training dataset 140, which may be used to train a head pose prediction machine learning model, such as Figure 4Further discussed. Notably, since the ground-truth pose data 136 has been mapped to image frame samples 134, each training sample 132 can be used to train a head pose prediction machine learning model, where the machine learning model must understand the extrinsic relationship between the depth sensor 104 and the image sensor 106 used in generating the training samples 132. Furthermore, the machine learning model can learn to infer head pose predictions from camera images captured by an optical image sensor located at a different location inside the vehicle than the image sensor 106.

[0051] In some embodiments, the head pose training dataset 140 can include training samples 132 generated from many (e.g., thousands) sets of synchronized sensor data 108 collected over multiple sessions, where different training subjects are operating a vehicle, such as by driving the vehicle on public roads over a period of time. As discussed above, for each individual training subject, the pose dataset collection system 120 can have corresponding registered head model point cloud data 114 (e.g., optimized for its facial features) that is used to calculate head pose measurements 126 based on the synchronized sensor data 108 collected from the training subject.

[0052] Now refer to Figure 3 , Figure 3 is a data flow diagram illustrating a registration process for generating registered head model point cloud data 114 for a training subject, such as may be described with respect to Figure 1 Use. A registered head model 114 can be generated using a registration optimization function 320 based on point cloud data 310 collected of the training subject's head and face (e.g., using a depth sensor 302). The depth sensor 302 can include a sensor such as the depth sensor 104, such as a stereo image camera depth sensor, a time-of-flight camera depth sensor, a millimeter wave radar sensor, a Kinect sensor, or other depth sensor, wherein the 3D point cloud data 310 includes or is derived from the output of the depth sensor 302. During the registration process, the training subject can occupy a car seat of the vehicle 800 and can be instructed to perform predefined head movements while capturing point cloud data 310 using the depth sensor 302. The registration optimization function 320 can use the 3D point cloud data 310 to fit the generic head model point cloud data 312 to personalize the model to reflect the head and facial characteristics of the training subject. This optimization results in the generation of a registered head model 114.

[0053] The registration optimization function 320 may input a 3D point cloud representation of the generic head phantom point cloud data 312 and a 3D point cloud representation of a training subject from the point cloud data 310. In some embodiments, the registration optimization function 320 may compute a rigid model transformation between the point cloud data 310 and the generic head phantom point cloud data 312, for example, by finding closest point correspondences using an iterative closest point (ICP) registration algorithm. The registration optimization function 320 may be used to align the point clouds in 3D space by iteratively minimizing the distance between corresponding points in the point clouds. In some embodiments, the registration optimization function 320 estimates updates by iteratively using least squares optimization to minimize the error between the point cloud data 310 and the generic head phantom point cloud data 312. The optimization produces a set of coefficients that are unique to the training subject on which the registration process is performed.

[0054] In some embodiments, the registration optimization function 320 can provide instructional feedback 332 to the training subject via a human-machine interface 330 (e.g., a display). The instructional feedback 332 can include, for example, visual guidance and / or auditory / voice instructions instructing the training subject on how to rotate their head to capture a specified range of viewing angles. The predefined head movements can include three directions of head rotation, including rotations in roll, pitch, and / or yaw. In some embodiments, the instructional feedback 332 can include displaying a rotatable head model on the human-machine interface (HMI) 330 that tracks the training subject's head pose. The registration optimization function 320 can perform optimization in real time to fit the generic head model point cloud data 312 to the training subject's facial features. The real-time display of the conversion of the generic head model point cloud data 312 to the registered head model 114 can be used to make the registration process interactive, wherein the training subject can visually observe the displayed head model's fit continuously improving and gradually adapting to their face.

[0055] Now refer to Figure 4 , Figure 4 An example head pose prediction training system 400 is shown for training a machine learning model to infer head pose predictions based on optical image data. Figure 4 As shown, the head pose training dataset 140 can be used to train the head pose prediction model 410, for example, Figure 1 The head pose training dataset 140 is generated by the pose dataset collection system 120. As an example, one or more training samples 402 can each include an image frame sample 404 (e.g., the image frame sample 134 of an optical image representing the head pose of the training subject) and corresponding GT pose data 406 (e.g., the GT pose data 136), the GT pose data 406 including ground-truth head pose measurements mapped to the viewpoint and / or reference frame of the image frame sample 404.

[0056] To train the head pose prediction model 410, in some embodiments, one or more training samples 402 may be input to the head pose prediction model 410, and a resulting head pose prediction 420 may be output from the head pose prediction model 410. The head pose prediction 420 may be input to a loss function 430 to measure the difference between the head pose prediction 420 and the GT pose data 406, and a loss score may be calculated based on the difference, the loss score representing the accuracy of the head pose prediction 420. The loss score may be used to generate loss feedback 432 to iteratively adjust the head pose prediction model 410 during training to iteratively reduce the loss score and converge to a configuration of the head pose prediction model 410 that accurately produces head pose predictions 420 that match the GT pose data 406. In some embodiments, the head pose prediction model 410 may be trained by further using the registered head model point cloud data 114 associated with the training subjects appearing in the image frame samples 404 as training input.

[0057] Now refer to Figure 5 , Figure 5 A head pose prediction system 500 is shown that includes a head pose prediction model 410 trained as described herein. In some embodiments, the head pose prediction system 500 can be incorporated into an ADAS system, such as a vehicle 800. In this example, the head pose prediction model 410 can receive one or more frames of optical image data 504 representing the heads of one or more occupants of the vehicle 800. The optical image data 504 can be generated from one or more optical image sensors 502 that capture images of the interior of the vehicle 800, such as images of the head of one or more occupants of the vehicle 800. Figure 1 The optical image sensor 502 may include sensors of the OMS and / or DMS, such as those associated with the vehicle 800 and Figures 8A-8D Based on optical image data 504 , head pose prediction model 410 may infer a 3D head pose prediction 520 , which represents a measurement of the vehicle occupant's head pose as captured by optical image data 504 .

[0058] As described above, 3D head pose prediction 520 can provide a 3D pose estimate (e.g., including yaw, pitch, and / or roll) relative to the local reference frame of optical image sensor 502 that captured optical image data 504. To enable use of 3D head pose prediction 520 in conjunction with sensor data from other sensors, 3D head pose prediction 520 can be mapped to another reference frame used by vehicle systems, such as a global reference frame corresponding to the physical structure of vehicle 800. In some embodiments, head pose prediction system 500 can include a head pose prediction mapping function 522 that inputs extrinsic calibration parameters of optical image sensor 502 that captured optical image data 504, such as vehicle geometry data 524. For example, in some embodiments, vehicle geometry data 524 can include vehicle geometry specifications (e.g., CAD model information) regarding the geometry of the vehicle's interior, combined with accurate extrinsic calibration parameters that calibrate optical image sensor 502 to the vehicle's geometry. The head pose prediction mapping function 522 can apply one or more extrinsic calibration parameters (e.g., RT transform) that map the 3D head pose prediction 520 to the optical image sensor 502 in a global reference frame, and can output the result as 3D head pose data 526, which can represent absolute head pose measurements of the vehicle occupant relative to the structure of the vehicle 800 (or other global reference frame used by the vehicle 800). Vehicle systems operating using the 3D head pose data 526 can obtain such data in a form that can be more directly used in conjunction with other OMS / DMS sensor data captured by other sensors or generated by other onboard systems.

[0059] For example, in some embodiments, 3D head pose data 526 can be used by interior ("in-cabin") monitoring system 530 for various interior monitoring functions, such as, but not limited to, ADAS functions, child and / or animal occupant detection, activity monitoring, attention monitoring, gaze prediction, digital assistant interaction monitoring, and / or other functions. Other interior monitoring functions can include, for example, recognizing the faces, facial features, eye information, and / or other information of one or more occupants of vehicle 800, recognizing occupants based on facial features, and / or detecting the gaze of occupants of vehicle 800. Based at least in part on head pose data 526, interior monitoring system 530 can generate output 534 for controlling one or more operations of vehicle 800. Output 534 can be generated using one or more machine learning models and / or deep neural networks (DNNs) 532 and / or other computer vision algorithms, image processing algorithms, machine learning models (e.g., machine learning algorithms), etc. to operate and / or control functions of the vehicle based at least on head pose data 526.

[0060] Now refer to Figure 6 , Figure 6FIG. 6 is a flow chart illustrating a method 600 for training a head pose prediction model according to some embodiments of the present disclosure. Figure 6 The features and elements described in method 600 may be used in conjunction with, in combination with, or as a substitute for elements of any other embodiment discussed herein, and vice versa. Figure 6 The functionality, structure, and other descriptions of the elements of the embodiments described in the specification may apply to the same or similarly named or described elements in any of the figures and / or embodiments described herein, and vice versa. Each block of the method 600 described herein includes a computational process that can be performed using any combination of hardware, firmware, and / or software. For example, the various functions may be performed by one or more processors including processing circuitry for executing instructions stored in a memory. The method may also be embodied as computer-usable instructions stored on a computer storage medium. The method may be provided by a standalone application, a service, or a hosted service (standalone or in combination with another hosted service), or a plug-in to another product, to name a few. Additionally, the method 600 is directed to Figure 1 Head pose dataset collection system and / or Figure 5 The head pose prediction system is described by way of example. However, the method may additionally or alternatively be performed by any system or any combination of systems, including but not limited to the systems described herein.

[0061] As discussed in more detail herein, in some embodiments, the method may include generating a prediction of a 3D pose of at least a portion of a vehicle occupant based on a machine learning model trained using: one or more 3D ground truth measurements determined using a 3D point cloud representing a model customized for at least a portion of a training subject, and optical image data of at least a portion of the training subject captured synchronously with the 3D point cloud, wherein the one or more 3D ground truth head pose measurements are transformed to a reference frame of the optical image data.

[0062] At block B602, method 600 includes determining at least one three-dimensional (3D) measurement corresponding to a head pose of a vehicle occupant based on at least a deviation between a first 3D point cloud representation of a model customized for the vehicle occupant and a second 3D point cloud representation of at least a portion of the vehicle occupant based on at least depth image data. As described above, in some embodiments, the pose dataset collection system 120 may include a geometric registration matching algorithm 124 that generates head pose measurements 126 based on the point cloud data 110. The geometric registration matching algorithm 124 may receive the point cloud data 110 and calculate a transformation between the point cloud data 110 and the registered head model point cloud data 114, the transformation representing a measurement of the head pose present in the point cloud data 110. For example, the geometric registration matching algorithm 124 may apply an ICP registration algorithm to calculate a transformation that accounts for deviations in angular rotation of the point cloud data 110 from the registered head model point cloud data 114 (e.g., with respect to yaw, pitch, and roll). A transform representing this deviation may be used to calculate a head pose measurement 126 output from the geometric registration matching algorithm 124 , which indicates a measurement of the head pose present in the synchronized sensor data 108 .

[0063] In some embodiments, a registered head model of the vehicle occupant's head may be generated based at least on optimizing the generic head model using a 3D point cloud representation of the vehicle occupant's head. Figure 3 As discussed, a registered head model 114 can be generated using registration optimization functionality 320 based on point cloud data 310 of a training subject's head and face collected (e.g., using depth sensor 302). Depth sensor 302 can include a sensor such as depth sensor 104, such as a stereoscopic camera depth sensor, a time-of-flight camera depth sensor, a millimeter-wave radar sensor, a Kinect sensor, or other depth sensor, with 3D point cloud data 310 including or derived from the output of depth sensor 302. During the registration process, a training subject can occupy a car seat in vehicle 800 and be instructed to perform predefined head gestures while capturing point cloud data 310 using depth sensor 302. Registration optimization functionality 320 can use 3D point cloud data 310 to fit generic head model point cloud data 312 to personalize the model to reflect the head and facial characteristics of the training subject. This optimization results in the generation of a registered head model 114. The deviation between the first 3D point cloud representation of the registered head model and the second 3D point cloud representation of the vehicle occupant's head may be calculated, for example, based at least on an iterative closest point algorithm.

[0064] In some embodiments, the registration process can provide instructional feedback to the training subject via a human-machine interface (e.g., a display). Thus, the method can include outputting instructions to the vehicle occupant to rotate their head during the registration process to generate a 3D point cloud representation of the vehicle occupant's head. The instructional feedback can include, for example, visual guidance and / or auditory / voice instructions instructing the training subject on how to rotate their head to capture a specified range of viewing angles.

[0065] Method 600 includes, at block B604, capturing optical image data representing at least a portion of a vehicle occupant, wherein the optical image data is synchronized with depth image data. As described herein, in order for head pose measurements to accurately represent measurements of head pose as they appear in camera image frames, the depth sensor and one or more camera sensors operate in a synchronized manner to obtain temporally synchronized image frames and point cloud data. Figure 1 As shown, a depth sensor 104 and one or more optical image sensors 106 can define a set of dataset collection sensors 102 that operate in a synchronized manner to capture temporally synchronized point cloud data 110 and optical image data 112 of a training subject. A pose dataset collection system 120 can use an onboard synchronization signal provided by the vehicle 800 to coordinate the synchronization of the depth sensor 104 with the one or more optical image sensors 106. For example, the OMS and / or DMS functions of the vehicle 800 can be synchronized via a vehicle synchronization signal generated by common hardware or software. The pose dataset collection system can use this synchronization signal to trigger the OMS and / or DMS camera and simultaneously trigger the depth sensor to generate infrared pulses (e.g., for generating time-of-flight depth data) based on an offset synchronization with the camera, where the IR pulse emission is close to, but not completely aligned with, the timing of the camera. Thus, the method can synchronize the operation of the depth sensor and one or more optical image sensors to capture depth image data and optical image data and / or synchronize the operation of the depth sensor and one or more optical image sensors based on a synchronization signal generated by the vehicle. The depth sensor can be operated to capture depth image data in a manner that is synchronized with the operation of one or more optical image sensors to capture optical image data. In some embodiments, the optical image data includes image frames that can be captured by multiple cameras within the vehicle with different viewpoints of the vehicle occupants, such as with respect to Figure 2A and Figure 2B as described in the examples.

