Neural network-based facial analysis using facial landmarks and associated confidence values

Abstract

A system and method for more accurately and robustly determining a characteristic of a subject from an image of the subject.SOLUTION: Determining, using the parallel processing circuitry, landmark points of the subject and corresponding confidence values, wherein the confidence values correspond to an amount of occlusion of portions of the subject image data corresponding to the respective landmark points, and using the landmark points and the confidence values to monitor a state of the subject, wherein the one or more machine learning models are a first set of one or more machine learning models; The method where the using further includes transmitting the landmark points and the confidence values for use with a second set of one or more machine learning models configured to generate as output at least one of the gaze direction of the subject, the head pose of the subject, or the drowsiness state of the subject.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Provisional Application No. 62 / 948,796, filed December 16, 2019, the disclosure of which is incorporated herein by reference in its entirety. [Background technology]

[0002] Recently, convolutional neural networks (CNNs) have been developed to estimate a subject's gaze direction. Such CNNs can determine, for example, from an input image of the subject, the direction in which the subject is looking. This enables systems using such CNNs to track the subject's gaze and react accordingly in real time.

[0003] However, conventional gaze determination systems are not without drawbacks. The overall performance and robustness of such systems remain limited, especially in edge cases or extreme situations such as large variations in head pose or partial occlusion of the face or eyes. Figure 1 shows examples of such extreme situations. More specifically, Figure 1 illustrates a situation in which a subject's face is partially or largely occluded, in this case by a breathing mask (left image), and a situation in which the subject has turned their head significantly in one direction (right image). In both examples, most of the subject's face (in the left image, the entire subject's face below the eyes; in the right image, the entire right side of the subject's face) is invisible, meaning that most of the subject's facial information is insensible. This limits the information available to a CNN for determining facial pose, often leading to inaccurate results. Summary of the Invention

[0004] Thus, systems and methods for implementing machine learning based on facial analysis in a more accurate and efficient manner are described herein. In one embodiment of the present disclosure, one or more machine learning models are used to determine both facial landmarks and associated confidence values, where the confidence values ​​represent the degree to which portions of a subject's face corresponding to the subject's facial landmarks are occluded, i.e., the uncertainty of the location of each landmark point. These landmark points and their associated confidence values ​​can be used to monitor the subject's state. For example, the landmark points and associated confidence values ​​can be input into another set of one or more machine learning models to output quantities characterizing the subject, such as the subject's gaze direction, head posture, drowsiness, cognitive load, or distraction state.

[0005] Embodiments of the present disclosure include a feedback mechanism that uses the determined confidence values ​​to improve the input image by selecting portions of the image that correspond to higher confidence values. These image portions may be deemed more reliable for generating accurate landmark values. The more reliable image portions may then be input into one or more machine learning models to generate more accurate landmarks and confidence values. In some embodiments, only the image portions that correspond to higher confidence values ​​may be input into one or more machine learning models, thereby ignoring portions with lower confidence values ​​and improving the resulting output. This process may be repeated as necessary to generate further improved landmarks and confidence values, which may be used to generate improved estimates of various characteristics of the subject.

[0006]

[0006] Embodiments of the present disclosure also include using only landmark values ​​having confidence values ​​corresponding to portions of the subject that are not occluded. That is, embodiments may use only landmarks derived from unoccluded portions of the subject and therefore having higher confidence values. In this manner, more accurate and efficient determination of subject characteristics, such as gaze direction, head posture, and drowsiness, may be made. To achieve this, only landmark values ​​derived from unoccluded portions of the subject may be input into a machine learning model that determines subject characteristics (e.g., the subject's gaze direction, head posture, drowsiness, cognitive load, distraction, etc.).

[0007] A weight value may be assigned to each landmark point according to its corresponding confidence value, with more confident landmark points being assigned a larger weight. That is, the assigned weight value may be inversely proportional to the level of occlusion of the corresponding part of the subject. To further improve the accuracy of the system according to embodiments of the present disclosure, a machine learning model that determines the subject's features may be trained to consider these weight values ​​along with their landmarks.

[0008] The characteristics or state of the subject determined by the above system may be used in any manner. In particular, the subject's gaze direction may be used in various applications, such as autonomous vehicles and augmented reality headsets. As an example, the subject's state may be used to initiate an action, such as a vehicle operation. For example, if it is determined that the driver's gaze is not directed toward the road, a warning message may be initiated to direct the driver's attention to the road. Embodiments of the present disclosure contemplate initiating any type of action responsive to any determined characteristic of the subject's state.

[0009] These and other objects and advantages of the present disclosure will become apparent from the following detailed description considered in conjunction with the accompanying drawings in which like reference characters refer to like parts throughout. [Brief explanation of the drawings]

[0010] [Figure 1]10. A conceptual illustration of subject's face occlusion. [Figure 2] FIG. 1 is a block diagram illustrating the operation of a system for determining and using subject landmarks and confidence values ​​according to an embodiment of the present disclosure. [Figure 3] 1 is a block diagram representation of an exemplary line-of-sight determination system according to an embodiment of the present disclosure. [Figure 4A] FIG. 1 is a diagram of an exemplary autonomous vehicle, according to some embodiments of the present disclosure. [Figure 4B] 4B is an example of camera positions and fields of view for the exemplary autonomous vehicle of FIG. 4A, according to some embodiments of the present disclosure. [Figure 4C] FIG. 4B is a block diagram of an example system architecture of the example autonomous vehicle of FIG. 4A, in accordance with some embodiments of the present disclosure. [Figure 4D] FIG. 4B is a system diagram for communication between a cloud-based server and the example autonomous vehicle of FIG. 4A, according to some embodiments of the present disclosure. [Figure 5] FIG. 1 is a block diagram of an exemplary computing device suitable for use in implementing some embodiments of the present disclosure. [Figure 6] 1 is a flowchart illustrating process steps for determining and using subject landmarks and confidence values ​​according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] In one embodiment, the present disclosure relates to a system and method for more accurate and robust determination of subject characteristics from images of the subject. One or more machine learning models receive as input images of the subject and output both facial landmarks and associated confidence values. The confidence values ​​represent the degree to which portions of the subject's face corresponding to the subject's facial landmarks are occluded, i.e., the certainty of the location of each landmark point. These landmark points and associated confidence values, and / or related information, are then input into another set of one or more machine learning models, which output facial analysis quantities such as the subject's gaze direction, head pose, drowsiness, cognitive load, or distraction state.

[0012] 2 is a block diagram illustrating an exemplary machine learning model-based architecture for determining landmark points and associated confidence values ​​according to an embodiment of the present disclosure. The model includes a face detection module 210, a facial landmark network 220 that outputs landmark values ​​and associated confidence values, and any number of machine learning models or networks that determine output quantities from the landmark and confidence value inputs.

[0013] The system of FIG. 2 also includes a sensor, such as a camera 200, that captures an image of a subject, such as a person, whose gaze direction is to be determined. The camera 200 sends image data from the captured image to a face detection module 210, which detects the subject's face in the image. The face detection module 210 can be any software module or set of instructions that can locate the subject's face in the image using any method or process, including known computer vision-based face detection processes that detect faces without using neural networks, such as edge detection methods, feature search methods, probabilistic face models, graph matching, support vector machines, and histograms of oriented gradients (HOG) fed into classifiers such as the HaarCascade classifier. Determining the face location can also be performed using neural network-based face recognition methods, such as methods that use deep neural network (DNN) face recognition schemes, as well as other schemes.

[0014] The detected faces may then be cropped for input to the facial landmark network 220. In this manner, a portion of the image or image data from the camera is selected for input to the facial landmark network 220. The section of the image may be selected, for example, by cropping the identified face according to a bounding box that may be drawn in a known manner.

[0015] The cropped facial portion is sent to a facial landmark network 220, which determines facial landmarks and corresponding confidence values. The facial landmark network 220 can be any machine learning network, e.g., any one or more machine learning models, capable of determining facial landmarks from facial input images. Such networks can include machine learning models constructed according to holistic methods for representing global facial appearance and shape information, as well as models constructed according to constrained local model methods that build local appearance models in addition to utilizing global shape models, generative networks, CNNs, and regression-based models for determining landmark locations as a function of facial shape and appearance information. Many such models are known, and embodiments of the present disclosure contemplate using any one or more such models, or any other suitable model or method, to determine facial landmarks from facial input images. The model can be constructed using any architecture and method suitable for determining facial landmarks from facial input images. For example, a CNN-based facial landmark network can be constructed using any convolutional kernel and pooling layer suitable for extracting facial features to determine corresponding landmark points.