[0066] Method 600 includes, at block B606, converting at least one 3D measurement result to a reference frame of the optical image data to generate at least one 3D ground truth measurement result. Figure 2CAs shown, for example, for each of the optical image sensors 210-1, 210-2, 210-3, and 210-4, the calibration parameters 128 may include sensor-specific calibration parameters (shown as calibration parameters 230-1, 230-2, 230-3, and 230-4) that calibrate the corresponding image sensor to the depth sensor 104. Such data from the 3D point cloud 110 captured by the depth sensor 104 may be mapped to the reference frame of the camera image frame from the corresponding image sensor. In some embodiments, the calibration parameters 128 may be calculated in the form of one or more RT transforms that account for the respective rotations and translations of the optical image sensor 106 and the depth sensor 104. For example, the geometric registration matching algorithm 124 may receive the point cloud data 110 (e.g., filtered by the outlier filter 122) and calculate a transform between the point cloud data 110 and the registered head model point cloud data 114, the transform representing a measurement of the head pose present in the point cloud data 110. For example, the geometric registration matching algorithm 124 can apply an ICP registration algorithm to calculate a transformation that accounts for the deviation of the angular rotation of the point cloud data 110 from the registered head model point cloud data 114 (e.g., with respect to yaw, pitch, and roll). The transformation representing this deviation can be used to calculate head pose measurements 126 output from the geometric registration matching algorithm 124, which indicate measurements of the head pose appearing in the synchronized sensor data 108. To associate the head pose measurements 126 with the head pose observed from the viewpoint of the individual image sensors 106, the pose dataset collection system 120 can include a head pose measurement mapping 130 that applies calibration parameters 128 to map the head pose measurements 126 into a reference frame associated with each of the one or more optical image sensors 106. Thus, for each optical image frame, a transformation is calculated that maps the head pose measurements 126 to the head pose measurements 126 in the reference frame of the field of view of the image frame, thereby defining a ground-truth head pose measurement for the head pose appearing in that particular image frame. Thus, an output of head pose measurement mapping 130 may include a set of training samples 132, wherein each individual training sample may include an image frame sample 134 from optical image data 112 and corresponding GT pose data 136, the GT pose data including head pose measurements 126 mapped to the viewpoint and / or reference frame of the image frame sample 134. Thus, the method may include generating training samples to train a machine learning model, wherein the training samples include image frames based on at least the optical image data, and 3D ground truth labels based on 3D ground truth head pose measurements.

[0067] Method 600 includes, at block B608, updating the machine learning model to generate a prediction of a 3D pose of at least a portion of the occupant based at least on the optical image data and at least one 3D ground truth measurement. Figure 4 As shown, a head pose training dataset 140 (e.g., Figure 1 The head pose prediction model 410 may be trained using a dataset generated by the pose dataset collection system 120 of the training subject. As an example, one or more training samples 402 may each include an image frame sample 404 (e.g., an image frame sample 134 of an optical image representing the head pose of the training subject) and corresponding GT pose data 406 (e.g., GT pose data 136) including ground-truth head pose measurements mapped to the viewpoint and / or reference frame of the image frame sample 404. To train the head pose prediction model 410, in some embodiments, one or more training samples 402 may be input to the head pose prediction model 410, and a resulting head pose prediction 420 may be output from the head pose prediction model 410. The head pose prediction 420 may be input to a loss function 430 to measure the difference between the head pose prediction 420 and the GT pose data 406, and a loss score may be calculated based on the difference representing the accuracy of the head pose prediction 420. This loss score can be used to generate loss feedback 432 to iteratively adjust the head pose prediction model 410 during training to iteratively reduce the loss score and converge to a configuration of the head pose prediction model 410 that accurately produces head pose predictions 420 that match the GT pose data 406. In some embodiments, the head pose prediction model 410 can be trained by further using the registered head model point cloud data 114 associated with the training subjects appearing in the image frame samples 404 as training input.

[0068] Now refer to Figure 7 , Figure 7 FIG. 7 is a flow chart illustrating a method 700 for head posture prediction according to some embodiments of the present disclosure. Figure 7 The features and elements described in the method 700 may be used in conjunction with, in combination with, or as a substitute for elements of any other embodiment discussed herein, and vice versa. Figure 7The functionality, structure, and other descriptions of the elements of the embodiments described in the specification may apply to the same or similarly named or described elements in any of the figures and / or embodiments described herein, and vice versa. Each block of the method 700 described herein includes a computational process that can be performed using any combination of hardware, firmware, and / or software. For example, the various functions may be performed by one or more processors that include processing circuitry for executing instructions stored in a memory. The method may also be embodied as computer-usable instructions stored on a computer storage medium. The method may be provided by a standalone application, a service, or a hosted service (standalone or in combination with another hosted service), or a plug-in to another product, to name a few. Additionally, the method 700 is about Figure 1 Head pose dataset collection system and / or Figure 5 The head pose prediction system is described by way of example. However, the method may additionally or alternatively be performed by any system or any combination of systems, including but not limited to the systems described herein.

[0069] Method 700 includes, at block B702, capturing optical image data representing at least a portion of a vehicle occupant. Head pose prediction model 410 may receive one or more frames of optical image data 504 representing the heads of one or more occupants of vehicle 800. Optical image data 504 may be generated by one or more optical image sensors 502 capturing images of the interior of vehicle 800, e.g., Figure 1 The optical image sensor 106 is depicted. In some embodiments, the head pose prediction system 500 can be incorporated into an ADAS system of, for example, a vehicle 800.

[0070] Method 700 includes, at block B704, generating a three-dimensional (3D) head pose prediction based at least on optical image data using a machine learning model. The machine learning model may infer the 3D head pose prediction based at least on: 3D ground truth measurement data representing a deviation between a first 3D point cloud representation of a model customized for a training subject and a second 3D point cloud representation of at least a portion of the training subject based at least on depth image data; and optical image data representing at least a portion of the training subject, wherein the optical image data is synchronized with the depth image data. The registered head model of the training subject's head may be generated based at least on optimizing a generic head model using the 3D point cloud representation of the training subject's head, as described with respect to FIG. Figure 3 discussed.

[0071] In some embodiments, the 3D head pose prediction 520 can provide a 3D pose estimate (e.g., including yaw, pitch, and / or roll) relative to a local reference frame of the optical image sensor 502 that captured the optical image data 504. In some embodiments, the head pose prediction system 500 can include a head pose prediction mapping function 522 that inputs extrinsic calibration parameters of the optical image sensor 502 that captured the optical image data 504, such as from the vehicle geometry data 524. In some embodiments, the head pose prediction mapping function 522 can apply one or more extrinsic calibration parameters (e.g., an RT transform) of the optical image sensor 502 that maps the 3D head pose prediction 520 to a global reference frame and output the result as 3D head pose data 526, which can represent absolute head pose measurements of the vehicle occupant relative to the structure of the vehicle 800 (or other global reference frame used by the vehicle 800). That is, in some embodiments, the method converts 3D ground truth head pose measurement data to a reference frame of optical image data based on applying one or more extrinsic calibration parameters, wherein the one or more extrinsic calibration parameters represent one or more rotation-translation (RT) transformations between a depth sensor that captures the depth image data and one or more optical image sensors that capture the optical image data.

[0072] The vehicle can be controlled to perform one or more operations based at least on the 3D head pose prediction. For example, the gaze direction of a vehicle occupant can be calculated based at least on the 3D head pose prediction. The interior ("in-cabin") monitoring system can use the 3D head pose data 526 to implement various other interior monitoring functions, such as, but not limited to, ADAS functions, child and / or animal occupant detection, activity monitoring, attention monitoring, digital assistant interaction monitoring, and / or other functions. Other interior monitoring functions may include, for example, identifying the faces, facial landmarks, eye information, and / or other information of one or more occupants of the vehicle 800, identifying occupants based on facial features, and / or detecting the gaze of an occupant of the vehicle 800.

[0073] The systems and methods described herein may be used by, but are not limited to, non-autonomous vehicles, semi-autonomous vehicles (e.g., in one or more adaptive driver assistance systems (ADAS)), manned and unmanned robots or robotic platforms, warehouse vehicles, off-road vehicles, vehicles coupled to one or more trailers, flying boats, ships, space shuttles, emergency response vehicles, motorcycles, electric or motorized bicycles, airplanes, construction vehicles, trains, underwater vehicles, remotely operated vehicles (e.g., drones), and / or other vehicle types. Furthermore, the systems and methods described herein may be used for a variety of purposes, such as, but not limited to, machine control, machine motion, machine driving, synthetic data generation, model training, perception, augmented reality, virtual reality, mixed reality, robotics, security and surveillance, simulation and digital twins, autonomous or semi-autonomous machine applications, deep learning, environmental simulation, object or participant simulation and / or digital twins, data center processing, conversational AI (artificial intelligence), light transport simulation (e.g., ray tracing, path tracing, etc.), collaborative content creation of 3D assets, cloud computing, generative AI, and / or any other suitable application.

[0074] The disclosed embodiments can be included in a variety of different systems, such as automotive systems (e.g., control systems for autonomous or semi-autonomous machines, perception systems for autonomous or semi-autonomous machines), systems implemented using robots, aviation systems, medical systems, boating systems, smart area monitoring systems, systems for performing deep learning operations, systems for performing simulation operations, systems for performing digital twin operations, systems implemented using edge devices, systems including one or more virtual machines (VMs), systems for performing synthetic data generation operations, systems implemented at least in part in a data center, systems for performing conversational artificial intelligence (AI) operations, systems implementing one or more language models (such as one or more large language models (LLMs)), systems for performing light transport simulations, systems for performing collaborative content creation of 3D assets, systems implemented at least in part using cloud computing resources, and / or other types of systems.

[0075] Example autonomous vehicle

[0076] Figure 8Ais an illustration of an example autonomous vehicle 800 according to some embodiments of the present disclosure. Autonomous vehicle 800 (alternatively referred to herein as "vehicle 800") may include, but is not limited to, a passenger vehicle such as a car, a truck, a bus, a first responder vehicle, a shuttle, an electric or motorized bicycle, a motorcycle, a fire truck, a police vehicle, an ambulance, a boat, a construction vehicle, an underwater vessel, a robotic vehicle, a drone, an airplane, a vehicle coupled to a trailer (e.g., a semi-tractor truck for hauling cargo), and / or another type of vehicle (e.g., a vehicle that is unmanned and / or accommodates one or more passengers). Autonomous vehicles are generally described in terms of levels of automation as defined by the National Highway Traffic Safety Administration (NHTSA), a division of the U.S. Department of Transportation, and the Society of Automotive Engineers (SAE), “Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles” (Standard No. J3016-201806, issued on June 15, 2018, Standard No. J3016-201609, issued on September 30, 2016, and previous and future versions of the same). The vehicle 800 may be capable of implementing one or more functions consistent with Levels 3-5 of autonomous driving levels. The vehicle 800 may be capable of implementing one or more functions according to Levels 1-5 of autonomous driving levels. For example, depending on the embodiment, the vehicle 800 may be capable of driver assistance (Level 1), partial automation (Level 2), conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5). As used herein, the term “autonomous” may include any and / or all types of autonomy of a vehicle 800 or other machine, such as fully autonomous, highly autonomous, conditionally autonomous, partially autonomous, with assisted autonomy, semi-autonomous, primarily autonomous, or other designations.

[0077] Vehicle 800 may include components such as a chassis, a body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other vehicle components. Vehicle 800 may include a propulsion system 850, such as an internal combustion engine, a hybrid power plant, an all-electric engine, and / or another type of propulsion system. Propulsion system 850 may be connected to a drivetrain of vehicle 800, which may include a transmission, to achieve propulsion of vehicle 800. Propulsion system 850 may be controlled in response to receiving a signal from throttle / accelerator 852.

[0078] A steering system 854, which may include a steering wheel, may be used to steer the vehicle 800 (e.g., along a desired path or route) when the propulsion system 850 is operating (e.g., when the vehicle is in motion). The steering system 854 may receive signals from a steering actuator 856. For fully automated (Level 5) functionality, a steering wheel may be optional.

[0079] Brake sensor system 846 may be used to operate vehicle brakes in response to receiving signals from brake actuator 848 and / or brake sensors.

[0080] May include one or more system on chip (SoC) 804 ( Figure 8C ) and / or one or more GPUs can provide signals (e.g., representing commands) to one or more components and / or systems of the vehicle 800. For example, the one or more controllers can send signals to operate the vehicle brakes via one or more brake actuators 848, to operate the steering system 854 via one or more steering actuators 856, and to operate the propulsion system 850 via one or more throttles / accelerators 852. The one or more controllers 836 can include one or more onboard (e.g., integrated) computing devices (e.g., supercomputers) that process sensor signals and output operational commands (e.g., signals representing commands) to enable autonomous driving and / or assist a human driver in driving the vehicle 800. The one or more controllers 836 can include a first controller 836 for autonomous driving functions, a second controller 836 for functional safety functions, a third controller 836 for artificial intelligence functions (e.g., computer vision), a fourth controller 836 for infotainment functions, a fifth controller 836 for redundancy in emergency situations, and / or other controllers. In some examples, a single controller 836 may handle two or more of the above functions, two or more controllers 836 may handle a single function, and / or any combination thereof.

[0081] The one or more controllers 836 may provide signals for controlling one or more components and / or systems of the vehicle 800 in response to sensor data (eg, sensor input) received from one or more sensors. Sensor data may be received from, for example and without limitation, global navigation satellite system (“GNSS”) sensors 858 (e.g., global positioning system sensors), RADAR sensors 860 , ultrasonic sensors 862 , LIDAR sensors 864 , inertial measurement unit (IMU) sensors 866 (e.g., accelerometers, gyroscopes, magnetic compasses, magnetometers, etc.), microphones 896 , stereo cameras 868 , wide-angle cameras 870 (e.g., fisheye cameras), infrared cameras 872 , surround cameras 874 (e.g., 360-degree cameras), long-range and / or mid-range cameras 898 , speed sensors 844 (e.g., for measuring the velocity of the vehicle 800 ), vibration sensors 842 , steering sensors 840 , brake sensors (e.g., as part of a brake sensor system 846 ), one or more occupant monitoring system (OMS) sensors 801 (e.g., one or more interior cameras), and / or other sensor types.

[0082] One or more of the controllers 836 may receive input (e.g., represented by input data) from the instrument cluster 832 of the vehicle 800 and provide output (e.g., represented by output data, display data, etc.) via a human machine interface (HMI) display 834, an audible annunciator, a speaker, and / or via other components of the vehicle 800. These outputs may include information such as vehicle speed, velocity, time, map data (e.g., Figure 8C The HMI display 834 may include information such as a high-definition ("HD") map 822 of the vehicle 800, location data (e.g., the location of the vehicle 800 on the map), directions, the locations of other vehicles (e.g., an occupancy grid), information about objects and object states as sensed by the controller 836, and the like. For example, the HMI display 834 may display information about the presence of one or more objects (e.g., street signs, warning signs, traffic light changes, etc.) and / or information about driving maneuvers that the vehicle has made, is making, or will make (e.g., changing lanes now, leaving 34B in two miles, etc.).

[0083] The vehicle 800 also includes a network interface 824 that can communicate over one or more networks using one or more wireless antennas 826 and / or a modem. For example, the network interface 824 can be capable of communicating over Long Term Evolution ("LTE"), Wideband Code Division Multiple Access ("WCDMA"), Universal Mobile Telecommunications System ("UMTS"), Global System for Mobile Communications ("GSM"), IMT-CDMA Multi-Carrier ("CDMA2000"), and the like. The one or more wireless antennas 826 can also enable communication between objects in the environment (e.g., vehicles, mobile devices, and the like) using one or more local area networks such as Bluetooth, Bluetooth Low Energy ("LE"), Z-Wave, ZigBee, and the like, and / or one or more low power wide area networks ("LPWAN") such as LoRaWAN, SigFox, and the like.

[0084] Figure 8B For use according to some embodiments of the present disclosure Figure 8A An example of camera positions and fields of view for autonomous vehicle 800 is shown. The cameras and respective fields of view are an example embodiment and are not intended to be limiting. For example, additional and / or alternative cameras may be included and / or located at different locations on vehicle 800.

[0085] The camera type used for the camera may include, but is not limited to, a digital camera that may be suitable for use with components and / or systems of the vehicle 800. The camera may operate at Automotive Safety Integrity Level (ASIL) B and / or at another ASIL. The camera type may have any image capture rate, such as 60 frames per second (fps), 120fps, 240fps, and the like, depending on the embodiment. The camera may be capable of using a rolling shutter, a global shutter, another type of shutter, or a combination thereof. In some examples, the color filter array may include a red-white-white-white (RCCC) color filter array, a red-white-white-blue (RCCB) color filter array, a red-blue-green-white (RBGC) color filter array, a Foveon X3 color filter array, a Bayer sensor (RGGB) color filter array, a monochrome sensor color filter array, and / or another type of color filter array. In some embodiments, a clear pixel camera such as a camera with an RCCC, RCCB, and / or RBGC color filter array may be used in an effort to improve light sensitivity.

[0086] In some examples, one or more of the cameras can be used to perform advanced driver assistance system (ADAS) functions (e.g., as part of a redundant or fail-safe design). For example, a multi-function monocular camera can be installed to provide functions including lane departure warning, traffic sign assistance, and intelligent headlight control. One or more of the cameras (e.g., all of the cameras) can simultaneously record and provide image data (e.g., video).

[0087] One or more of the cameras can be mounted in a mounting assembly, such as a custom-designed (three-dimensional ("3D") printed) assembly, to cut out stray light and reflections from within the car (e.g., reflections from the dashboard reflected in the windshield mirror) that might interfere with the camera's ability to capture image data. With respect to the wing mirror mounting assembly, the wing mirror assembly can be custom 3D printed so that the camera mounting plate matches the shape of the wing mirror. In some examples, one or more cameras can be integrated into the wing mirror. For side-view cameras, one or more cameras can also be integrated into the four pillars at each corner of the cabin.

[0088] A camera (e.g., a front-facing camera) having a field of view that includes a portion of the environment in front of the vehicle 800 can be used for surround vision to help identify the forward path and obstacles, as well as assist in providing information critical to generating an occupancy grid and / or determining a preferred vehicle path with the help of one or more controllers 836 and / or control SoCs. The front-facing camera can be used to perform many of the same ADAS functions as LIDAR, including emergency braking, pedestrian detection, and collision avoidance. The front-facing camera can also be used for ADAS functions and systems, including lane departure warning ("LDW"), autonomous cruise control ("ACC"), and / or other functions such as traffic sign recognition.

[0089] A variety of cameras may be used in the front-facing configuration, including, for example, a monocular camera platform including a complementary metal oxide semiconductor ("CMOS") color imager. Another example may be a wide-angle camera 870, which may be used to sense objects entering the field of view from the periphery (e.g., pedestrians, traffic at an intersection, or bicycles). Although Figure 8B The figure shows only one wide-angle camera, but there can be any number (including zero) of wide-angle cameras 870 on the vehicle 800. In addition, any number of long-range cameras 898 (e.g., a pair of long-view stereo cameras) can be used for depth-based object detection, especially for objects for which neural networks have not yet been trained. Long-range cameras 898 can also be used for object detection and classification and basic object tracking.

[0090] Any number of stereo cameras 868 may also be included in the front configuration. In at least one embodiment, one or more stereo cameras 868 may include an integrated control unit including a scalable processing unit that may provide a multi-core microprocessor and programmable logic ("FPGA") with an integrated controller area network ("CAN") or Ethernet interface on a single chip. Such a unit may be used to generate a 3D map of the vehicle's environment, including distance estimates for all points in the image. Alternative stereo cameras 868 may include a compact stereo vision sensor that may include two camera lenses (one on the left and one on the right) and an image processing chip that may measure the distance from the vehicle to a target object and use the generated information (e.g., metadata) to activate autonomous emergency braking and lane departure warning functions. Other types of stereo cameras 868 may be used in addition to or alternatively to those described herein.

[0091] Cameras with a field of view that includes portions of the environment to the sides of the vehicle 800 (e.g., side-view cameras) can be used for surround viewing, providing information used to create and update occupancy grids and generate side impact collision warnings. For example, surround cameras 874 (e.g., Figure 8B Four surround cameras 874 (shown in FIG) can be placed on the vehicle 800. The surround cameras 874 can include a wide-angle camera 870, a fisheye camera, a 360-degree camera, and / or the like. For example, the four fisheye cameras can be placed on the front, rear, and sides of the vehicle. In an alternative arrangement, the vehicle can use three surround cameras 874 (e.g., left, right, and rear), and can utilize one or more other cameras (e.g., a forward-facing camera) as a fourth surround-view camera.

[0092] A camera having a field of view that includes a portion of the environment behind the vehicle 800 (e.g., a rearview camera) can be used to assist with parking, surround view, rear collision warning, and creating and updating occupancy grids. A variety of cameras can be used, including but not limited to cameras that are also suitable as front-facing cameras as described herein (e.g., long-range and / or mid-range cameras 898, stereo cameras 868, infrared cameras 872, etc.).

[0093] A camera (e.g., one or more OMS sensors 801) whose field of view includes a portion of the interior environment within the cabin of vehicle 800 can be used as part of an occupant monitoring system (OMS), such as, but not limited to, a driver monitoring system (DMS). For example, an OMS sensor (e.g., OMS sensor 801) can be used (e.g., by controller 836) to track the gaze direction, head posture, and / or blinking of an occupant and / or driver. This gaze information can be used to determine the occupant's or driver's attention level (e.g., to detect drowsiness, fatigue, and / or distraction) and / or take responsive action to prevent harm to the occupant or operator. In some embodiments, data from the OMS sensor can be used to implement gaze-controlled operations triggered by the driver and / or non-driver passengers, such as, but not limited to, adjusting cabin temperature and / or airflow, opening and closing windows, controlling cabin lighting, controlling the entertainment system, adjusting rearview mirrors, adjusting seat position, and / or other operations. In some embodiments, the OMS can be used for applications such as determining when an object and / or passenger remains in the cabin (e.g., by detecting the presence of an occupant after the driver has exited the vehicle). In some embodiments, one or more of the optical image sensor 106 and / or the optical image sensor 502 may include an optical image sensor of the vehicle 800 , such as, but not limited to, an OMS sensor 801 .

[0094] Figure 8C For use according to some embodiments of the present disclosure Figure 8A 800. It will be understood that this arrangement and other arrangements described herein are set forth merely as examples. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, functional groupings, etc.) may be used in addition to or in place of those shown, and some elements may be omitted entirely. Further, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in conjunction with other components, and in any appropriate combination and location. The various functions described herein as being performed by entities may be implemented by hardware, firmware, and / or software. For example, the various functions may be implemented by a processor executing instructions stored in memory.

[0095] Figure 8CEach of the components, features, and systems of vehicle 800 is illustrated as being connected via bus 802. Bus 802 may include a controller area network (CAN) data interface (alternatively, referred to herein as a "CAN bus"). CAN may be a network internal to vehicle 800 that assists in controlling various features and functions of vehicle 800, such as actuation of brakes, acceleration, braking, steering, windshield wipers, and the like. The CAN bus may be configured to have tens or even hundreds of nodes, each with its own unique identifier (e.g., CAN ID). The CAN bus may be read to find steering wheel angle, ground speed, engine revolutions per minute (RPM), button positions, and / or other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0096] Although bus 802 is described here as a CAN bus, this is not intended to be limiting. For example, FlexRay and / or Ethernet may be used in addition to or in lieu of a CAN bus. Furthermore, although bus 802 is represented by a single line, this is not intended to be limiting. For example, there may be any number of buses 802, which may include one or more CAN buses, one or more FlexRay buses, one or more Ethernet buses, and / or one or more other types of buses using different protocols. In some examples, two or more buses 802 may be used to perform different functions and / or may be used for redundancy. For example, a first bus 802 may be used for collision avoidance functionality, and a second bus 802 may be used for drive control. In any example, each bus 802 may communicate with any component of vehicle 800, and two or more buses 802 may communicate with the same component. In some examples, each SoC 804, each controller 836, and / or each computer within the vehicle may have access to the same input data (e.g., input from sensors of vehicle 800) and may be connected to a common bus such as a CAN bus.

[0097] The vehicle 800 may include one or more controllers 836, such as those described herein. Figure 8A Controller 836 may be used for a variety of functions. Controller 836 may be coupled to any of the other various components and systems of vehicle 800 and may be used for control of vehicle 800, artificial intelligence of vehicle 800, infotainment for vehicle 800, and / or the like.

[0098] The vehicle 800 may include one or more system-on-chips (SoCs) 804. The SoC 804 may include a CPU 806, a GPU 808, a processor 810, a cache 812, an accelerator 814, a data store 816, and / or other components and features not shown. The SoC 804 may be used to control the vehicle 800 in a variety of platforms and systems. For example, the one or more SoCs 804 may be combined with an HD map 822 in a system (e.g., a system of the vehicle 800), which may be downloaded from one or more servers (e.g., a server) via a network interface 824. Figure 8D one or more servers 878) to obtain map refreshes and / or updates.

[0099] The CPU 806 may include a CPU cluster or CPU complex (alternatively, referred to herein as a "CCPLEX"). The CPU 806 may include multiple cores and / or L2 caches. For example, in some embodiments, the CPU 806 may include eight cores in a coherent multiprocessor configuration. In some embodiments, the CPU 806 may include four dual-core clusters, each with a dedicated L2 cache (e.g., a 2MB L2 cache). The CPU 806 (e.g., CCPLEX) may be configured to support simultaneous cluster operations such that any combination of CPU 806 clusters can be active at any given time.

[0100] The CPU 806 may implement power management capabilities including one or more of the following features: each hardware block may be automatically clock gated when idle to conserve dynamic power; each core clock may be gated when the core is not actively executing instructions due to the execution of WFI / WFE instructions; each core may be independently power gated; each core cluster may be independently clock gated when all cores are clock gated or power gated; and / or each core cluster may be independently power gated when all cores are power gated. The CPU 806 may further implement an enhanced algorithm for managing power states, in which allowed power states and expected wakeup times are specified, and hardware / microcode determines the optimal power state to enter for the core, cluster, and CCPLEX. The processing core may support a simplified power state entry sequence in software, with this work being offloaded to the microcode.

[0101] The GPU 808 may include an integrated GPU (alternatively referred to herein as an "iGPU"). The GPU 808 may be programmable and efficient for parallel workloads. In some examples, the GPU 808 may use an enhanced tensor instruction set. The GPU 808 may include one or more streaming microprocessors, each of which may include an L1 cache (e.g., an L1 cache with at least 96KB of storage capacity), and two or more of these streaming microprocessors may share an L2 cache (e.g., an L2 cache with 512KB of storage capacity). In some embodiments, the GPU 808 may include at least eight streaming microprocessors. The GPU 808 may use a computing application programming interface (API). In addition, the GPU 808 may use one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0102] In the case of automotive and embedded use, GPU 808 can be power optimized to achieve optimal performance. For example, GPU 808 can be manufactured on fin field effect transistors (FinFETs). However, this is not intended to be limiting, and GPU 808 can be manufactured using other semiconductor manufacturing processes. Each streaming microprocessor can merge several mixed precision processing cores divided into multiple blocks. For example and without limitation, 64 PF32 cores and 32 PF64 cores can be divided into four processing blocks. In such an example, each processing block can be allocated 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two mixed precision NVIDIA tensor cores for deep learning matrix arithmetic, L0 instruction cache, warp scheduler, dispatch unit and / or 64KB register file. In addition, the streaming microprocessor may include independent parallel integer and floating point data paths to provide efficient execution of workloads using a mix of computation and addressing calculations. The streaming microprocessor may include independent thread scheduling capabilities to allow for finer-grained synchronization and collaboration between parallel threads. Streaming microprocessors may include a combined L1 data cache and shared memory unit to increase performance while simplifying programming.