[0016] The model can be further modified to generate as an output a confidence value for each output facial landmark. Such a confidence value corresponds to the detection probability of each landmark, i.e., the likelihood that the landmark point is actually located on the facial feature it is supposed to represent. Thus, the probability or confidence value for each landmark may indicate the level of occlusion of a particular facial region. This is because the more occluded (i.e., less visible) a facial feature is, the less likely the landmark will be correctly located on that feature. As an example, if the facial landmark network uses a classifier to determine landmark locations, the confidence value may be or correspond to the confidence score generated by the classifier in determining the landmark classification.

[0017] Thus, for an input image of a subject's face, the output of the facial landmark network 220 of FIG. 2 can be a set of landmark points, i.e., the two-dimensional locations of the subject's facial features, and a confidence value for each landmark point. These outputs can be sent to any number and type of networks, models, or other software modules to serve as inputs. Such networks can include, by way of example, a gaze network 230 that determines the subject's gaze direction, a head pose network 240 that determines the subject's head pose in the image, and a drowsiness model 250 that determines the likelihood that the subject is asleep or lethargic. Embodiments of the present disclosure also contemplate any other network that can have facial landmarks as inputs. The confidence values ​​can also be input to the face detection module 210 as a form of feedback to improve facial feature selection and, therefore, landmark point generation. In particular, the face detection module 210 can select only facial features corresponding to sufficiently high confidence values, i.e., select only sufficiently unoccluded portions of the subject's face for bounding box drawing and face cropping. This allows for more accurate or reliable determination of landmark points.

[0018] The exemplary gaze network 230 shown in FIG. 2 may be any network that determines a subject's gaze direction from at least landmark points on the subject's face. As one example, the gaze network 230 may be one or more machine learning models constructed to receive both landmark values ​​and corresponding confidence values ​​as inputs and output a gaze direction. Such networks may include known gaze networks configured to receive landmark values ​​as inputs, modified to incorporate confidence values ​​as additional inputs. The confidence values ​​may then be used, for example, to modify weights of features determined from the input landmarks or weights of the input landmarks themselves (the weight values ​​of input landmarks are typically inversely proportional to the level of occlusion associated with the landmarks, i.e., proportional to their corresponding confidence values), thereby decreasing the contribution of features corresponding to landmark points with low confidence values ​​and / or increasing the contribution of features corresponding to landmark points with higher confidence values. As another example, the gaze network 230 may be one or more machine learning models constructed to receive landmark values ​​as inputs excluding confidence values. In this latter example, the confidence values ​​may be used to filter out landmark points with insufficient confidence before input to the gaze network. That is, the confidence value can be used as a filter prior to gaze network input, so that only landmark points with sufficient confidence values ​​are input to gaze network 230. In this way, gaze network 230 can consider only unoccluded areas of the subject's face when determining gaze, and therefore produce more accurate results. This can produce more accurate results compared to conventional gaze detection models that do not have the ability to distinguish between occluded and unoccluded areas of the face when determining gaze.

[0019] The exemplary head pose network 240 shown in FIG. 2 can be any network that determines a subject's head pose from at least landmark points on the subject's face. As one example, the head pose network 240 can be one or more machine learning models constructed to receive both landmark values ​​and corresponding confidence values ​​as inputs and output head pose values. Such networks can include known head pose networks configured to receive landmark values ​​as inputs, modified to incorporate confidence values ​​as additional inputs. The confidence values ​​can then be used, for example, to modify weights of features determined from the input landmarks or weights of the input landmarks themselves (the weight values ​​of input landmarks are typically inversely proportional to the level of occlusion associated with the landmarks, i.e., proportional to their corresponding confidence values), thereby decreasing the contribution of features corresponding to landmark points with low confidence values ​​and / or increasing the contribution of features corresponding to landmark points with higher confidence values. As another example, the head pose network 240 can be one or more machine learning models constructed to receive landmark values ​​as inputs excluding confidence values. In this latter example, the confidence values ​​can be used to filter out landmark points with insufficient confidence before input to the head pose network. That is, the confidence values ​​can be used as a filter before input to head pose network 240, so that only landmark points with sufficient confidence values ​​are input to head pose network 240. In this way, head pose network 240 can consider only unoccluded regions of the subject's head when determining head pose, and therefore produce more accurate results. This can produce more accurate results compared to conventional head pose determination models that do not have the ability to distinguish between occluded and unoccluded regions of the subject's head when determining pose.

[0020] The exemplary somnolence model 250 shown in FIG. 2 can be any one or more machine learning models that determine a drowsiness state from at least landmark points on a subject's face. As one example, somnolence model 250 can be one or more machine learning models constructed to receive both landmark values ​​and corresponding confidence values ​​as inputs and output a drowsiness state, such as a binary drowsy / non-drowsy state, one of a finite number of drowsiness states, etc. Such networks can include known somnolence detection networks configured to receive landmark values ​​as inputs, with these networks modified to incorporate confidence values ​​as additional inputs. The confidence values ​​can then be used, for example, to modify weights of features determined from the input landmarks or weights of the input landmarks themselves (weight values ​​of input landmarks are typically inversely proportional to the level of occlusion associated with the landmarks, i.e., proportional to their corresponding confidence values), thereby decreasing the contribution of features corresponding to landmark points with low confidence values ​​and / or increasing the contribution of features corresponding to landmark points with higher confidence values. As another example, somnolence detection network 250 can be one or more machine learning models constructed to receive landmark values ​​as inputs excluding confidence values. In this latter example, the confidence value can be used to filter out landmark points with insufficient confidence before input to somnolence detection network 250. That is, the confidence value can be used as a filter before somnolence model input, so that only landmark points with sufficient confidence values ​​are input to somnolence detection network 250. In this manner, somnolence detection network 250 considers only unoccluded regions of the subject's head when determining somnolence, thereby producing more accurate and efficient results given the available data. This can produce more accurate results compared to conventional somnolence detection models that do not have the ability to distinguish between occluded and unoccluded regions of the subject's head when determining somnolence.

[0021] The outputs of gaze network 230, head pose network 240, and drowsiness model 250 may be transmitted to any system or module for any use. For example, the outputs of these networks may be transmitted to visualizer 260, which may be any display suitable for displaying electronic information. Visualizer 260 may display the outputs of gaze network 230, head pose network 240, and drowsiness model 250 in graphical or visual form and / or may display alerts, warning messages, or any other indicators based on these outputs so that a user can see the indicators and react accordingly.

[0022] It should be noted that the system of FIG. 2 may include other networks in addition to the gaze network 230, the head pose network 240, and the drowsiness model 250. In particular, the landmark points and confidence values ​​output by the facial landmark network 220 may be input to any number of other networks. For example, in addition to the gaze network 230, the head pose network 240, and the drowsiness model 250, the system of FIG. 2 may further include any other networks, such as a cognitive load network. An exemplary cognitive load network may be one or more machine learning models that determine a cognitive load state, or at least an estimate of the amount of working memory currently being used, from the landmark points on the subject's face. As an example, the cognitive load network may include one or more machine learning models configured to receive both the landmark values ​​and corresponding confidence values ​​as inputs and output a cognitive load state, such as a binary excessive / non-excessive cognitive load state, one of a finite number of cognitive load states, or the like. Such networks may include known cognitive load detection networks configured to receive landmark values ​​as inputs, with these networks modified to incorporate confidence values ​​as additional inputs. The confidence values ​​can then be used to, for example, modify weights of features determined from the input landmarks or weights of the input landmarks themselves (the weight values ​​of input landmarks are typically inversely proportional to the level of occlusion associated with the landmarks, i.e., proportional to their corresponding confidence values), reducing the contribution of features corresponding to landmark points with low confidence values ​​and / or increasing the contribution of features corresponding to landmark points with higher confidence values. As another example, the cognitive load detection network can be one or more machine learning models constructed to receive landmark values ​​as input, excluding the confidence values. In this latter example, the confidence values ​​can be used to filter out landmark points with insufficient confidence before input to the cognitive load detection network. That is, the confidence values ​​can be used as a filter before the cognitive load model input, such that only landmark points with sufficient confidence values ​​are input to the cognitive load detection network.In this way, the cognitive load detection network considers only the unoccluded areas of the subject's head when determining the cognitive load state, and thus may produce more accurate results compared to conventional cognitive load detection models that do not have the ability to distinguish between occluded and unoccluded areas of the subject's head when determining the cognitive load state.