[0103] The GPU 808 can include high bandwidth memory (HBM) and / or a 16GB HBM2 memory subsystem that provides a peak memory bandwidth of approximately 900 GB / s in some examples. In some examples, synchronous graphics random access memory (SGRAM), such as fifth generation graphics double data rate synchronous random access memory (GDDR5), can be used in addition to or in lieu of HBM memory.

[0104] The GPU 808 may include unified memory technology that includes access counters to allow memory pages to be more accurately migrated to the processor that accesses them most frequently, thereby improving the efficiency of memory ranges shared between processors. In some examples, address translation services (ATS) support may be used to allow the GPU 808 to directly access the CPU 806 page tables. In such an example, when the GPU 808 memory management unit (MMU) experiences a miss, an address translation request may be transmitted to the CPU 806. In response, the CPU 806 may look up the virtual-to-physical mapping for the address in its page table and transmit the translation back to the GPU 808. In this way, unified memory technology may allow a single unified virtual address space to be used for memory of both the CPU 806 and the GPU 808, thereby simplifying GPU 808 programming and porting applications to the GPU 808.

[0105] In addition, GPU 808 can include access counters that can track how often GPU 808 accesses the memory of other processors. Access counters can help ensure that memory pages are moved to the physical memory of the processor that accesses them most frequently.

[0106] SoC 804 may include any number of caches 812, including those described herein. For example, cache 812 may include an L3 cache available to both CPU 806 and GPU 808 (e.g., connected to both CPU 806 and GPU 808). Cache 812 may include a write-back cache that can track the state of lines, for example, using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). Depending on the embodiment, the L3 cache may include 4MB or more, although smaller cache sizes may also be used.

[0107] The SoC 804 may include an arithmetic logic unit (ALU) that can be utilized in performing any of a variety of tasks or operations associated with the vehicle 800, such as processing a DNN. Furthermore, the SoC 804 may include a floating point unit (FPU) (or other math coprocessor or digital coprocessor type) for performing mathematical operations within the system. For example, the SoC 804 may include one or more FPUs integrated as execution units within the CPU 806 and / or GPU 808.

[0108] SoC 804 may include one or more accelerators 814 (e.g., hardware accelerators, software accelerators, or a combination thereof). For example, SoC 804 may include a hardware accelerator cluster, which may include optimized hardware accelerators and / or large on-chip memory. The large on-chip memory (e.g., 4MB SRAM) may enable the hardware accelerator cluster to accelerate neural networks and other calculations. The hardware accelerator cluster may be used to supplement the GPU 808 and offload some tasks of the GPU 808 (e.g., freeing up more cycles of the GPU 808 for performing other tasks). As an example, the accelerator 814 may be used for targeted workloads (e.g., perception, convolutional neural networks (CNNs), etc.) that are sufficiently stable to be easily controlled for acceleration. When used herein, the term "CNN" may include all types of CNNs, including region-based or regional convolutional neural networks (RCNNs) and fast RCNNs (e.g., for object detection).

[0109] In some embodiments, one or more aspects of pose dataset collection system 120 and / or head pose prediction system 500 may be implemented using code executed on one or more of CPU 806, GPU 808, and / or SoC 804. For example, head pose prediction model 410 may be implemented by one or more of GPUs 808.

[0110] The accelerator 814 (e.g., a hardware accelerator cluster) may include a deep learning accelerator (DLA). The DLA may include one or more tensor processing units (TPUs) that can be configured to provide an additional 10 trillion operations per second for deep learning applications and reasoning. The TPU may be an accelerator configured to perform image processing functions (e.g., for CNN, RCNN, etc.) and optimized for performing image processing functions. The DLA may be further optimized for a specific set of neural network types and floating-point operations and reasoning. The design of the DLA may provide higher performance per millimeter than a general-purpose GPU and far exceed the performance of the CPU. The TPU may perform several functions, including a single-instance convolution function, support for INT8, INT16, and FP16 data types for both features and weights, and post-processor functions.

[0111] DLA can quickly and efficiently execute neural networks, particularly CNNs, on processed or unprocessed data for any of a wide variety of functions, such as, but not limited to: CNNs for object recognition and detection using data from camera sensors; CNNs for distance estimation using data from camera sensors; CNNs for emergency vehicle detection and recognition and detection using data from microphones; CNNs for facial recognition and vehicle owner identification using data from camera sensors; and / or CNNs for safety and / or security-related events.

[0112] The DLA can perform any function of the GPU 808, and by using an inference accelerator, for example, the designer can target any function to either the DLA or the GPU 808. For example, the designer can focus the processing of CNNs and floating-point operations on the DLA and leave other functions to the GPU 808 and / or other accelerators 814.

[0113] The accelerator 814 (e.g., a hardware accelerator cluster) may include a programmable vision accelerator (PVA), which may be alternatively referred to herein as a computer vision accelerator. The PVA may be designed and configured to accelerate computer vision algorithms for advanced driver assistance systems (ADAS), autonomous driving, and / or augmented reality (AR) and / or virtual reality (VR) applications. The PVA may provide a balance between performance and flexibility. For example, each PVA may include, for example and without limitation, any number of reduced instruction set computer (RISC) cores, direct memory access (DMA), and / or any number of vector processors.

[0114] The RISC core can interact with an image sensor (e.g., an image sensor of any camera described herein), an image signal processor, and / or the like. Each of these RISC cores can include any amount of memory. Depending on the embodiment, the RISC core can use any of a number of protocols. In some examples, the RISC core can execute a real-time operating system (RTOS). The RISC core can be implemented using one or more integrated circuit devices, application specific integrated circuits (ASICs), and / or memory devices. For example, the RISC core can include an instruction cache and / or tightly coupled RAM.

[0115] The DMA can enable components of the PVA to access system memory independently of the CPU 806. The DMA can support any number of features used to provide optimizations for the PVA, including but not limited to support for multi-dimensional addressing and / or circular addressing. In some examples, the DMA can support addressing in up to six or more dimensions, which can include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0116] A vector processor can be a programmable processor that can be designed to efficiently and flexibly execute programming for computer vision algorithms and provide signal processing capabilities. In some examples, the PVA can include a PVA core and two vector processing subsystem partitions. The PVA core can include a processor subsystem, one or more DMA engines (e.g., two DMA engines), and / or other peripherals. The vector processing subsystem can operate as the main processing engine of the PVA and can include a vector processing unit (VPU), an instruction cache, and / or a vector memory (e.g., VMEM). The VPU core can include a digital signal processor, such as, for example, a single instruction multiple data (SIMD), a very long instruction word (VLIW) digital signal processor. The combination of SIMD and VLIW can enhance throughput and speed.

[0117] Each of the vector processors can include an instruction cache and can be coupled to dedicated memory. As a result, in some examples, each of the vector processors can be configured to execute independently of the other vector processors. In other examples, the vector processors included in a particular PVA can be configured to employ data parallelism. For example, in some embodiments, multiple vector processors included in a single PVA can execute the same computer vision algorithm, but on different regions of an image. In other examples, the vector processors included in a particular PVA can execute different computer vision algorithms simultaneously on the same image, or even execute different algorithms on sequential images or portions of images. Among other things, any number of PVAs can be included in a hardware accelerator cluster, and any number of vector processors can be included in each of these PVAs. In addition, the PVAs can include additional error correction code (ECC) memory to enhance overall system security.

[0118] The accelerator 814 (e.g., a hardware accelerator cluster) may include an on-chip computer vision network and SRAM to provide high bandwidth, low latency SRAM for the accelerator 814. In some examples, the on-chip memory may include at least 4MB of SRAM consisting of, for example and without limitation, eight field-configurable memory blocks that can be accessed by both the PVA and the DLA. Each pair of memory blocks may include an advanced peripheral bus (APB) interface, configuration circuitry, a controller, and a multiplexer. Any type of memory may be used. The PVA and DLA may access the memory via a backbone that provides high-speed memory access to the PVA and DLA. The backbone may include an on-chip computer vision network that interconnects the PVA and DLA to the memory (e.g., using APB).

[0119] The on-chip computer vision network can include an interface that ensures that both the PVA and DLA provide ready and valid signals before transmitting any control signals / addresses / data. Such an interface can provide separate phases and separate channels for transmitting control signals / addresses / data, as well as burst-based communication for continuous data transmission. This type of interface can comply with ISO 26262 or IEC 61508 standards, but other standards and protocols can also be used.

[0120] In some examples, the SoC 804 may include a real-time ray tracing hardware accelerator such as that described in U.S. patent application Ser. No. 16 / 101,232 filed on Aug. 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the position and extent of objects (e.g., within a world model) to generate real-time visualization simulations for use in RADAR signal interpretation, for sound propagation synthesis and / or analysis, for SONAR system simulations, for general wave propagation simulations, for comparison with LIDAR data for positioning and / or other functions, and / or for other uses. In some embodiments, one or more tree traversal units (TTUs) may be used to perform one or more ray tracing related operations.

[0121] The accelerator 814 (e.g., a hardware accelerator cluster) has a wide range of uses in autonomous driving. The PVA can be a programmable vision accelerator that can be used in key processing stages in ADAS and autonomous vehicles. The capabilities of the PVA are a good match for algorithmic domains that require predictable processing, low power, and low latency. In other words, the PVA performs well on semi-intensive or intensive rule computations, and even on small data sets that require predictable runtimes with low latency and low power. Therefore, in the context of a platform for autonomous vehicles, the PVA is designed to run classic computer vision algorithms because they are efficient at object detection and integer math operations.

[0122] For example, according to one embodiment of the technology, PVA is used to perform computer stereo vision. In some examples, a semi-global matching-based algorithm can be used, but this is not intended to be limiting. Many applications for Level 3-5 autonomous driving require on-the-fly motion estimation / stereo matching (e.g., structure from motion, pedestrian recognition, lane detection, etc.). PVA can perform computer stereo vision functions on input from two monocular cameras.

[0123] In some examples, PVA can be used to perform dense optical flow, by processing raw RADAR data (e.g., using a 4D Fast Fourier Transform) to provide processed RADAR. In other examples, PVA is used for time-of-flight depth processing, by processing raw time-of-flight data to provide processed time-of-flight data.

[0124] The DLA can be used to run any type of network to enhance control and driving safety, including, for example, a neural network that outputs a confidence measure for each object detection. Such confidence values ​​can be interpreted as probabilities, or as providing a relative "weight" of each detection compared to other detections. This confidence value enables the system to make further decisions about which detections should be considered true positives versus false positives. For example, the system can set a threshold for confidence and only consider detections that exceed the threshold as true positives. In an automatic emergency braking (AEB) system, a false positive detection could cause the vehicle to automatically apply emergency braking, which is clearly undesirable. Therefore, only the most confident detections should be considered triggers for AEB. The DLA can run a neural network to regress the confidence value. This neural network can take as its input at least some subset of parameters, such as bounding box dimensions, a ground plane estimate obtained (e.g., from another subsystem), inertial measurement unit (IMU) sensor 866 output related to the vehicle 800's orientation and distance, and 3D position estimates of objects obtained from the neural network and / or other sensors (e.g., LIDAR sensor 864 or RADAR sensor 860).

[0125] The SoC 804 may include one or more data stores 816 (e.g., memory). The data store 816 may be on-chip memory of the SoC 804 that may store neural networks to be executed on the GPU and / or DLA. In some examples, the data store 816 may be large enough to store multiple instances of the neural network for redundancy and safety. The data store 816 may include an L2 or L3 cache 812. References to the data store 816 may include references to memory associated with the PVA, DLA, and / or other accelerators 814 as described herein.

[0126] The SoC 804 may include one or more processors 810 (e.g., embedded processors). The processors 810 may include a boot and power management processor, which may be a dedicated processor and subsystem for handling boot power and management functions and related safety implementations. The boot and power management processor may be part of the SoC 804 boot sequence and may provide runtime power management services. The boot power and management processor may provide clock and voltage programming, auxiliary system low power state transitions, SoC 804 thermal and temperature sensor management, and / or SoC 804 power state management. Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to temperature, and the SoC 804 may use the ring oscillator to detect the temperature of the CPU 806, GPU 808, and / or accelerator 814. If it is determined that the temperature exceeds a threshold, the boot and power management processor may enter a temperature fault routine and place the SoC 804 in a lower power state and / or place the vehicle 800 in a driver safety parking mode (e.g., to safely park the vehicle 800).

[0127] The processor 810 may also include a set of embedded processors that can be used as an audio processing engine. The audio processing engine can be an audio subsystem that allows for full hardware support for multi-channel audio through multiple interfaces and a wide range of flexible audio I / O interfaces. In some examples, the audio processing engine is a dedicated processor core having a digital signal processor with dedicated RAM.

[0128] The processor 810 may also include an always-on processor engine that can provide the necessary hardware features to support low-power sensor management and wake-up use cases. The always-on processor engine may include a processor core, tightly coupled RAM, supporting peripherals (such as timers and interrupt controllers), various I / O controller peripherals, and routing logic.

[0129] The processor 810 may also include a safety cluster engine, which includes a dedicated processor subsystem that handles safety management of automotive applications. The safety cluster engine may include two or more processor cores, tightly coupled RAM, supporting peripherals (such as timers, interrupt controllers, etc.), and / or routing logic. In safety mode, the two or more cores can operate in lockstep mode and act as a single core with comparison logic to detect any differences between their operations.

[0130] Processor 810 may also include a real-time camera engine, which may include a dedicated processor subsystem for handling real-time camera management.

[0131] Processor 810 may also include a high dynamic range signal processor, which may include an image signal processor, which is a hardware engine that is part of the camera processing pipeline.

[0132] The processor 810 may include a video image compositer, which may be a processing block (e.g., implemented on a microprocessor) that implements the video post-processing functions required by the video playback application to produce the final image for the player window. The video image compositer may perform lens distortion correction for the wide-angle camera 870, the surround camera 874, and / or for the in-cab monitoring camera sensor. The in-cab monitoring camera sensor is preferably monitored by a neural network running on another instance of the advanced SoC, configured to recognize in-cab events and respond accordingly. The in-cab system may perform lip reading to activate mobile phone service and place calls, dictate emails, change vehicle destinations, activate or change the vehicle's infotainment system and settings, or provide voice-activated web surfing. Certain functions are only available to the driver when the vehicle is operating in autonomous mode and are disabled in other circumstances.

[0133] The video image compositer can include enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, in the presence of motion in the video, the noise reduction appropriately weights spatial information and downweights information provided by neighboring frames. In the case where an image or portion of an image does not include motion, the temporal noise reduction performed by the video image compositer can use information from previous images to reduce noise in the current image.

[0134] The video image compositor can also be configured to perform stereo rectification on the input stereo footage frames. The video image compositor can further be used for user interface composition when the operating system desktop is in use and the GPU 808 does not need to continuously render new surfaces. Even when the GPU 808 is powered on and active for 3D rendering, the video image compositor can be used to offload the GPU 808 to improve performance and responsiveness.

[0135] The SoC 804 may also include a Mobile Industry Processor Interface (MIPI) camera serial interface, a high-speed interface for receiving video and input from a camera, and / or a video input block that may be used for camera and related pixel input functions. The SoC 804 may also include an input / output controller that may be controlled by software and may be used to receive I / O signals that are not assigned to a specific role.

[0136] The SoC 804 may also include a wide range of peripheral interfaces to enable communication with peripherals, audio codecs, power management, and / or other devices. The SoC 804 may be used to process data from cameras (connected via Gigabit multimedia serial links and Ethernet), sensors (e.g., LIDAR sensor 864, RADAR sensor 860, etc., which may be connected via Ethernet), data from the bus 802 (e.g., vehicle 800 speed, steering wheel position, etc.), and data from the GNSS sensor 858 (connected via Ethernet or a CAN bus). The SoC 804 may also include dedicated high-performance mass storage controllers, which may include their own DMA engines and which may be used to free the CPU 806 from routine data management tasks.

[0137] SoC 804 can be an end-to-end platform with a flexible architecture that spans levels 3-5 of automation, providing a comprehensive functional safety architecture that leverages and efficiently uses computer vision and ADAS technologies for diversity and redundancy, along with deep learning tools to provide a platform for a flexible and reliable driving software stack. SoC 804 can be faster, more reliable, and even more energy- and space-efficient than conventional systems. For example, when combined with CPU 806, GPU 808, and data storage 816, accelerator 814 can provide a fast and efficient platform for level 3-5 autonomous vehicles.

[0138] This technology therefore provides capabilities and functionality that cannot be achieved with conventional systems. For example, computer vision algorithms can be executed on CPUs, which can be configured using high-level programming languages ​​such as the C programming language to execute a wide variety of processing algorithms across a wide variety of visual data. However, CPUs often fail to meet the performance requirements of many computer vision applications, such as those related to execution time and power consumption. In particular, many CPUs are unable to execute complex object detection algorithms in real time, a requirement for in-vehicle ADAS applications and practical Level 3-5 autonomous vehicles.

[0139] In contrast to conventional systems, by providing a CPU complex, a GPU complex, and a cluster of hardware accelerators, the technology described herein allows multiple neural networks to be executed simultaneously and / or sequentially, and the results to be combined to achieve Level 3-5 autonomous driving capabilities. For example, a CNN executed on a DLA or dGPU (e.g., GPU 820) can include text and word recognition, allowing the supercomputer to read and understand traffic signs, including signs for which a neural network has not been specifically trained. The DLA can also include a neural network that can recognize, interpret, and provide semantic understanding of the signs, and pass that semantic understanding to a path planning module running on the CPU complex.

[0140] As another example, as required for Level 3, 4, or 5 driving, multiple neural networks can be run simultaneously. For example, a warning sign consisting of "Caution: Flashing lights indicate icing conditions" along with a light can be interpreted by several neural networks, either independently or collectively. The sign itself can be identified as a traffic sign by a first neural network deployed (e.g., a trained neural network), and the text "Flashing lights indicate icing conditions" can be interpreted by a second neural network deployed, which informs the vehicle's path planning software (preferably executing on a CPU complex) that icing conditions exist when the flashing lights are detected. The flashing lights can be identified by operating a third neural network deployed over multiple frames, which informs the vehicle's path planning software of the presence (or absence) of the flashing lights. All three neural networks can run simultaneously, for example, within the DLA and / or on GPU 808.

[0141] In some examples, a CNN for facial recognition and owner recognition can use data from a camera sensor to identify the presence of an authorized driver and / or owner of the vehicle 800. The always-on sensor processing engine can be used to unlock the vehicle and turn on the lights when the owner approaches the driver's door, and in security mode, disable the vehicle when the owner leaves the vehicle. In this way, the SoC 804 provides security against theft and / or carjacking.

[0142] In another example, a CNN for emergency vehicle detection and identification can use data from microphone 896 to detect and identify emergency vehicle sirens. In contrast to conventional systems that use general classifiers to detect sirens and manually extract features, SoC 804 uses CNN to classify environmental and urban sounds and to classify visual data. In a preferred embodiment, the CNN running on the DLA is trained to identify the relative closing speed of emergency vehicles (for example, by using the Doppler effect). The CNN can also be trained to identify emergency vehicles specific to the local area in which the vehicle is operating, as identified by the GNSS sensor 858. Thus, for example, when operating in Europe, the CNN will seek to detect European sirens, and when in the United States, the CNN will seek to identify only North American sirens. Once an emergency vehicle is detected, with the assistance of the ultrasonic sensor 862, the control program can be used to execute the emergency vehicle safety routine to slow the vehicle, pull over, stop the vehicle, and / or idle the vehicle until the emergency vehicle passes.

[0143] The vehicle may include a CPU 818 (e.g., a discrete CPU or dCPU) that may be coupled to the SoC 804 via a high-speed interconnect (e.g., PCIe). The CPU 818 may include, for example, an X86 processor. The CPU 818 may be used to perform any of a variety of functions, including, for example, arbitrating potentially inconsistent results between ADAS sensors and the SoC 804, and / or monitoring the status and health of the controller 836 and / or the infotainment SoC 830.

[0144] The vehicle 800 may include a GPU 820 (e.g., a discrete GPU or dGPU) that may be coupled to the SoC 804 via a high-speed interconnect (e.g., NVIDIA's NVLINK). The GPU 820 may provide additional artificial intelligence functionality, for example, by executing redundant and / or different neural networks, and may be used to train and / or update the neural network based at least in part on input from sensors of the vehicle 800 (e.g., sensor data).

[0145] The vehicle 800 may also include a network interface 824, which may include one or more wireless antennas 826 (e.g., one or more wireless antennas for different communication protocols, such as a cellular antenna, a Bluetooth antenna, etc.). The network interface 824 can be used to enable wireless connections to the cloud (e.g., to a server 878 and / or other network devices), to other vehicles, and / or to computing devices (e.g., a passenger's client device) via the Internet. In order to communicate with other vehicles, a direct link can be established between the two vehicles, and / or an indirect link can be established (e.g., across a network and through the Internet). The direct link can be provided using a vehicle-to-vehicle communication link. The vehicle-to-vehicle communication link can provide the vehicle 800 with information about vehicles approaching the vehicle 800 (e.g., vehicles in front of, to the side of, and / or behind the vehicle 800). This functionality can be part of the cooperative adaptive cruise control functionality of the vehicle 800.

[0146] The network interface 824 may include a SoC that provides modulation and demodulation functions and enables the controller 836 to communicate over a wireless network. The network interface 824 may include an RF front-end for up-conversion from baseband to radio frequency and down-conversion from radio frequency to baseband. The frequency conversion may be performed by a well-known process and / or may be performed using a super-heterodyne process. In some examples, the RF front-end function may be provided by a separate chip. The network interface may include wireless functions for communicating over LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-wave, ZigBee, LoRaWAN, and / or other wireless protocols.

[0147] The vehicle 800 may also include data storage 828, which may include off-chip storage (e.g., outside the SoC 804). The data storage 828 may include one or more storage elements, including RAM, SRAM, DRAM, VRAM, flash memory, a hard disk, and / or other components and / or devices that can store at least one bit of data.

[0148] The vehicle 800 may also include a GNSS sensor 858. The GNSS sensor 858 (e.g., GPS, assisted GPS sensor, differential GPS (DGPS) sensor, etc.) is used to assist with mapping, perception, occupancy grid generation, and / or path planning functions. Any number of GNSS sensors 858 may be used, including, for example and without limitation, GPS using a USB connector with an Ethernet to serial (RS-232) bridge.

[0149] The vehicle 800 may also include a RADAR sensor 860. The RADAR sensor 860 can be used by the vehicle 800 for remote vehicle detection even in darkness and / or in adverse weather conditions. The RADAR functional safety level can be ASIL B. The RADAR sensor 860 can use CAN and / or bus 802 (e.g., to transmit data generated by the RADAR sensor 860) for control and access to object tracking data, and in some examples access Ethernet to access raw data. A variety of RADAR sensor types can be used. For example and without limitation, the RADAR sensor 860 can be suitable for front, rear, and side RADAR use. In some examples, a pulsed Doppler RADAR sensor is used.

[0150] The RADAR sensor 860 can include different configurations, such as long-range with a narrow field of view, short-range with a wide field of view, short-range side coverage, and so on. In some examples, the long-range RADAR can be used for adaptive cruise control functions. The long-range RADAR system can provide a wide field of view (e.g., within a range of 250m) achieved by two or more independent scans. The RADAR sensor 860 can help distinguish between static objects and moving objects and can be used by the ADAS system for emergency braking assistance and forward collision warnings. The long-range RADAR sensor may include a single-station multimode RADAR with multiple (e.g., six or more) fixed RADAR antennas and high-speed CAN and FlexRay interfaces. In the example with six antennas, the central four antennas can create a focused beam pattern that is designed to record the surroundings of the vehicle 800 at a higher rate with minimal traffic interference from adjacent lanes. The other two antennas can expand the field of view, making it possible to quickly detect vehicles entering or leaving the lane of the vehicle 800.

[0151] As an example, a medium-range RADAR system may include a range of up to 860m (front) or 80m (rear) and a field of view of up to 42 degrees (front) or 850 degrees (rear). A short-range RADAR system may include, but is not limited to, a RADAR sensor designed to be mounted on both ends of the rear bumper. When mounted on both ends of the rear bumper, such a RADAR sensor system can create two beams that continuously monitor the blind spots behind and beside the vehicle.

[0152] Short-range RADAR systems can be used in ADAS systems for blind spot detection and / or lane change assistance.

[0153] Vehicle 800 may also include ultrasonic sensors 862. Ultrasonic sensors 862, which may be located on the front, rear, and / or sides of vehicle 800, may be used for parking assistance and / or for creating and updating an occupancy grid. A variety of ultrasonic sensors 862 may be used, and different ultrasonic sensors 862 may have different detection ranges (e.g., 2.5 m, 4 m). Ultrasonic sensors 862 may operate at functional safety level ASIL B.

[0154] Vehicle 800 may include a LIDAR sensor 864. LIDAR sensor 864 may be used for object and pedestrian detection, emergency braking, collision avoidance, and / or other functions. LIDAR sensor 864 may be ASIL B functional safety level. In some examples, vehicle 800 may include multiple LIDAR sensors 864 (e.g., two, four, six, etc.) that may use Ethernet (e.g., to provide data to a Gigabit Ethernet switch).

[0155] In some examples, the LIDAR sensor 864 may be capable of providing a list of objects and their distances for a 360-degree field of view. Commercially available LIDAR sensors 864 may have, for example, an advertised range of approximately 800 meters, an accuracy of 2-3 cm, and support for 800 Mbps Ethernet connections. In some examples, one or more non-obtrusive LIDAR sensors 864 may be used. In such examples, the LIDAR sensor 864 may be implemented as a small device that can be embedded in the front, back, sides, and / or corners of the vehicle 800. In such examples, the LIDAR sensor 864 may provide a field of view of up to 120 degrees horizontally and 35 degrees vertically, with a range of 200 meters, even for low-reflectivity objects. The front-mounted LIDAR sensor 864 may be configured for a horizontal field of view between 45 and 135 degrees.

[0156] In some examples, LIDAR technologies such as 3D flash LIDAR may also be used. 3D flash LIDAR uses flashes of laser as an emission source to illuminate the vehicle's surroundings up to about 200 m. The flash LIDAR unit includes a receiver that records the laser pulse transmission time and reflected light on each pixel, which in turn corresponds to the range from the vehicle to the object. Flash LIDAR can allow highly accurate and distortion-free images of the surrounding environment to be generated with each laser flash. In some examples, four flash LIDAR sensors can be deployed, one on each side of the vehicle 800. Available 3D flash LIDAR systems include solid-state 3D staring array LIDAR cameras (e.g., non-scanning LIDAR devices) with no moving parts other than a fan. The flash LIDAR device can use 5 nanosecond Class I (eye-safe) laser pulses per frame and can capture the reflected laser light in the form of a 3D range point cloud and co-registered intensity data. By using flash LIDAR, and because flash LIDAR is a solid-state device with no moving parts, the LIDAR sensor 864 may be less susceptible to motion blur, vibration, and / or shock.