[0023] The system of FIG. 2 may also include a distraction state network. An exemplary distraction state network may be any one or more machine learning models that determine a distraction state or an estimate of a distraction level of a subject, such as a vehicle driver. As an example, the distraction state network may include one or more machine learning models constructed to receive both landmark values ​​and corresponding confidence values ​​as inputs and output a distraction state—e.g., a binary distracted / non-distracted state, one of a finite number of distraction states, etc.—based on features such as the subject's gaze direction. Such networks may include known distraction state networks configured to receive landmark values ​​as inputs, these networks modified to incorporate confidence values ​​as additional inputs. The confidence values ​​may then be used, for example, to modify weights of features determined from the input landmarks or weights of the input landmarks themselves (weight values ​​of input landmarks are typically inversely proportional to the level of occlusion associated with the landmarks, i.e., proportional to their corresponding confidence values), thereby decreasing the contribution of features corresponding to landmark points with low confidence values ​​and / or increasing the contribution of features corresponding to landmark points with higher confidence values. As another example, the distraction state network can be one or more machine learning models constructed to receive landmark values ​​as input, excluding a confidence value. In this latter example, the confidence value can be used to filter out landmark points with insufficient confidence before input to the distraction state network. That is, the confidence value can be used as a filter before the distraction state network input, so that only landmark points with sufficient confidence values ​​are input to the distraction state network. In this way, the distraction state network can consider only the unoccluded areas of the subject's head when determining the distraction state, thus generating more accurate results. This can generate more accurate results than conventional distraction state detection models that do not have the ability to distinguish between occluded and unoccluded areas of the subject's head when determining the distraction state.

[0024] FIG. 3 is a block diagram representation of one exemplary gaze determination system according to an embodiment of the present disclosure. Here, a computing device 300, which may be any electronic computing device including processing circuitry capable of performing the landmark and confidence value determination operations of an embodiment of the present disclosure, is in electronic communication with both a camera 310 and a gaze assistance system 320. During operation, the camera 310, which may correspond to the camera 200 of FIG. 2 and the cabin camera 441 of FIGS. 4A and 4C below, captures and transmits an image of a subject to the computing device 300, which then implements, for example, the machine learning model of FIG. 2 to determine output landmarks and confidence values ​​of the subject's face from the image of the camera 310. The computing device 300 transmits the landmarks and confidence values ​​to the gaze assistance system 320, which in response performs an action or performs one or more operations. The computing device 300 may be any one or more electronic computing devices suitable for implementing the machine learning models of an embodiment of the present disclosure, such as computing device 500, described in further detail below.

[0025] The gaze assistance system 320 may be any system capable of performing one or more actions based on landmarks and confidence values ​​received from the computing device 300, such as determining a subject's gaze direction and performing various actions accordingly. The gaze assistance system 320 may include, for example, the gaze network 230, head pose network 240, drowsiness model 250, and visualizer 260 of FIG. 2 . Any configuration of the camera 310, computing device 300, and gaze assistance system 320 is contemplated. As an example, the gaze assistance system 320 may be an autonomous vehicle capable of determining and reacting to the gaze direction of a driver or another passenger, such as the autonomous vehicle 400 described in further detail below. In this example, the camera 310 and computing device 300 may be located within the vehicle, while the gaze assistance system 320 may represent the vehicle itself. The camera 310 may be located anywhere within the vehicle from which it can view the driver or passenger. Thus, the camera 310 can capture images of the driver or passenger and transmit them to the computing device 300, which calculates the corresponding subject's landmarks and confidence values. The vehicle can include a gaze network, such as that shown in FIG. 2, to determine the resulting gaze direction of the subject. This gaze direction can then be transmitted to another software module, which determines, for example, what action the vehicle can take in response. For example, the vehicle can determine that the gaze direction represents a distracted driver or a driver not paying attention to the road and initiate any type of action accordingly. Such action may include any type of warning issued to the driver (e.g., a visual or audible warning, a warning on a head-up display, etc.), initiating autopilot, braking or turning, or other action. The computing device 300 can be located within the vehicle of the gaze assistance system 320 as a local processor, or it can be a remote processor that receives images from the camera 310 and wirelessly transmits gaze directions or commands to the vehicle of the gaze assistance system 320.

[0026] As another example, the gaze assist system 320 may be a virtual reality or augmented reality system capable of displaying images in response to a user's movements and gaze. In this example, the gaze assist system 320 includes a virtual reality or augmented reality display, such as a headset worn by a user and configured to project images thereon. The camera 310 and computing device 300 may be disposed within the headset, with the camera 310 capturing images of the user's eyes and the computing device 300 determining landmarks and confidence values, as well as the user's gaze direction, from which the camera 310 determines landmarks and confidence values, as well as the user's gaze direction. This gaze direction is then transmitted to the virtual reality or augmented reality display, which may perform any action in response. For example, to conserve computational resources, the gaze assist system 320 may render only virtual reality or augmented reality elements that are within the user's field of view, as determined using the determined gaze direction. Similarly, the gaze assist system 320 may alert the user to objects or events that are determined to be outside the user's field of view but that the user may wish to avoid or may be interested in. Similar to the autonomous vehicle example above, the computing device 300 of the virtual reality or augmented reality system can be located within the system 320, e.g., within the headset itself, or it can be located remotely such that images are wirelessly transmitted to the computing device 300 and a calculated gaze direction is wirelessly transmitted to the headset, which then performs various actions accordingly. In this manner, the systems and methods of embodiments of the present disclosure can result in a virtual reality or augmented reality system that more accurately and efficiently determines a user's gaze direction. In particular, the use of more reliable landmark points improves the accuracy of gaze determination and reduces wasted computational resources associated with rendering images with incorrectly determined gaze directions.

[0027] As yet another example, the gaze assistance system 320 may be a computer-based advertising system that determines which advertisements a user is viewing. More specifically, the gaze assistance system 320 may be any electronic computing system or device, such as a desktop computer, a laptop computer, a smartphone, a server computer, or the like. The camera 310 and the computing device 300 may be incorporated into the computing device so as to face the user, such as in or near the display of the computing device. The camera 310 may capture an image of the user, and the computing device 300 may determine landmarks and confidence values, as well as the user's gaze direction, therefrom. The determined gaze direction may then be transmitted to the gaze assistance system 320, e.g., the computing device 300 that displays advertisements to the user, a remote computing device, or the like. The computing device may then use the calculated gaze direction to determine which advertisements the user is focusing on and provide information about the effectiveness of various advertisements.

[0028] 4A is a diagram of an example autonomous vehicle 400 according to some embodiments of the present disclosure. The autonomous vehicle 400 (sometimes referred to herein as “vehicle 400”) 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 bicycle or moped, a motorcycle, a fire engine, a police vehicle, an ambulance, a boat, a construction vehicle, an underwater vehicle, a drone, and / or another type of vehicle (e.g., a vehicle that is unmanned and / or accommodates one or more occupants). 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) in their “Level Classification and Definition of Driving Automation Systems for Motor Vehicles” (Standard No. J3016-201806, issued June 15, 2018, Standard No. J3016-201609, issued September 30, 2016, and previous and future versions of this standard). Vehicle 400 may be capable of functioning according to one or more of levels of automation, from Level 3 to Level 5. For example, vehicle 400 may be capable of conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5), depending on the embodiment.

[0029] Vehicle 400 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 400 may include a propulsion system 450, such as an internal combustion engine, a hybrid power plant, an all-electric engine, and / or another propulsion system type. Propulsion system 450 may be connected to a drivetrain of vehicle 400, which may include a transmission, to enable propulsion of vehicle 400. Propulsion system 450 may be controlled in response to receiving a signal from a throttle / accelerator 452.

[0030] A steering system 454, which may include a steering wheel, may be used to steer the vehicle 400 (e.g., along a desired path or road) when the propulsion system 450 is operating (e.g., when the vehicle is moving). The steering system 454 may receive signals from a steering actuator 456. The steering wheel may be optional for fully automated (Level 5) functionality.

[0031] Brake sensor system 446 may be used to operate vehicle brakes in response to receiving signals from brake actuator 448 and / or brake sensors.

[0032] A controller 436, which may include one or more CPUs, system-on-chip (SoC) 404 (FIG. 4C), and / or GPUs, may provide signals (e.g., representing commands) to one or more components and / or systems of the vehicle 400. For example, the controller may send signals to operate the vehicle's brakes via one or more brake actuators 448, the steering system 454 via one or more steering actuators 456, and / or the propulsion system 450 via one or more throttles / accelerators 452. The controller 436 may include one or more on-board (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 400. The controllers 436 may include a first controller 436 for autonomous driving functions, a second controller 436 for functional safety functions, a third controller 436 for artificial intelligence functions (e.g., computer vision), a fourth controller 436 for infotainment functions, a fifth controller 436 for redundancy in emergency situations, and / or other controllers. In some examples, a single controller 436 may govern two or more of the above functions, and two or more controllers 436 may govern a single function and / or any combination thereof.

[0033] Controller 436 may provide signals to control one or more components and / or systems of vehicle 400 in response to sensor data (e.g., sensor inputs) received from one or more sensors. The sensor data may be received from, by way of example and not limitation, global navigation satellite system sensors 458 (e.g., global positioning system sensors), radar sensors 460, ultrasonic sensors 462, lidar sensors 464, inertial measurement unit (IMU) sensors 466 (e.g., accelerometers, gyroscopes, magnetic compasses, magnetometers, etc.), microphones 496, stereo cameras 468, wide-view cameras 470 (e.g., fisheye cameras), infrared cameras 472, surround cameras 474 (e.g., 360-degree cameras), long-range and / or medium-range cameras 498, speed sensors 444 (e.g., measuring the speed of vehicle 400), vibration sensors 442, steering sensors 440, brake sensors 446 (e.g., as part of brake sensor system 446), and / or other sensor types.