[0157] The vehicle may also include an IMU sensor 866. In some examples, the IMU sensor 866 may be located at the center of the rear axle of the vehicle 800. The IMU sensor 866 may include, for example and without limitation, an accelerometer, a magnetometer, a gyroscope, a magnetic compass, and / or other sensor types. In some examples, such as in a six-axis application, the IMU sensor 866 may include an accelerometer and a gyroscope, while in a nine-axis application, the IMU sensor 866 may include an accelerometer, a gyroscope, and a magnetometer.

[0158] In some embodiments, the IMU sensor 866 can be implemented as a miniature high-performance GPS-assisted inertial navigation system (GPS / INS) that combines micro-electromechanical systems (MEMS) inertial sensors, a high-sensitivity GPS receiver, and advanced Kalman filtering algorithms to provide estimates of position, velocity, and attitude. Thus, in some examples, the IMU sensor 866 can enable the vehicle 800 to estimate heading without the need for input from a magnetic sensor by directly observing and correlating velocity changes from the GPS to the IMU sensor 866. In some examples, the IMU sensor 866 and the GNSS sensor 858 can be combined into a single integrated unit.

[0159] The vehicle may include microphones 896 positioned in and / or around the vehicle 800. The microphones 896 may be used for, among other things, emergency vehicle detection and identification.

[0160] The vehicle may also include any number of camera types, including stereo cameras 868, wide angle cameras 870, infrared cameras 872, surround cameras 874, long and / or medium range cameras 898, and / or other camera types. These cameras may be used to capture image data around the entire periphery of the vehicle 800. The type of camera used depends on the embodiment and the requirements of the vehicle 800, and any combination of camera types may be used to provide the necessary coverage around the vehicle 800. Additionally, the number of cameras may vary depending on the embodiment. For example, the vehicle may include six cameras, seven cameras, ten cameras, twelve cameras, and / or another number of cameras. As an example and not limitation, the cameras may support Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet. Each of the cameras described herein may include a GMSL and / or Gigabit Ethernet network. Figure 8A and Figure 8B Described in more detail.

[0161] Vehicle 800 may also include a vibration sensor 842. Vibration sensor 842 can measure the vibration of a component of the vehicle, such as an axle. For example, changes in vibration can indicate changes in the road surface. In another example, when two or more vibration sensors 842 are used, the difference between the vibrations can be used to determine friction or slip on the road surface (e.g., when there is a vibration difference between a powered drive shaft and a freely rotating shaft).

[0162] The vehicle 800 may include an ADAS system 838. In some examples, the ADAS system 838 may include a SoC. The ADAS system 838 may include autonomous / adaptive / automatic cruise control (ACC), cooperative adaptive cruise control (CACC), forward collision warning (FCW), automatic emergency braking (AEB), lane departure warning (LDW), lane keeping assist (LKA), blind spot warning (BSW), rear cross traffic warning (RCTW), collision warning system (CWS), lane centering (LC), and / or other features and functions.

[0163] The ACC system can utilize RADAR sensor 860, LIDAR sensor 864, and / or cameras. The ACC system can include longitudinal ACC and / or lateral ACC. Longitudinal ACC monitors and controls the distance to the vehicle immediately in front of vehicle 800, automatically adjusting the vehicle speed to maintain a safe distance from the vehicle in front. Lateral ACC maintains distance and, when necessary, recommends that vehicle 800 change lanes. Lateral ACC is related to other ADAS applications such as LCA and CWS.

[0164] CACC uses information from other vehicles, which can be received from other vehicles indirectly via a wireless link or through a network connection (e.g., through the Internet) via the network interface 824 and / or the wireless antenna 826. A direct link can be provided by a vehicle-to-vehicle (V2V) communication link, while an indirect link can be an infrastructure-to-vehicle (I2V) communication link. Typically, the V2V communication concept provides information about the vehicle immediately ahead (e.g., the vehicle immediately ahead of the vehicle 800 and in the same lane as it), while the I2V communication concept provides information about traffic further ahead. The CACC system can include either or both of the I2V and V2V information sources. Given information about the vehicle ahead of the vehicle 800, CACC can be more reliable, and it has the potential to improve the smoothness of traffic flow and reduce road congestion.

[0165] The FCW system is designed to alert the driver to hazards so that the driver can take corrective action. The FCW system uses a front-facing camera and / or RADAR sensor 860 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to driver feedback such as a display, speaker, and / or vibrating component. The FCW system can provide warnings in the form of, for example, audible, visual warnings, vibrations, and / or rapid brake pulses.

[0166] The AEB system detects an impending forward collision with another vehicle or other object and can automatically apply the brakes if the driver does not take corrective action within specified time or distance parameters. The AEB system can use a front-facing camera and / or RADAR sensor 860 coupled to a dedicated processor, DSP, FPGA and / or ASIC. When the AEB system detects a hazard, it typically first alerts the driver to take corrective action to avoid the collision, and if the driver does not take corrective action, the AEB system can automatically apply the brakes in an effort to prevent or at least mitigate the effects of the predicted collision. The AEB system can include technologies such as dynamic brake support and / or collision approach braking.

[0167] The LDW system provides visual, audible, and / or tactile warnings, such as steering wheel or seat vibrations, to alert the driver when the vehicle 800 crosses a lane marking. When the driver indicates an intention to leave the lane by activating a turn signal, the LDW system is deactivated. The LDW system may utilize a front-facing camera coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0168] The LKA system is a variation of the LDW system. If the vehicle 800 begins to leave its lane, the LKA system provides steering input or braking to correct the vehicle 800.

[0169] The BSW system detects and warns the driver of vehicles in the car's blind spot. The BSW system can provide visual, audible, and / or tactile alerts to indicate that merging or changing lanes is unsafe. The system can provide additional warnings when the driver uses a turn signal. The BSW system can use a rear-facing camera and / or RADAR sensor 860 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0170] The RCTW system can provide visual, audible, and / or tactile notifications when an object is detected outside the range of the rear-mounted camera while the vehicle 800 is in reverse. Some RCTW systems include automatic emergency braking (AEB) to ensure that the vehicle brakes are applied to avoid a collision. The RCTW system can use one or more rear-mounted RADAR sensors 860 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0171] Conventional ADAS systems may be prone to false positive results, which may be annoying and distracting to the driver, but are typically not catastrophic because the ADAS system alerts the driver and allows the driver to decide whether a safety condition actually exists and take action accordingly. However, in the autonomous vehicle 800, in the event of conflicting results, the vehicle 800 itself must decide whether to pay attention to the results from the main computer or the auxiliary computer (e.g., the first controller 836 or the second controller 836). For example, in some embodiments, the ADAS system 838 can be a backup and / or auxiliary computer for providing perception information to the backup computer rationality module. The backup computer rationality monitor can run redundant and diverse software on hardware components to detect failures in perception and dynamic driving tasks. The output from the ADAS system 838 can be provided to the supervisory MCU. If the outputs from the main computer and the auxiliary computer conflict, the supervisory MCU must determine how to reconcile the conflict to ensure safe operation.

[0172] In some examples, the primary computer can be configured to provide a confidence score to the supervisory MCU, indicating the primary computer's confidence in the selected result. If the confidence score exceeds a threshold, the supervisory MCU can follow the primary computer's direction, regardless of whether the secondary computer provides conflicting or inconsistent results. In the event that the confidence score does not meet the threshold and the primary and secondary computers indicate different results (e.g., a conflict), the supervisory MCU can arbitrate between these computers to determine the appropriate result.

[0173] The supervisory MCU can be configured to run a neural network that is trained and configured to determine conditions under which the secondary computer provides a false alarm based, at least in part, on outputs from the primary computer and the secondary computer. Thus, the neural network in the supervisory MCU can learn when the output of the secondary computer can be trusted and when it cannot. For example, when the secondary computer is a RADAR-based FCW system, the neural network in the supervisory MCU can learn when the FCW system is identifying a metal object that is not actually a danger, such as a drain grate or manhole cover, which triggers an alarm. Similarly, when the secondary computer is a camera-based LDW system, the neural network in the supervisory MCU can learn to disregard the LDW when a cyclist or pedestrian is present and lane departure is actually the safest strategy. In embodiments that include a neural network running on the supervisory MCU, the supervisory MCU can include at least one of a DLA or a GPU suitable for running the neural network with associated memory. In a preferred embodiment, the supervisory MCU can include and / or be included as a component of the SoC 804.

[0174] In other examples, the ADAS system 838 may include an auxiliary computer that uses traditional computer vision rules to perform ADAS functions. In this way, the auxiliary computer can use classic computer vision rules (if-then), and the presence of a neural network in the supervisory MCU can improve reliability, safety, and performance. For example, diverse implementations and intentional non-identity make the entire system more fault-tolerant, especially with respect to failures caused by software (or software-hardware interface) functions. For example, if there is a software vulnerability or bug in the software running on the main computer and the non-identical software code running on the auxiliary computer provides the same overall result, the supervisory MCU can be more confident that the overall result is correct and that the vulnerability in the software or hardware on the main computer did not cause a substantial error.

[0175] In some examples, the output of the ADAS system 838 can be fed into the primary computer's perception block and / or the primary computer's dynamic driving task block. For example, if the ADAS system 838 indicates a forward collision warning due to an object immediately ahead, the perception block can use this information when identifying the object. In other examples, the secondary computer can have its own neural network that is trained and thus reduces the risk of false positives as described herein.

[0176] The vehicle 800 may also include an infotainment SoC 830 (e.g., an in-vehicle infotainment system (IVI)). Although illustrated and described as an SoC, the infotainment system may not be an SoC and may include two or more separate components. The infotainment SoC 830 may include a combination of hardware and software that can be used to provide audio (e.g., music, personal digital assistant, navigation instructions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), phone (e.g., hands-free calling), network connectivity (e.g., LTE, WiFi, etc.), and / or information services (e.g., a navigation system, rear parking assistance, radio data system, vehicle-related information such as fuel level, total distance covered, brake fuel level, oil level, door open / closed, air filter information, etc.) to the vehicle 800. For example, the infotainment SoC 830 may include a radio, a disc player, a navigation system, a video player, USB and Bluetooth connectivity, an onboard computer, in-vehicle entertainment, WiFi, steering wheel audio controls, hands-free voice controls, a head-up display (HUD), an HMI display 834, a telematics device, a control panel (e.g., for controlling and / or interacting with various components, features, and / or systems), and / or other components. The infotainment SoC 830 may further be used to provide information (e.g., visual and / or auditory) to a user of the vehicle, such as information from an ADAS system 838, autonomous driving information such as planned vehicle maneuvers, trajectories, surrounding environment information (e.g., intersection information, vehicle information, road information, etc.), and / or other information.

[0177] The infotainment SoC 830 may include GPU functionality. The infotainment SoC 830 may communicate with other devices, systems, and / or components of the vehicle 800 via a bus 802 (e.g., a CAN bus, Ethernet, etc.). In some examples, the infotainment SoC 830 may be coupled to a supervisory MCU so that in the event of a failure of a primary controller 836 (e.g., a primary and / or backup computer of the vehicle 800), the infotainment system's GPU may perform some self-driving functions. In such an example, the infotainment SoC 830 may place the vehicle 800 in a driver-safe parking mode as described herein.

[0178] The vehicle 800 may also include an instrument cluster 832 (e.g., a digital instrument panel, an electronic instrument cluster, a digital instrument panel, etc.). The instrument cluster 832 may include a controller and / or a supercomputer (e.g., a separate controller or a supercomputer). The instrument cluster 832 may include a set of instruments, such as a speedometer, fuel level, oil pressure, a tachometer, an odometer, a turn indicator, a shift position indicator, a seat belt warning light, a parking brake warning light, an engine check light, airbag (SRS) system information, lighting controls, safety system controls, navigation information, etc. In some examples, information can be displayed and / or shared between the infotainment SoC 830 and the instrument cluster 832. In other words, the instrument cluster 832 may be included as part of the infotainment SoC 830, or vice versa.

[0179] Figure 8D For cloud-based servers and Figure 8A 880). Depending on the embodiment, each of the servers 878 can include any number of GPUs 884, CPUs 880, and / or PCIe switches. For example, each of the servers 878 can include eight, sixteen, thirty-two, and / or more GPUs 884.

[0180] Server 878 can receive image data from a vehicle via network 890 that represents images showing unexpected or changed road conditions, such as recently begun road construction. Server 878 can transmit neural network 892, updated neural network 892, and / or map information 894, including information about traffic and road conditions, via network 890 and to the vehicle. Updates to map information 894 can include updates to HD map 822, such as information about construction sites, potholes, curves, flooding, or other obstacles. In some examples, neural network 892, updated neural network 892, and / or map information 894 can have been generated from new training and / or data received from any number of vehicles in the environment and / or based on experience with training performed at a data center (e.g., using server 878 and / or other servers).

[0181] Server 878 can be used to train a machine learning model (e.g., a neural network) based on training data. The training data can be generated by the vehicle and / or can be generated in simulation (e.g., using a game engine). In some examples, the training data is labeled (e.g., in cases where the neural network benefits from supervised learning) and / or undergoes other preprocessing, while in other examples, the training data is not labeled and / or preprocessed (e.g., in cases where the neural network does not require supervised learning). Training can be performed according to any one or more categories of machine learning techniques, including but not limited to the following categories: supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, joint learning, transfer learning, feature learning (including principal component and cluster analysis), multilinear subspace learning, manifold learning, representation learning (including alternative dictionary learning), rule-based machine learning, anomaly detection, and any variants or combinations thereof. Once the machine learning model is trained, the machine learning model can be used by the vehicle (e.g., transmitted to the vehicle via network 890), and / or the machine learning model can be used by server 878 to remotely monitor the vehicle.

[0182] In some examples, server 878 can receive data from the vehicle and apply the data to the latest real-time neural network for real-time intelligent reasoning. Server 878 can include a deep learning supercomputer and / or a dedicated AI computer powered by GPU 884, such as the DGX and DGX Station machines developed by NVIDIA. However, in some examples, server 878 can include the deep learning infrastructure of a data center using only CPU power.