[0034] One or more controllers 436 may receive input (e.g., represented by input data) from the instrument cluster 432 of the vehicle 400 and provide output (e.g., represented by output data, display data, etc.) via a human-machine interface (HMI) display 434, an audible annunciator, a speaker, and / or other components of the vehicle 400. The output may include information such as vehicle speed, velocity, time, map data (e.g., HD map 422 of FIG. 4C ), position data (e.g., the location of the vehicle 400 on a map, etc.), direction, the location of other vehicles (e.g., an occupancy grid), information about objects, and the state of objects as perceived by the controller 436. For example, the HMI display 434 may display information about the presence of one or more objects (e.g., road signs, caution signs, traffic light switches, etc.) and / or information about a driving maneuver the vehicle has performed, is performing, or is about to perform (e.g., change lanes here, take exit 34B in 2 miles, etc.).

[0035] Vehicle 400 further includes a network interface 424 capable of communicating over one or more networks using one or more wireless antennas 426 and / or a modem. For example, network interface 424 may be capable of communicating via LTE, WCDMA, UMTS, GSM, CDMA2000, etc. Wireless antenna 426 may also enable communication between objects in the environment (e.g., vehicles, mobile devices, etc.) using local area networks such as Bluetooth, Bluetooth LE, Z-Wave, ZigBee, etc., and / or low power wide area networks (LPWANs) such as LoRaWAN, SigFox, etc.

[0036] 4B is an example of camera positions and fields of view for the exemplary autonomous vehicle 400 of FIG. 4A, according to some embodiments of the present disclosure. The cameras and their respective fields of view are an exemplary embodiment and are not intended to be limiting. For example, additional and / or alternative cameras may be included and / or the cameras may be located at different locations on the vehicle 400.

[0037] The camera type may include, but is not limited to, a digital camera that may be adapted for use with components and / or systems of vehicle 400. The camera may be capable of operating at Automotive Safety Integrity Level (ASIL) B and / or another ASIL. The camera type may be capable of any image capture rate, such as 60 frames per second (fps), 120 fps, 240 fps, etc., 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-clear-clear-clear (RCCC) color filter array, a red-clear-clear-blue (RCCB) color filter array, a red-blue-green-clear (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 to improve light sensitivity.

[0038] In some examples, one or more cameras may 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 mono camera may be installed to provide features such as lane departure warning, traffic sign assist, intelligent headlamp control, etc. One or more cameras (e.g., all cameras) may simultaneously record and provide image data (e.g., video).

[0039] To block stray light and reflections from the interior of the vehicle (e.g., reflections from the dashboard reflected in the windshield mirror) that may interfere with the camera's image data capture function, one or more cameras may be mounted on a mounting assembly, such as a custom-designed (3D printed) assembly. With respect to wing mirror mounting assemblies, the wing mirror assembly may be custom 3D printed so that the camera mounting plate matches the shape of the wing mirror. In some examples, the camera may be integrated into the wing mirror. In the case of side-view cameras, the cameras may be integrated into four pillars at each corner of the cabin.

[0040] A camera (e.g., a forward-facing camera) with a field of view that includes a portion of the environment ahead of the vehicle 400 can be used to obtain a surround view to help identify the path and obstacles ahead and, with the aid of one or more controllers 436 and / or control SoCs, to help provide important information for generating an occupancy grid and / or determining a preferred vehicle path. Forward-facing cameras can be used to perform many of the same ADAS functions as lidar, such as emergency braking, pedestrian detection, and collision avoidance. Forward-facing cameras can also be used for ADAS functions and systems, including other functions such as lane departure warning (LDW), autonomous cruise control (ACC), and / or traffic sign recognition.

[0041] Various cameras can be used in a forward-facing configuration, such as a monocular camera platform including a CMOS (complementary metal-oxide semiconductor) color imager. Another example can be a wide-view camera 470 that can be used to perceive objects (e.g., pedestrians, crossing vehicles, or bicyclists) that enter the field of view from the surroundings. While only one wide-view camera is shown in FIG. 4B, there can be any number of wide-view cameras 470 on the vehicle 400. Additionally, a long-range camera 498 (e.g., a long-field-of-view stereo camera pair) can be used for depth-based object detection, particularly for objects for which a neural network has not yet been trained. The long-range camera 498 can also be used for object detection and classification, as well as basic object tracking.

[0042] One or more stereo cameras 468 may also be included in the forward-facing configuration. The stereo camera 468 may include an integrated control unit with a scalable processing unit that may provide programmable logic (e.g., FPGA) and a multi-core microprocessor with an integrated 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. An alternative stereo camera 468 may include a miniature stereo vision sensor that includes two camera lenses (one on each side) and an image processing chip that can measure the distance from the vehicle to target objects and use the generated information (e.g., metadata) to activate autonomous emergency braking and lane departure warning features. Other types of stereo cameras 468 may be used in addition to or instead of those described herein.

[0043] Cameras having a field of view that includes a portion of the environment to the sides of the vehicle 400 (e.g., side-view cameras) can be used for surround view to provide information used to create and update the occupancy grid and generate side impact collision warnings. For example, surround cameras 474 (e.g., four surround cameras 474 as shown in FIG. 4B ) can be positioned around the vehicle 400. The surround cameras 474 can include wide-view cameras 470, fisheye cameras, 360-degree cameras, etc. For example, four fisheye cameras can be positioned at the front, rear, and sides of the vehicle. In an alternative configuration, the vehicle may use three surround cameras 474 (e.g., left, right, and rear) and utilize one or more other cameras (e.g., a forward-facing camera) as a fourth surround view camera.

[0044] A camera having a field of view that includes a portion of the environment behind vehicle 400 (e.g., a rearview camera) may be used for parking assistance, surround view, rear collision warning, and creating and updating an occupancy grid. A wide variety of cameras can be used, including, but not limited to, cameras that are also suitable as forward-facing cameras (e.g., long-range and / or mid-range camera 498, stereo camera 468, infrared camera 472, etc.), as described herein.

[0045] A camera having a field of view that includes a portion of the interior or cabin of vehicle 400 may be used to monitor one or more conditions of the driver, passengers, or objects within the cabin. Any type of camera may be used, including but not limited to cabin camera 441, which may be any type of camera described herein and may be located anywhere in vehicle 400 to provide a field of view of the cabin or interior of vehicle 400. For example, cabin camera 441 may be located within or part of the dashboard, rearview mirror, side mirror, seat, or door of vehicle 400 and oriented to capture an image of any driver, passenger, or other object or portion of vehicle 400.

[0046] FIG. 4C is a block diagram of an example system architecture for the example autonomous vehicle 400 of FIG. 4A , in accordance with some embodiments of the present disclosure. It should be understood that this and other configurations described herein are presented by way of example only. Other configurations and elements (e.g., machines, interfaces, functions, sequences, groupings of functions, etc.) may be used in addition to or in place of those shown, and some elements may be omitted entirely. Furthermore, many of the elements described herein are functional entities that may be implemented as individual or distributed components, or in combination with other components, in any suitable combination and location. Various functions described herein as performed by entities may be performed by hardware, firmware, and / or software. For example, various functions may be performed by a processor executing instructions stored in a memory.

[0047] Each of the components, features, and systems of the vehicle 400 in FIG. 4C is shown as connected via a bus 402. The bus 402 may include a controller area network (CAN) data interface (alternatively sometimes referred to herein as a “CAN bus”). The CAN may be a network internal to the vehicle 400 used to help control various features and functions of the vehicle 400, such as brake application, acceleration, braking, steering, windshield wipers, etc. The CAN bus may be configured to have tens or hundreds of nodes, each with its own unique identifier (e.g., a CAN ID). The CAN bus may be read to find steering wheel angle, ground speed, engine revolutions per minute (RPM), button position, and other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0048] Although the bus 402 is described herein as being a CAN bus, this is not intended to be limiting. For example, FlexRay and / or Ethernet may be used in addition to or instead of a CAN bus. Furthermore, although a single line is used to represent the bus 402, this is not intended to be limiting. There may be any number of buses 402, which may include, for example, 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 402 may be used to perform different functions and / or for redundancy. For example, a first bus 402 may be used for collision avoidance functions and a second bus 402 may be used for actuation control. In any example, each bus 402 may communicate with any of the components of the vehicle 400, or two or more buses 402 may communicate with the same component. In some examples, each SoC 404, each controller 436, and / or each computer in the vehicle may have access to the same input data (e.g., input from sensors in the vehicle 400) and may be connected to a common bus, such as a CAN bus.

[0049] Vehicle 400 may include one or more controllers 436, such as those described herein with respect to FIG. 4A. Controller 436 may be used for a variety of functions. Controller 436 may be coupled to any of a variety of other components and systems of vehicle 400 and may be used for control of vehicle 400, artificial intelligence of vehicle 400, infotainment of vehicle 400, etc.

[0050] Vehicle 400 may include a system on a chip (SoC) 404. SoC 404 may include a CPU 406, a GPU 408, a processor 410, a cache 412, an accelerator 414, a data store 416, and / or other components and features not shown. SoC 404 may be used to control vehicle 400 in a variety of platforms and systems. For example, SoC 404 may be combined in a system (e.g., that of vehicle 400) with an HD map 422 that can obtain map refreshes and / or updates from one or more servers (e.g., server 478 of FIG. 4D ) via a network interface 424.

[0051] CPU 406 may include a CPU cluster or CPU complex (also sometimes referred to herein as a "CCPLEX"). CPU 406 may include multiple cores and / or L2 caches. For example, in some embodiments, CPU 406 may include eight cores in a coherent multiprocessor configuration. In some embodiments, CPU 406 may include four dual-core clusters, each with its own dedicated L2 cache (e.g., 2 MB of L2 cache). CPU 406 (e.g., a CCPLEX) may be configured to support simultaneous cluster operation, allowing any combination of clusters of CPUs 406 to be active at a given time.

[0052] The CPU 406 can implement power management features including one or more of the following features: individual hardware blocks can be automatically clock gated when idle to conserve dynamic power; each core clock can be gated when the core is not actively executing instructions by executing a WFI / WFE instruction; each core can be independently power gated; if all cores are clock gated or power gated, each core cluster can be independently clock gated; and / or if all cores are power gated, each core cluster can be independently power gated. The CPU 406 can further implement an enhanced algorithm for managing power states, where allowable power states and desired wake-up times are specified and hardware / microcode determines the best power state to enter for the cores, clusters, and CCPLEX. Processing cores can offload work to microcode to support simplified power state entry sequences in software.

[0053] GPU 408 may include an integrated GPU (also sometimes referred to herein as an "iGPU"). GPU 408 may be programmable and efficient for parallel workloads. In some examples, GPU 408 may use an extended tensor instruction set. GPU 408 may include one or more streaming microprocessors, where each streaming microprocessor may include an L1 cache (e.g., an L1 cache having at least 96 KB of storage capacity) and two or more streaming microprocessors may share an L2 cache (e.g., an L2 cache having 512 KB of storage capacity). In some embodiments, GPU 408 may include at least eight streaming microprocessors. GPU 408 may use a computer-based application programming interface (API). Additionally, GPU 408 may use one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0054] The GPU 408 may be power-optimized for best performance in automotive and embedded use cases. For example, the GPU 408 may be fabricated on Fin Field Effect Transistors (FinFETs). However, this is not intended to be limiting, and the GPU 408 may be fabricated using other semiconductor manufacturing processes. Each streaming microprocessor may incorporate several mixed-precision processing cores divided into multiple blocks. By way of example and not limitation, 64 PF32 cores and 32 PF64 cores may be divided into four processing blocks. In such an example, each processing block may be assigned 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two mixed-precision NVIDIA tensor cores for deep learning matrix operations, an L0 instruction cache, a warp scheduler, a dispatch unit, and / or a 64KB register file. Additionally, the streaming microprocessor may include independent parallel integer and floating-point data paths to efficiently execute workloads that combine computation and addressing calculations. Streaming microprocessors may include independent thread scheduling capabilities to allow finer synchronization and coordination between parallel threads. Streaming microprocessors may include a combination of an L1 data cache and a shared memory unit to improve performance while simplifying programming.

[0055] The GPU 408 may include a high-bandwidth memory (HBM) and / or 16 GB HBM2 memory subsystem to provide a peak memory bandwidth of approximately 900 GB / s in some examples. In some examples, synchronous graphics random access memory (SGRAM), such as graphics double data rate type 5 synchronous random access memory (GDDR5), may be used in addition to or instead of the HBM memory.

[0056] The GPU 408 may include unified memory technology, including access counters that enable more accurate migration of memory pages 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 408 to directly access the CPU 406 page tables. In such examples, if the GPU 408 memory management unit (MMU) experiences a failure, an address translation request may be sent to the CPU 406. In response, the CPU 406 may consult its page table for virtual-to-physical mapping of addresses and send the translation back to the GPU 408. The unified memory technology thus enables a single, unified virtual address space for both the CPU 406 and GPU 408 memories, thereby simplifying programming the GPU 408 and porting applications to the GPU 408.

[0057] Additionally, GPU 408 may include access counters that can track the frequency of GPU 408's accesses to the memory of other processors, which can help ensure that memory pages are moved to the physical memory of the processor that is accessing the pages most frequently.

[0058] The SoC 404 may include any number of caches 412, including those described herein. For example, the cache 412 may include an L3 cache available to both the CPU 406 and the GPU 408 (e.g., connected to both the CPU 406 and the GPU 408). The cache 412 may include a write-back cache that can track line states, such as by using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). In some embodiments, the L3 cache may include 4 MB or more, although smaller cache sizes may also be used.

[0059] The SoC 404 may include an arithmetic logic unit (ALU) that may be utilized in performing processing related to any of various tasks or operations of the vehicle 400, such as processing DNNs. Additionally, the SoC 404 may include a floating-point unit (FPU) or other type of mathematical or numeric co-processor that performs mathematical operations within the system. For example, the SoC 104 may include one or more FPUs integrated as execution units within the CPU 406 and / or GPU 408.

[0060] The SoC 404 may include one or more accelerators 414 (e.g., hardware accelerators, software accelerators, or a combination thereof). For example, the SoC 404 may include a hardware acceleration cluster, which may include optimized hardware accelerators and / or large on-chip memory. Using large on-chip memory (e.g., 4 MB of SRAM), the hardware acceleration cluster can accelerate neural networks and other calculations. The hardware acceleration cluster can be used to complement the GPU 408 and offload some of the GPU 408's tasks (e.g., freeing up more cycles of the GPU 408 to perform other tasks). As an example, the accelerator 414 can be used for target workloads (e.g., perception, convolutional neural networks (CNNs), etc.) that are stable enough to accommodate acceleration. As used herein, the term “CNN” may include all types of CNNs, including region-based or region-convolutional neural networks (RCNNs) and Fast RCNNs (e.g., used for object detection).

[0061] The accelerator 414 (e.g., a hardware acceleration cluster) may include a deep learning accelerator (DLA). The DLA may include one or more tensor processing units (TPUs), which can be configured to provide an additional 10 trillion operations per second for deep learning applications and inference. The TPU may be an accelerator configured and optimized to perform image processing functions (e.g., for CNN, RCNN, etc.). The DLA may be further optimized for a specific set of neural network types and floating-point operations and inference. The DLA's design provides higher performance per millisecond than general-purpose GPUs and significantly exceeds the performance of CPUs. The TPU can execute several functions, such as single-instance convolution functions, supporting INT8, INT16, and FP16 data types for both features and weights, as well as post-processor functions.

[0062] The DLA can quickly and efficiently run neural networks, particularly CNNs, on processed or unprocessed data for any of a variety of functions, including, by way of example and not limitation, CNNs for object recognition and detection using data from camera sensors, CNNs for distance estimation using data from camera sensors, CNNs for emergency vehicle identification and detection using data from microphones, CNNs for face recognition and vehicle owner identification using data from camera sensors, and / or CNNs for security and / or safety related events.

[0063] The DLA can perform any function of the GPU 408, and by using an inference accelerator, for example, a designer can direct either the DLA or the GPU 408 to any function. For example, a designer may focus CNN and floating-point processing in the DLA, and offload other functions to the GPU 408 and / or other accelerators 414.

[0064] The accelerator 414 (e.g., a hardware acceleration cluster) may include programmable vision accelerators (PVAs), which may also be referred to herein as computer vision accelerators. The PVAs 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 PVAs may provide a balance of performance and flexibility. For example, each PVA may include, by way of example and not limitation, any number of reduced instruction set computer (RISC) cores, direct memory access (DMA), and / or any number of vector processors.

[0065] The RISC cores may interact with an image sensor (e.g., an image sensor in any of the cameras described herein), an image signal processor, etc. Each RISC core may include any amount of memory. The RISC cores may use any of several protocols, depending on the embodiment. In some examples, the RISC cores may execute a real-time operating system (RTOS). The RISC cores may be implemented using one or more integrated circuit devices, application specific integrated circuits (ASICs), and / or memory devices. For example, the RISC cores may include an instruction cache and / or tightly coupled RAM.