[0183] The deep learning infrastructure of server 878 may be capable of rapid real-time inference and may use this capability to assess and verify the health of the processors, software, and / or associated hardware in vehicle 800. For example, the deep learning infrastructure may receive periodic updates from vehicle 800, such as an image sequence and / or objects that vehicle 800 has located in the image sequence (e.g., via computer vision and / or other machine learning object classification techniques). The deep learning infrastructure may run its own neural network to identify objects and compare them to the objects identified by vehicle 800, and if the results do not match and the infrastructure concludes that the AI ​​in vehicle 800 has malfunctioned, server 878 may transmit a signal to vehicle 800 instructing the vehicle's 800 fail-safe computer to take control, notify passengers, and complete a safe parking maneuver.

[0184] For inference, the server 878 may include a GPU 884 and one or more programmable inference accelerators (e.g., NVIDIA's TensorRT). The combination of GPU-powered servers and inference acceleration can enable real-time responses. In other examples, such as where performance is less important, CPU, FPGA, and other processor-powered servers can be used for inference.

[0185] Example computing device

[0186] Figure 9 9 is a block diagram of an example computing device 900 suitable for implementing some embodiments of the present disclosure. Computing device 900 may include an interconnect system 902 that directly or indirectly couples the following devices: memory 904, one or more central processing units (CPUs) 906, one or more graphics processing units (GPUs) 908, a communication interface 910, input / output (I / O) ports 912, input / output components 914, a power supply 916, one or more presentation components 918 (e.g., display(s)), and one or more logic units 920. In at least one embodiment, computing device(s) 900 may include one or more virtual machines (VMs), and / or any of its components may include virtual components (e.g., virtual hardware components). For non-limiting examples, one or more of GPUs 908 may include one or more vGPUs, one or more of CPUs 906 may include one or more vCPUs, and / or one or more of logic units 920 may include one or more virtual logic units. As such, computing device(s) 900 may include discrete components (e.g., a full GPU dedicated to computing device 900), virtual components (e.g., a portion of a GPU dedicated to computing device 900), or a combination thereof.

[0187] although Figure 9 The various blocks of are shown as being connected via interconnect system 902 using wires, but this is not intended to be limiting and is provided for clarity only. For example, in some embodiments, presentation component 918 (such as a display device) may be considered to be I / O component 914 (e.g., if the display is a touch screen). As another example, CPU 906 and / or GPU 908 may include memory (e.g., memory 904 may represent a storage device in addition to the memory of GPU 908, CPU 906, and / or other components). In other words, Figure 9 The computing devices are illustrative only. No distinction is made between such categories as "workstation," "server," "laptop," "desktop," "tablet," "client device," "mobile device," "handheld device," "game console," "electronic control unit (ECU)," "virtual reality system," and / or other device or system types, as all are considered Figure 9 within the range of computing devices.

[0188] Interconnect system 902 can represent one or more links or buses, such as address bus, data bus, control bus or its combination.Interconnect system 902 can include one or more bus or link types, such as industry standard architecture (ISA) bus, extended industry standard architecture (EISA) bus, video electronics standard association (VESA) bus, peripheral component interconnect (PCI) bus, peripheral component interconnect express (PCIe) bus and / or another type of bus or link. In some embodiments, there is a direct connection between components. As an example, CPU 906 can be directly connected to memory 904. Further, CPU 906 can be directly connected to GPU 908. In the case of direct or point-to-point connection between components, interconnect system 902 can include PCIe link to perform connection. In these examples, PCI bus does not need to be included in computing device 900.

[0189] Memory 904 may include any of a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by computing device 900. Computer-readable media can include volatile and non-volatile media, as well as removable and non-removable media. By way of example and not limitation, computer-readable media can include computer storage media and communication media.

[0190] Computer storage media may include volatile and nonvolatile media and / or removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, and / or other data types. For example, memory 904 may store computer-readable instructions (e.g., representing (one or more) programs and / or (one or more) program elements, such as an operating system). Computer storage media may include, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by the computing device 900. As used herein, computer storage media does not include signals themselves.

[0191] Computer storage media can embody computer-readable instructions, data structures, program modules, and / or other data types in a modulated data signal such as a carrier wave or other transport mechanism, and includes any information delivery media. The term "modulated data signal" may refer to a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, computer storage media can include wired media (such as a wired network or direct-wired connection) and wireless media (such as acoustic, RF, infrared, and other wireless media). Combinations of any of the above should also be included within the scope of computer-readable media.

[0192] The CPU 906 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 900 to perform one or more of the methods and / or processes described herein. The CPUs 906 may each include one or more cores (e.g., one, two, four, eight, twenty-eight, seventy-two, etc.) capable of handling numerous software threads simultaneously. The CPU 906 may include any type of processor and may include different types of processors depending on the type of computing device 900 being implemented (e.g., a processor with fewer cores for a mobile device and a processor with more cores for a server). For example, depending on the type of computing device 900, the processor may be an Advanced RISC Machine (ARM) processor implemented using Reduced Instruction Set Computing (RISC) or an x86 processor implemented using Complex Instruction Set Computing (CISC). The computing device 900 may also include one or more CPUs 906 in addition to one or more microprocessors or supplemental coprocessors (such as a math coprocessor).

[0193] In addition to or in lieu of CPU(s) 906, GPU(s) 908 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 900 to perform one or more of the methods and / or processes described herein. One or more of the GPUs 908 may be integrated GPUs (e.g., with one or more of the CPUs 906) and / or one or more of the GPUs 908 may be discrete GPUs. In embodiments, one or more of the GPUs 908 may be coprocessors for one or more of the CPUs 906. The GPU 908 may be used by the computing device 900 to render graphics (e.g., 3D graphics) or perform general-purpose computations. For example, the GPU 908 may be used for general-purpose computing on a GPU (GPGPU). The GPU 908 may include hundreds or thousands of cores capable of handling hundreds or thousands of software threads simultaneously. The GPU 908 may generate pixel data for an output image in response to rendering commands (e.g., rendering commands received from the CPU 906 via a host interface). The GPU 908 may include graphics memory (e.g., display memory) for storing pixel data or any other suitable data (e.g., GPGPU data). The display memory may be included as part of the memory 904. The GPU 908 may include two or more GPUs operating in parallel (e.g., via a link). The link may connect the GPUs directly (e.g., using NVLINK) or may connect the GPUs through a switch (e.g., using NVSwitch). When combined, each GPU 908 may generate pixel data or GPGPU data for different portions or for different outputs (e.g., a first GPU for a first image and a second GPU for a second image). Each GPU may include its own memory or may share memory with other GPUs.

[0194] In addition to or in lieu of the CPU 906 and / or GPU 908, the logic unit 920 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 900 to perform one or more of the methods and / or processes described herein. In embodiments, the CPU(s) 906, the GPU(s) 908, and / or the logic unit(s) 920 may perform any combination of methods, processes, and / or portions thereof, either discretely or jointly. One or more of the logic units 920 may be part of and / or integrated into one or more of the CPU 906 and / or GPU 908, and / or one or more of the logic units 920 may be discrete components or otherwise external to the CPU 906 and / or GPU 908. In embodiments, one or more of logic units 920 may be a co-processor of one or more of CPUs 906 and / or one or more of GPUs 908 .

[0195] In some embodiments, one or more aspects of the pose dataset collection system 120 and / or the head pose prediction system 500 may be implemented using code executed on one or more of the CPU 906 , the GPU 908 , and / or the logic unit 920 .

[0196] Examples of logic unit 920 include one or more processing cores and / or components thereof, such as a data processing unit (DPU), a tensor core (TC), a tensor processing unit (TPU), a pixel vision core (PVC), a vision processing unit (VPU), a graphics processing cluster (GPC), a texture processing cluster (TPC), a streaming multiprocessor (SM), a tree transverse unit (TTU), an artificial intelligence accelerator (AIA), a deep learning accelerator (DLA), an arithmetic logic unit (ALU), an application-specific integrated circuit (ASIC), a floating point unit (FPU), an input / output (I / O) element, a peripheral component interconnect (PCI) or a peripheral component interconnect express (PCIe) element, etc.

[0197] The communication interface 910 may include one or more receivers, transmitters, and / or transceivers that enable the computing device 900 to communicate with other computing devices via an electronic communication network (including wired and / or wireless communications). The communication interface 910 may include components and functions that implement communication over any of a number of different networks, such as a wireless network (e.g., Wi-Fi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), a wired network (e.g., via Ethernet or Wi-Fi), a low-power wide-area network (e.g., LoRaWAN, SigFox, etc.), and / or the Internet. In one or more embodiments, the one or more logic units 920 and / or the communication interface 910 may include one or more data processing units (DPUs) to transmit data received over the network and / or through the interconnect system 902 directly to one or more GPUs 908 (e.g., memory).

[0198] The I / O ports 912 can enable the computing device 900 to be logically coupled to other devices including I / O components 914, (one or more) presentation components 918, and / or other components, some of which can be built into (e.g., integrated into) the computing device 900. Illustrative I / O components 914 include a microphone, a mouse, a keyboard, a joystick, a game pad, a game controller, a satellite dish, a scanner, a printer, a wireless device, and the like. The I / O components 914 can provide a natural user interface (NUI) that processes air gestures, voice, or other physiological input generated by the user. In some cases, the input can be transmitted to an appropriate network element for further processing. The NUI can implement any combination of voice recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition on and near the screen, air gestures, head and eye tracking, and touch recognition (as described in more detail below) associated with the display of the computing device 900. The computing device 900 may include a depth camera for gesture detection and recognition, such as a stereo camera system, an infrared camera system, an RGB camera system, touch screen technology, and combinations thereof. Additionally, computing device 900 may include an accelerometer or gyroscope that enables detection of motion (e.g., as part of an inertial measurement unit (IMU)). In some examples, computing device 900 may use the output of the accelerometer or gyroscope to render immersive augmented or virtual reality.

[0199] The power supply 916 may include a hardwired power supply, a battery power supply, or a combination thereof. The power supply 916 may provide power to the computing device 900 to enable the components of the computing device 900 to operate.

[0200] The presentation component 918 may include a display (e.g., a monitor, a touch screen, a television screen, a head-up display (HUD), other display types, or a combination thereof), speakers, and / or other presentation components. The presentation component 918 may receive data from other components (e.g., GPU 908, CPU 906, DPU, etc.) and output the data (e.g., as images, video, sound, etc.).

[0201] Sample Data Center

[0202] Figure 10 An example data center 1000 that can be used in at least one embodiment of the present disclosure is shown. The data center 1000 can include a data center infrastructure layer 1010, a framework layer 1020, a software layer 1030, and / or an application layer 1040.

[0203] like Figure 10 As shown, the data center infrastructure layer 1010 may include a resource coordinator 1012, grouped computing resources 1014, and node computing resources ("node CRs") 1016(1)-1016(N), where "N" represents any complete positive integer. In at least one embodiment, the node CRs 1016(1)-1016(N) may include, but are not limited to, any number of central processing units (CPUs) or other processors (including DPUs, accelerators, field programmable gate arrays (FPGAs), graphics processors or graphics processing units (GPUs), etc.), memory devices (e.g., dynamic read-only memories), storage devices (e.g., solid-state or disk drives), network input / output (NW I / O) devices, network switches, virtual machines (VMs), power modules and / or cooling modules, etc. In some embodiments, one or more node CRs from the node CRs 1016(1)-1016(N) may correspond to servers having one or more of the above-mentioned computing resources. Furthermore, in some embodiments, nodes CR1016(1)-10161(N) may include one or more virtual components, such as vGPUs, vCPUs, etc., and / or one or more of nodes CR1016(1)-1016(N) may correspond to virtual machines (VMs). In some embodiments, one or more aspects of the pose dataset collection system 120 and / or the head pose prediction system 500 may be implemented using code executed on one or more of nodes CR1016(1)-1016(N).

[0204] In at least one embodiment, the grouped computing resources 1014 may include separate groups of node CR1016 housed in one or more racks (not shown), or many racks housed in data centers at different geographical locations (also not shown). The separate groups of node CR1016 within the grouped computing resources 1014 may include grouped computing, network, memory, or storage resources that can be configured or allocated to support one or more workloads. In at least one embodiment, several node CR1016 including CPUs, GPUs, DPUs, and / or other processors may be grouped in one or more racks to provide computing resources to support one or more workloads. One or more racks may also include any number of power modules, cooling modules, and / or network switches in any combination.

[0205] The resource coordinator 1012 may configure or otherwise control one or more node CRs 1016(1)-1016(N) and / or grouped computing resources 1014. In at least one embodiment, the resource coordinator 1012 may comprise a software design infrastructure (SDI) management entity for the data center 1000. The resource coordinator 1012 may comprise hardware, software, or some combination thereof.

[0206] In at least one embodiment, Figure 10 As shown, the framework layer 1020 may include a job scheduler 1033, a configuration manager 1034, a resource manager 1036 and / or a distributed file system 1038. The framework layer 1020 may include a framework that supports software 1032 of the software layer 1030 and / or one or more applications 1042 of the application layer 1040. The software 1032 or the application 1042 may include network-based service software or applications, such as those provided by Amazon Web Services, Google Cloud, and Microsoft Azure. The framework layer 1020 may be, but is not limited to, a free and open source software network application framework (such as Apache Spark) that can utilize the distributed file system 1038 for large-scale data processing (e.g., "big data"). TM(hereinafter referred to as "Spark"). In at least one embodiment, the job scheduler 1033 may include a Spark driver to facilitate scheduling workloads supported by the different layers of the data center 1000. The configuration manager 1034 may be capable of configuring the different layers, such as the software layer 1030 and the framework layer 1020 (which includes Spark and a distributed file system 1038 for supporting large-scale data processing). The resource manager 1036 may be capable of managing clustered or grouped computing resources that are mapped to the distributed file system 1038 and the job scheduler 1032 or allocated to support the distributed file system 1038 and the job scheduler 1033. In at least one embodiment, the clustered or grouped computing resources may include the grouped computing resources 1014 at the data center infrastructure layer 1010. The resource manager 1036 may coordinate with the resource coordinator 1012 to manage these mapped or allocated computing resources.