[0066] The DMA may allow components of the PVA to access system memory independent of the CPU 406. The DMA may support any number of features used to provide optimizations to the PVA, including, but not limited to, support for multi-dimensional addressing and / or circular addressing. In some examples, the DMA may support up to six or more dimensions of addressing, which may include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0067] A vector processor may be a programmable processor that can be designed to efficiently and flexibly program computer vision algorithms and provide signal processing functions. In some examples, a PVA may include a PVA core and two vector processing subsystem partitions. The PVA core may include a processor subsystem, a DMA engine (e.g., two DMA engines), and / or other peripherals. The vector processing subsystem may act as the PVA's main processing engine and may include a vector processing unit (VPU), an instruction cache, and / or a vector memory (e.g., VMEM). The VPU core may include a digital signal processor, such as a single instruction, multiple data (SIMD), very long instruction word (VLIW) digital signal processor. The combination of SIMD and VLIW can improve throughput and speed.

[0068] Each of the vector processors may include an instruction cache and may be coupled to dedicated memory. As a result, in some examples, each of the vector processors may be configured to execute independently of the other vector processors. In other examples, the vector processors included in a particular PVA may be configured to use data parallelism. For example, in some embodiments, multiple vector processors included in a single PVA may execute the same computer vision algorithm on different regions of an image. In other examples, the vector processors included in a particular PVA may execute different computer vision algorithms simultaneously on the same image, or even execute different algorithms on consecutive images or portions of an image. Notably, a hardware acceleration cluster may include any number of PVAs, and each PVA may include any number of vector processors. Furthermore, the PVAs may include additional error-correcting code (ECC) memory to increase the overall security of the system.

[0069] The accelerator 414 (e.g., a hardware acceleration cluster) may include a computer vision network on-chip and SRAM that provides the accelerator 414 with high-bandwidth, low-latency SRAM. In some examples, the on-chip memory may include, by way of example and not limitation, at least 4 MB of SRAM consisting of eight field-configurable memory blocks accessible by both the PVA and 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 through a backbone that provides the PVA and DLA with high-speed access to the memory. The backbone may include an on-chip computer vision network that interconnects the PVA and DLA to the memory (e.g., using the APB).

[0070] The on-chip computer vision network may include an interface that verifies that both the PVA and DLA are providing ready and valid signals before transmitting control signals, addresses, and data. Such an interface may provide separate phases and separate channels for transmitting control signals, addresses, and data, as well as burst-type communication for continuous data transfer. This type of interface may conform to ISO 26262 or IEC 61508 standards, but may also use other standards and protocols.

[0071] In some examples, SoC 404 may include a real-time ray tracing hardware accelerator, such as that described in U.S. patent application Ser. No. 16 / 101,232, filed Aug. 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the location and range of objects (e.g., within a world model) and generate real-time visualization simulations that may be used for radar signal interpretation, synthesis and / or analysis of sound propagation, simulation of sonar systems, general wave propagation simulation, comparison with lidar data for localization and / or other functions, and / or other uses. In some embodiments, one or more tree traversal units (TTUs) may be used to perform one or more ray tracing-related operations.

[0072] The accelerator 414 (e.g., a hardware accelerator cluster) has a wide range of applications in autonomous driving. The PVA can be a programmable vision accelerator that can be used for key processing stages in ADAS and autonomous vehicles. The PVA's capabilities are ideally suited to algorithm domains that require predictable processing with low power and low latency. In other words, the PVA works well for semi-dense or dense regular computations, even on small datasets that require predictable runtime with low latency and low power. Therefore, in the context of an autonomous vehicle platform, the PVA is designed to run classical computer vision algorithms due to its efficiency in object detection and operating on integer arithmetic.

[0073] For example, according to one embodiment of the present technology, PVA is used to perform computer stereo vision. In some examples, a semi-global matching-based algorithm may 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 (structure derived from motion, pedestrian recognition, lane detection, etc.). PVA can perform computer stereo vision functions with input from two monocular cameras.

[0074] In some examples, PVAs may be used to perform dense optical flow. For example, PVAs may be used to process raw radar data (e.g., using a 4D fast Fourier transform) to provide a processed radar signal before emitting the next radar pulse. In other examples, PVAs may be used to perform time-of-flight depth processing, for example, by processing raw time-of-flight data to provide processed time-of-flight data.

[0075] DLA can be used to implement any type of network that enhances control and driving safety, such as a neural network that outputs a confidence value for each object detection. Such a confidence value can be interpreted as a probability or as providing the relative "weight" of each detection compared to other detections. This confidence value allows the system to make further decisions regarding which detections are considered true positives rather than false positives. For example, the system can set a confidence threshold and only consider detections above the threshold as true positives. In an automatic emergency braking (AEB) system, a false positive detection would cause the vehicle to automatically perform emergency braking, which is clearly undesirable. Therefore, only the most reliable detections should be considered to trigger AEB. DLA can implement a neural network that regresses a confidence value. The neural network can use as input at least a subset of parameters, such as bounding box dimensions, an estimate of the ground plane obtained (e.g., from another subsystem), the orientation of the vehicle 400, the output of an inertial measurement unit (IMU) sensor 466 that correlates with distance, and a 3D position estimate of the object obtained from the neural network and / or other sensors (e.g., a lidar sensor 464 or a radar sensor 460), among others.

[0076] The SoC 404 may include a data store 416 (e.g., memory). The data store 416 may be on-chip memory of the SoC 404 and may store neural networks running on the GPU and / or DLA. In some examples, the data store 416 may be large enough to store multiple instances of the neural network for redundancy and safety. The data store 416 may include an L2 or L3 cache 412. As described herein, references to the data store 416 may include references to memory associated with the GPU, DLA, and / or other accelerators 414.

[0077] The SoC 404 may include one or more processors 410 (e.g., embedded processors). The processors 410 may include a startup and power management processor, which may be a dedicated processor and subsystem for handling startup power and management functions and associated security enforcement. The startup and power management processor may be part of the SoC 404 boot sequence and may provide runtime power management services. The startup power and management processor may provide clock and voltage programming, assist in system low-power state transitions, manage the SoC 404's thermal and temperature sensors, and / or manage the SoC 404's power state. Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to temperature, and the SoC 404 may use the ring oscillator to detect the temperature of the CPU 406, GPU 408, and / or accelerator 414. If the temperature is determined to exceed a threshold, the startup and power management processor may enter a temperature fault routine and place the SoC 404 in a low-power state and / or place the vehicle 400 in a driver safety shutdown mode (e.g., safely shut down the vehicle 400).

[0078] The processor 410 may further include a set of embedded processors that can function as an audio processing engine. The audio processing engine can be an audio subsystem that enables full hardware support for multi-channel audio through multiple interfaces and a wide and flexible range of audio I / O interfaces. In some examples, the audio processing engine is a dedicated processor core with a digital signal processor with dedicated RAM.

[0079] The processor 410 may further include an always-on processor engine that can provide the necessary hardware functionality to support low-power sensor management and wake use cases. The always-on processor engine may include a processor core, tightly coupled RAM, supporting peripherals (e.g., timers and interrupt controllers), various I / O controller peripherals, and routing logic.

[0080] The processor 410 may further include a safety cluster engine that includes a dedicated processor subsystem for handling safety management for automotive applications. The safety cluster engine may include two or more processor cores, tightly coupled RAM, supporting peripherals (e.g., timers, interrupt controllers, etc.), and / or routing logic. In safety mode, the two or more cores operate in lockstep mode and function as a single core with comparison logic to detect differences between operations.

[0081] The processor 410 may further include a real-time camera engine, which may include a dedicated processor subsystem for handling real-time camera management.

[0082] The processor 410 may further 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.

[0083] The processor 410 may include a video image compositor, which may be a processing block (e.g., implemented in a microprocessor) that implements video post-processing functions required by the video playback application to generate the final image in the player window. The video image compositor may perform lens distortion correction on the wide-view camera 470, the surround camera 474, and / or the in-cabin surveillance camera sensor. The in-cabin surveillance camera sensor is preferably monitored by a neural network running on another instance of a high-performance SoC and configured to identify in-cabin events and respond accordingly. The in-cabin system may perform lip reading to activate and place calls to mobile phone services, dictate emails, change the vehicle's destination, activate or change the vehicle's infotainment system and settings, or provide voice-activated web surfing. Certain features are only available to the driver when the vehicle is operating in autonomous mode and are disabled otherwise.

[0084] The video image synthesizer may include enhanced temporal noise reduction that reduces both spatial and temporal noise. For example, when motion occurs in the video, the noise reduction appropriately weights spatial information and reduces the weight of information provided by adjacent frames. When an image or portion of an image does not contain motion, the temporal noise reduction performed by the video image synthesizer can use information from previous images to reduce noise in the current image.

[0085] The video image compositor may also be configured to perform stereo correction on the input stereo lens frames. The video image compositor may further be used for user interface composition when the operating system desktop is in use, and the GPU 408 does not need to continuously render new surfaces. Even when the GPU 408 is driven and actively performing 3D rendering, the video image compositor can be used to offload the GPU 408, improving performance and responsiveness.

[0086] The SoC 404 may further include a Mobile Industrial Processor Interface (MIPI) camera serial interface to receive video and input from a camera, a high-speed interface, and / or a video input block that may be used for camera and related pixel input functions. The SoC 404 may further include an input / output controller that may be controlled by software and that may be used to receive I / O signals not associated with a specific role. The SoC 404 may further include peripherals, audio codecs, power management, and / or a wide range of peripheral interfaces to enable communication with other devices. The SoC 404 may be used to process data from cameras (e.g., connected via gigabit multimedia serial links and Ethernet), sensors (e.g., lidar sensors 464, radar sensors 460, etc., which may be connected via Ethernet), data from bus 402 (e.g., vehicle 400 speed, steering wheel position, etc.), and GNSS sensors 458 (e.g., connected via Ethernet or a CAN bus). The SoC 404 may further include a dedicated high-performance mass storage controller, which may include its own DMA engine, that may be used to offload routine data management tasks from the CPU 406.

[0087] The SoC 404 can be an end-to-end platform with a flexible architecture spanning 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. The SoC 404 can be faster, more reliable, and potentially more energy- and space-efficient than traditional systems. For example, the accelerator 414, when combined with the CPU 406, GPU 408, and data store 416, can provide a fast and efficient platform for a level 3-5 autonomous vehicle.

[0088] This technology therefore offers capabilities not possible with conventional systems. For example, computer vision algorithms can be executed on a CPU, which can be configured using a high-level programming language, such as the C programming language, to perform a wide variety of processing algorithms across a wide variety of visual data. However, CPUs often cannot meet the performance requirements of many computer vision applications, such as those related to execution time and power consumption. In particular, many CPUs cannot execute complex object detection algorithms in real time, a requirement for in-vehicle ADAS applications and practical Level 3-5 autonomous vehicles.

[0089] In contrast to conventional systems, by providing a CPU complex, a GPU complex, and a hardware acceleration cluster, the technology described herein allows multiple neural networks to run simultaneously and / or sequentially and combine the results to enable Level 3-5 autonomous driving capabilities. For example, a CNN running on the DLA or dGPU (e.g., GPU 420) may include text and word recognition, enabling the supercomputer to read and understand traffic signs, including signs for which the neural network was not specifically trained. The DLA may further include a neural network that can identify, interpret, and understand the meaning of signs and pass that semantic understanding to a path planning module running on the CPU complex.

[0090] As another example, multiple neural networks required for Level 3, 4, or 5 driving can be run simultaneously. For example, a warning sign reading "Caution: Flashing lights indicate icy conditions," along with a light, can be interpreted individually or jointly by several neural networks. The sign itself can be identified as a traffic sign by a first deployed neural network (e.g., a trained neural network), and the text "Flashing lights indicate icy conditions" can be interpreted by a second deployed neural network, which notifies the vehicle's route planning software (preferably running on the CPU complex) that icy conditions exist if a flashing light is detected. The flashing light can be identified by running a third deployed neural network over multiple frames and notifying the vehicle's route planning software of the presence (or absence) of the flashing light. All three neural networks can run simultaneously, such as within the DLA and / or on the GPU 408.

[0091] In some examples, a CNN for facial recognition and vehicle owner identification can use data from a camera sensor to identify the presence of an authorized driver and / or owner of the vehicle 400. An always-on sensor processing engine can be used to unlock the vehicle when the owner approaches the driver's door and turns on the lights, and in security mode, disable the vehicle when the owner leaves the vehicle. In this way, the SoC 404 provides security against theft and / or carjacking.

[0092] In another example, a CNN for emergency vehicle detection and identification can use data from microphone 496 to detect and identify emergency vehicle sirens. Unlike conventional systems that use general classifiers to detect sirens and manually extract features, SoC 404 uses a CNN to classify ambient and urban sounds, as well as visual data. In a preferred embodiment, the CNN running on the DLA is trained to identify the relative closing speed of an emergency vehicle (e.g., using the Doppler effect). The CNN may also be trained to identify emergency vehicles specific to the local area in which the vehicle is operating, as identified by GNSS sensor 458. Thus, for example, when operating in Europe, the CNN attempts to detect European sirens, and when operating in the United States, the CNN attempts to identify only North American sirens. Once an emergency vehicle is detected, with the assistance of ultrasonic sensor 462, the control program can execute emergency vehicle safety routines, such as slowing the vehicle, pulling over to the side of the road, parking, and / or idling the vehicle until the emergency vehicle has passed.

[0093] The vehicle may include a CPU 418 (e.g., a discrete CPU, or dCPU), which may be coupled to the SoC 404 via a high-speed interconnect (e.g., PCIe). The CPU 418 may include, for example, an X86 processor. The CPU 418 may be used to perform any of a variety of functions, such as reconciling potentially inconsistent results between the ADAS sensors and the SoC 404 and / or monitoring the status and health of the controller 436 and / or infotainment SoC 430.

[0094] Vehicle 400 may include a GPU 420 (e.g., a discrete GPU, or dGPU), which may be coupled to SoC 404 via a high-speed interconnect (e.g., NVIDIA NVLINK). GPU 420 may provide additional artificial intelligence functionality, such as by running redundant and / or different neural networks, and may be used to train and / or update neural networks based on input (e.g., sensor data) from sensors of vehicle 400.

[0095] Vehicle 400 may further include a network interface 424, which may include one or more wireless antennas 426 (e.g., one or more wireless antennas for different communication protocols, such as a cellular antenna, a Bluetooth antenna, etc.). Network interface 424 may be used to enable wireless connectivity over the Internet with the cloud (e.g., server 478 and / or other network devices), other vehicles, and / or computing devices (e.g., passenger client devices). To communicate with other vehicles, a direct link may be established between the two vehicles and / or an indirect link may be established (e.g., between networks and via the Internet). A direct link may be provided using a vehicle-to-vehicle communication link. The vehicle-to-vehicle communication link may provide vehicle 400 with information about vehicles in proximity to vehicle 400 (e.g., vehicles in front of, to the sides of, and / or behind vehicle 400). This functionality may be part of a cooperative adaptive cruise control function of vehicle 400.

[0096] The network interface 424 may include an SoC that provides modulation and demodulation functionality, enabling the controller 436 to communicate over a wireless network. The network interface 424 may include a radio frequency front end for upconversion from baseband to radio frequency and downconversion from radio frequency to baseband. The frequency conversion may be performed via well-known processes and / or may be performed using a superheterodyne process. In some examples, the radio frequency front end functionality may be provided by a separate chip. The network interface may include radio functionality for communicating via LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-Wave, ZigBee, LoRaWAN, and / or other wireless protocols. The vehicle 400 may further include a data store 428, which may include off-chip (e.g., off-SoC 404) storage. The data store 428 may include one or more storage elements, including RAM, SRAM, DRAM, VRAM, flash, hard disk, and / or other components and / or devices capable of storing at least one bit of data.

[0097] Vehicle 400 may further include GNSS sensors 458 (e.g., GPS and / or assisted GPS sensors) to assist in mapping, perception, occupancy grid generation, and / or route planning functions. Any number of GNSS sensors 458 may be used, including, by way of example and not limitation, a GPS using a USB connector with an Ethernet to serial (RS232) bridge. Vehicle 400 may further include radar sensor 460. Radar sensor 460 may be used in vehicle 400 for long-range vehicle detection, even in darkness and / or severe weather conditions. The radar functional safety level may be ASIL B. Radar sensor 460 may access control and object tracking data using CAN and / or bus 402 (e.g., transmitting data generated by radar sensor 460), and in some examples, access Ethernet to access raw data. A wide variety of radar sensor types may be used. By way of example and not limitation, radar sensor 460 may be suitable for use with front, rear, and side radar. In some examples, pulse Doppler radar sensors are used.

[0098] The radar sensor 460 may include different configurations, such as long-range with a narrow field of view, short-range with a wide field of view, or short-range lateral coverage. In some examples, long-range radar may be used for adaptive cruise control functions. Long-range radar systems may provide a wide field of view achieved by two or more independent scans, such as within a 250-meter range. The radar sensor 460 may help distinguish between stationary and moving objects and may be used by ADAS systems for emergency brake assistance and forward collision warning. Long-range radar sensors may include monostatic multimodal radar with multiple (e.g., six or more) fixed radar antennas and high-speed CAN and FlexRay interfaces. In an example with six antennas, the center four antennas may generate a focused beam pattern designed to record the vehicle 400's surroundings at higher speeds while minimizing interference from traffic in adjacent lanes. The other two antennas may expand the field of view, allowing for quick detection of vehicles entering and exiting the vehicle's 400 lane.