[0207] In at least one embodiment, the software 1032 included in the software layer 1030 may include software used by at least a portion of the node CRs 1016(1)-1016(N), the grouped computing resources 1014, and / or the distributed file system 1038 of the framework layer 1020. The one or more types of software may include, but are not limited to, Internet web search software, email virus scanning software, database software, and streaming video content software.

[0208] In at least one embodiment, the applications 1042 included in the application layer 1040 may include one or more types of applications used by at least a portion of the node CRs 1016(1)-1016(N), the grouped computing resources 1014, and / or the distributed file system 1038 of the framework layer 1020. The one or more types of applications may include, but are not limited to, any number of genomic applications, cognitive computing, and machine learning applications, including training or inference software, machine learning framework software (e.g., PyTorch, TensorFlow, Caffe, etc.), and / or other machine learning applications used in conjunction with one or more embodiments.

[0209] In at least one embodiment, any of the configuration manager 1034, the resource manager 1036, and the resource coordinator 1012 can implement any number and type of self-modification actions based on any amount and type of data acquired in any technically feasible manner. The self-modification actions can save the data center operator of the data center 1000 from making potentially poor configuration decisions and potentially avoiding underutilized and / or poorly performing portions of the data center.

[0210] According to one or more embodiments described herein, the data center 1000 may include tools, services, software, or other resources to train one or more machine learning models or to predict or infer information using one or more machine learning models. For example, the machine learning model(s) may be trained by computing weight parameters according to a neural network architecture using the software and / or computing resources described above with respect to the data center 1000. In at least one embodiment, a trained or deployed machine learning model corresponding to one or more neural networks may be used to infer or predict information using the resources described above with respect to the data center 1000 using weight parameters computed using one or more training techniques, such as, but not limited to, those described herein. One or more aspects of the head pose prediction training system 400 may be implemented using the data center 1000 to train the head pose prediction model 410.

[0211] In at least one embodiment, data center 1000 may use a CPU, an application-specific integrated circuit (ASIC), a GPU, an FPGA, and / or other hardware (or virtual computing resources corresponding thereto) to perform training and / or inference using the aforementioned resources. In addition, one or more of the software and / or hardware resources described above may be configured to allow a user to train or perform inference services on information, such as image recognition, speech recognition, or other artificial intelligence services.

[0212] Sample network environment

[0213] A network environment suitable for implementing embodiments of the present disclosure may include one or more client devices, servers, network attached storage (NAS), other backend devices, and / or other device types. The client devices, servers, and / or other device types (e.g., each device) may be configured to: Figure 9 The data center 1000 may be implemented on one or more instances of the computing device(s) 900 - for example, each device may include similar components, features, and / or functionality of the computing device(s) 900. In addition, in the case of implementing a backend device (e.g., a server, NAS, etc.), the backend device may be included as part of the data center 1000, an example of which is described herein with respect to FIG. Figure 10 Describe in more detail.

[0214] The components of the network environment can communicate with each other via a network, which can be wired, wireless, or both. The network can include multiple networks or one of multiple networks. For example, the network can include one or more wide area networks (WANs), one or more local area networks (LANs), one or more public networks (such as the Internet and / or the Public Switched Telephone Network (PSTN)), and / or one or more private networks. In the case where the network includes a wireless telecommunications network, components such as base stations, communication towers, or even access points (and other components) can provide wireless connectivity.

[0215] Compatible network environments may include one or more peer-to-peer network environments (in which case the server may not be included in the network environment) and one or more client-server network environments (in which case one or more servers may be included in the network environment). In a peer-to-peer network environment, the functionality described herein for the server may be implemented on any number of client devices.

[0216] In at least one embodiment, the network environment may include one or more cloud-based network environments, distributed computing environments, combinations thereof, and the like. The cloud-based network environment may include a framework layer, a job scheduler, a resource manager, and a distributed file system implemented on one or more servers, which may include one or more core network servers and / or edge servers. The framework layer may include software supporting the software layer and / or a framework for one or more applications at the application layer. The software or application may include network-based service software or applications, respectively. In an embodiment, one or more client devices may use network-based service software or applications (e.g., by accessing the service software and / or application via one or more application programming interfaces (APIs)). The framework layer may be, but is not limited to, a free and open source software network application framework that can use a distributed file system for large-scale data processing (e.g., "big data").

[0217] A cloud-based network environment can provide cloud computing and / or cloud storage that performs any combination of the computing and / or data storage functions described herein (or one or more portions thereof). Any of these different functions can be distributed across multiple locations from a central or core server (e.g., one or more data centers that can be distributed across a state, region, country, global, etc.). If the connection to the user (e.g., client device) is relatively close to an edge server, the core server can assign at least a portion of the functionality to the edge server. A cloud-based network environment can be private (e.g., limited to a single organization), public (e.g., available to many organizations), and / or a combination thereof (e.g., a hybrid cloud environment).

[0218] The client device(s) may include the Figure 9 At least some of the components, features, and functionality of the described example computing device(s) 900. By way of example and not limitation, the client device may be implemented as a personal computer (PC), a laptop computer, a mobile device, a smartphone, a tablet computer, a smartwatch, a wearable computer, a personal digital assistant (PDA), an MP3 player, a virtual reality headset, a global positioning system (GPS) or device, a video player, a camera, a surveillance device or system, a vehicle, a boat, a spacecraft, a virtual machine, a drone, a robot, a handheld communication device, a hospital device, a gaming device or system, an entertainment system, a vehicle computer system, an embedded system controller, a remote control, an appliance, a consumer electronic device, a workstation, an edge device, any combination of the depicted devices, or any other suitable device.

[0219] The present disclosure can be described in the general context of machine-usable instructions or computer code executed by a computer or other machine such as a personal digital assistant or other handheld device, including computer-executable instructions such as program modules. Generally, program modules including routines, programs, objects, components, data structures, etc. refer to code that performs a specific task or implements a specific abstract data type. The present disclosure can be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, more specialized computing devices, etc. The present disclosure can also be practiced in a distributed computing environment where tasks are performed by remote processing devices linked through a communication network.

[0220] As used herein, the phrase "and / or" with respect to two or more elements should be interpreted as referring to only one element or combination of elements. For example, "element A, element B, and / or element C" may include only element A, only element B, only element C, element A and element B, element A and element C, element B and element C, or elements A, B, and C. In addition, "at least one of element A or element B" may include at least one of element A, at least one of element B, or at least one of element A and at least one of element B. Further, "at least one of element A and element B" may include at least one of element A, at least one of element B, or at least one of element A and at least one of element B.

[0221] The subject matter of the present disclosure is described in detail herein to meet statutory requirements. However, the description itself is not intended to limit the scope of the present disclosure. On the contrary, the present inventors have contemplated that the claimed subject matter may also be embodied in other ways to include steps that are different from the steps described herein in conjunction with other current or future technologies, or combinations of similar steps. Moreover, although the terms "step" and / or "block" may be used herein to imply different elements of the method employed, these terms should not be interpreted as implying any particular order among or between the various steps disclosed herein, unless the order of the steps is explicitly described.

Claims

1. One or more processors, including processing circuitry configured to: determining at least one three-dimensional (3D) measurement corresponding to a head pose of an occupant of the vehicle based at least on a deviation between a first 3D point cloud representation of a model customized for an occupant of the vehicle and a second 3D point cloud representation of at least a portion of the occupant of the vehicle based at least on depth image data; capturing optical image data representative of at least a portion of the occupant of the vehicle, wherein the optical image data is synchronized with the depth image data; transforming the at least one 3D measurement to a reference frame of the optical image data to generate at least one 3D ground truth measurement; as well as The machine learning model is updated to generate a prediction of a 3D pose of at least a portion of the occupant based at least on the optical image data and the at least one 3D ground truth measurement.

2. The one or more processors of claim 1 , wherein the circuitry is further configured to: The model customized for the occupant of the vehicle is generated based at least on optimizing a generic model using a 3D point cloud representation of at least a portion of the occupant of the vehicle.

3. The one or more processors of claim 2, wherein the circuitry is further configured to: Instructions are output to the occupant of the vehicle to rotate their head during a registration process to generate the first 3D point cloud representation.

4. The one or more processors of claim 1 , wherein the circuitry is further configured to: A depth sensor and one or more optical image sensors are synchronously operated to capture the depth image data and the optical image data.

5. The one or more processors of claim 4, wherein the circuitry is further configured to: The depth sensor and the one or more optical image sensors are operated in synchronization based on a synchronization signal generated by the vehicle.

6. The one or more processors of claim 4, wherein the circuitry is further configured to: The depth sensor is triggered to capture the depth image data based on an offset synchronization with triggering the one or more optical image sensors to capture the optical image data.

7. The one or more processors of claim 1 , wherein the circuitry is further configured to: The machine learning model is further updated based on inputs including the model customized for the occupants of the vehicle, the inputs being transformed to the reference frame of the optical image data.

8. The one or more processors of claim 1 , wherein the circuitry is further configured to: The at least one 3D measurement result is converted to the reference frame of the optical image data based on applying one or more extrinsic calibration parameters, the one or more extrinsic calibration parameters representing one or more rotation-translation RT transformations between a depth sensor capturing the depth image data and one or more optical image sensors capturing the optical image data.

9. The one or more processors of claim 1 , wherein the circuitry is further configured to: A deviation between the first and second 3D point cloud representations of at least a portion of a head of the occupant of the vehicle is calculated based at least on an iterative closest point algorithm.

10. The one or more processors of claim 1 , wherein the circuitry is further configured to: Generating training samples to train the machine learning model, wherein the training samples include image frames based on at least the optical image data and 3D ground truth labels based on the at least one 3D ground truth measurement result.

11. The one or more processors of claim 1 , wherein the optical image data comprises image frames captured by a plurality of cameras within the vehicle having different viewpoints of the occupants of the vehicle.

12. The one or more processors of claim 1 , wherein the one or more processors are included in at least one of: control systems for autonomous or semi-autonomous machines; Perception systems for autonomous or semi-autonomous machines; a system for performing simulation operations; Systems for performing digital twin operations; a system for performing light transport simulations; A system for performing collaborative content creation of three-dimensional assets; Systems for performing deep learning operations; systems for performing remote operations; A system for performing real-time streaming; Systems for generating or presenting one or more of augmented reality content, virtual reality content, or mixed reality content; Systems implemented using edge devices; Systems implemented using robots; Systems for performing conversational artificial intelligence (AI) operations; A system implementing one or more language models; A system implementing one or more large language models (LLMs); Systems for generating synthetic data; Systems for generating synthetic data using AI; A system comprising one or more virtual machines VM; A system implemented at least in part in a data center; or A system implemented at least in part using cloud computing resources.

13. A system comprising one or more processors, wherein the one or more processors are configured to: capturing optical image data representative of at least a portion of an occupant of the vehicle; and generating a prediction of a three-dimensional (3D) pose corresponding to the occupant based at least on the optical image data using a machine learning model, wherein the machine learning model is configured to infer the predicted 3D pose based at least on: 3D ground truth measurement data representing a deviation between a first 3D point cloud representation of a model customized for a training subject and a second 3D point cloud representation of at least a portion of the training subject based at least on depth image data; and Optical image data representing at least a portion of the training subject, wherein the optical image data is synchronized with the depth image data.

14. The system of claim 13, wherein the one or more processors are further configured to: The vehicle is controlled to perform one or more operations based at least on the predicted 3D pose.

15. The system of claim 13, wherein the one or more processors are further configured to: The model customized for the training subject is generated based at least on optimizing a generic model using a 3D point cloud representation of at least a portion of the training subject.

16. The system of claim 13, wherein the 3D ground truth measurement data is transformed to a reference frame of the optical image data based on applying one or more extrinsic calibration parameters, the one or more extrinsic calibration parameters representing one or more rotation-translation (RT) transformations between a depth sensor capturing the depth image data and one or more optical image sensors capturing the optical image data.

17. The system of claim 13, wherein the one or more processors are further configured to: One or more extrinsic calibration parameters are applied to transform the predicted 3D pose to a global reference frame of the vehicle.

18. The system of claim 13, wherein the one or more processors are further configured to: A gaze direction of the occupant of the vehicle is calculated based on at least the predicted 3D pose.

19. The system of claim 13, wherein the system is included in at least one of the following: control systems for autonomous or semi-autonomous machines; Perception systems for autonomous or semi-autonomous machines; a system for performing simulation operations; Systems for performing digital twin operations; a system for performing light transport simulations; A system for performing collaborative content creation of three-dimensional assets; Systems for performing deep learning operations; systems for performing remote operations; A system for performing real-time streaming; Systems for generating or presenting one or more of augmented reality content, virtual reality content, or mixed reality content; Systems implemented using edge devices; Systems implemented using robots; Systems for performing conversational artificial intelligence (AI) operations; A system implementing one or more language models; A system implementing one or more large language models (LLMs); Systems for generating synthetic data; Systems for generating synthetic data using AI; A system comprising one or more virtual machines VM; A system implemented at least in part in a data center; or A system implemented at least in part using cloud computing resources.

20. A method comprising: Generate a prediction of a 3D pose of at least a portion of an occupant of a vehicle based on a machine learning model trained using: one or more 3D ground truth measurements determined using a 3D point cloud representing a model customized for at least a portion of a training subject, and optical image data of at least a portion of the training subject captured synchronously with the 3D point cloud, wherein the one or more 3D ground truth head pose measurements are transformed to a reference frame of the optical image data.

Images