[0099] A medium-range radar system may include, by way of example, a range of up to 460 meters (forward) or 80 meters (rearward) and a field of view of up to 42 degrees (forward) or 450 degrees (rearward). A short-range radar system may include, but is not limited to, a radar sensor designed to be mounted on either end of the rear bumper. When installed on either end of the rear bumper, such a radar sensor system can generate two beams and constantly monitor the blind spots behind and adjacent to the vehicle.

[0100] ADAS systems may use short-range radar systems for blind spot detection and / or lane change assistance.

[0101] Vehicle 400 may further include ultrasonic sensors 462. The ultrasonic sensors 462, which may be located on the front, rear, and / or sides of vehicle 400, may be used for parking assistance and / or creating and updating an occupancy grid. A variety of ultrasonic sensors 462 may be used, and different ultrasonic sensors 462 may be used for different detection ranges (e.g., 2.5 m, 4 m). Ultrasonic sensors 462 may operate at an ASIL B functional safety level.

[0102] The vehicle 400 may include a lidar sensor 464. The lidar sensor 464 may be used for object and pedestrian detection, emergency braking, collision avoidance, and / or other functions. The lidar sensor 464 may be functional safety level ASIL B. In some examples, the vehicle 400 may include multiple lidar sensors 464 (e.g., 2, 4, 6, etc.) that can use Ethernet (e.g., to provide data to a Gigabit Ethernet switch).

[0103] In some examples, the lidar sensor 464 may be capable of providing a list of objects and their distances for a 360-degree field of view. Commercially available lidar sensors 464 may have an advertised range of approximately 100 meters, with an accuracy of 2 cm to 3 cm, supporting, for example, a 100 Mbps Ethernet connection. In some examples, one or more non-protruding lidar sensors 464 may be used. In such examples, the lidar sensor 464 may be implemented as a small device that can be embedded in the front, rear, sides, and / or corners of the vehicle 400. In such examples, the lidar sensor 464 may provide a horizontal field of view of up to 120 degrees and a vertical field of view of up to 35 degrees, with a range of 200 meters, even for objects with low reflectivity. A forward-mounted lidar sensor 464 may be configured to obtain a horizontal field of view of 45 degrees to 135 degrees.

[0104] In some examples, lidar technology such as 3D flash lidar may also be used. 3D flash lidar uses a laser flash as a transmitter to illuminate the vehicle's surroundings up to approximately 200 meters. Flash lidar includes a receptor that records the transit time of the laser pulse and the reflected light for each pixel, which corresponds to the distance from the vehicle to the object. Flash lidar enables high-precision, distortion-free images of the surroundings to be generated with every laser flash. In some examples, four flash lidar sensors may be placed, one on each side of the vehicle 400. 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. Flash lidar devices use Class I (eye-safe) laser pulses of 5 nanoseconds per frame and can capture reflected laser light in the form of a 3D range point cloud and co-registered intensity data. By using flash lidar, the lidar sensor 464 may be less susceptible to motion blur, vibration, and / or shock because flash lidar is a solid-state device with no moving parts.

[0105] The vehicle may further include an IMU sensor 466. In some examples, the IMU sensor 466 may be located at the center of the rear axle of the vehicle 400. The IMU sensor 466 may include, by way of example and not limitation, an accelerometer, a magnetometer, a gyroscope, a magnetic compass, and / or other types of sensors. In some examples, such as a six-axis application, the IMU sensor 466 may include an accelerometer and a gyroscope, while in a nine-axis application, the IMU sensor 466 may include an accelerometer, a gyroscope, and a magnetometer.

[0106] In some embodiments, the IMU sensor 466 may be implemented as a compact, high-performance GPS-aided Inertial Navigation System (GPS / INS) that combines microelectromechanical systems (MEMS) inertial sensors, a highly sensitive GPS receiver, and advanced Kalman filtering algorithms to provide estimates of position, velocity, and attitude. Thus, in some examples, the IMU sensor 466 may enable the vehicle 400 to estimate heading without requiring input from a magnetic sensor by directly observing and correlating changes in velocity from the GPS to the IMU sensor 466. In some examples, the IMU sensor 466 and the GNSS sensor 458 may be combined into a single integrated unit. The vehicle may include microphones 496 positioned within and / or around the vehicle 400. The microphones 496 may be used for, among other things, emergency vehicle detection and identification.

[0107] The vehicle may further include any number of camera types, including stereo cameras 468, wide-view cameras 470, infrared cameras 472, surround cameras 474, long-range and / or mid-range cameras 498, and / or other camera types. The cameras may be used to capture image data around the entire perimeter of the vehicle 400. The types of cameras used depend on the embodiment and requirements of the vehicle 400, and any combination of camera types may be used to provide the necessary coverage around the vehicle 400. The number of cameras may also 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. The cameras may support Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet, by way of example and not limitation. Each of the cameras is described in more detail herein with respect to FIGS. 4A and 4B.

[0108] The vehicle 400 may further include a vibration sensor 442. The vibration sensor 442 may measure vibrations of vehicle components, such as axles. For example, changes in vibration may indicate changes in the road surface. In another example, when two or more vibration sensors 442 are used, the difference between the vibrations (e.g., when there is a difference in vibration between a powered axle and a free-spinning axle) may be used to determine friction or slippage of the road surface.

[0109] The vehicle 400 may include an ADAS system 438. In some examples, the ADAS system 438 may include an SoC. The ADAS system 438 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.

[0110] The ACC system may use radar sensors 460, lidar sensors 464, and / or cameras. The ACC system may include longitudinal ACC and / or lateral ACC. Longitudinal ACC monitors and controls the distance to the vehicle immediately preceding the vehicle 400 and automatically adjusts vehicle speed to maintain a safe distance from the vehicle ahead. Lateral ACC performs distance maintenance and advises the vehicle 400 to change lanes if necessary. Lateral ACC is relevant to other ADAS applications such as LC and CWS.

[0111] CACC uses information from other vehicles, which may be received from other vehicles via a wireless link or indirectly via a network connection (e.g., via the Internet) via network interface 424 and / or wireless antenna 426. A direct link may be provided by a vehicle-to-vehicle (V2V) communication link, while an indirect link may be an infrastructure-to-vehicle (I2V) communication link. Generally, the V2V communication concept provides information about the immediately preceding vehicle (e.g., a vehicle immediately preceding and in the same lane as vehicle 400), while the I2V communication concept provides information about traffic further ahead. A CACC system may include either or both I2V and V2V information sources. Information about vehicles ahead of vehicle 400 may make CACC more reliable, potentially improving traffic flow and reducing road congestion.

[0112] The FCW system is designed to warn the driver of hazards so that the driver can take corrective action. The FCW system uses a forward-facing camera and / or radar sensor 460 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which are electrically coupled to driver feedback such as a display, speaker, and / or vibration components. The FCW system can provide warnings such as in the form of an audible, visual warning, vibration, and / or quick brake pulse.

[0113] An AEB system can detect an imminent forward collision with another vehicle or other object and automatically apply the brakes if the driver does not take corrective action within specified time or distance parameters. The AEB system can use a forward-facing camera and / or radar sensor 460 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. If the driver does not take corrective action, the AEB system automatically applies the brakes to attempt 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 crash imminent braking.

[0114] The LDW system provides visual, audible, and / or tactile warnings, such as vibration of the steering wheel or seat, to alert the driver if the vehicle 400 crosses a lane marking. If the driver indicates an intentional lane departure by activating a turn signal, the LDW system will not activate. The LDW system may use a forward-facing camera coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to driver feedback such as a display, speaker, and / or vibration components.

[0115] An LKA system is a variation of an LDW system, which provides steering input or braking to correct the vehicle 400 if it begins to drift out of its lane. The BSW system detects vehicles in the vehicle's blind spot and alerts the driver. The BSW system can provide visual, audible, and / or tactile warnings to indicate that merging or changing lanes is unsafe. If the driver uses a turn signal, the system can provide an additional warning. The BSW system can use a rear-facing camera and / or radar sensor 460 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0116] The RCTW system may provide visual, audible, and / or tactile notifications when an object is detected outside the range of the rear camera when the vehicle 400 is backing up. Some RCTW systems include AEB to ensure vehicle braking is applied to avoid a collision. The RCTW system may use one or more rear-facing radar sensors 460 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which are electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0117] Conventional ADAS systems are prone to false-positive detection results, which can be annoying and distracting to the driver, but are typically not fatal because the ADAS system alerts the driver, allowing the driver to determine whether a truly safe condition exists and act accordingly. However, in autonomous vehicle 400, when results conflict, vehicle 400 itself must decide whether to heed the results from the primary computer or the secondary computer (e.g., first controller 436 or second controller 436). For example, in some embodiments, ADAS system 438 may be a backup and / or secondary computer that provides perception information to a backup computer rationality module. The backup computer rationality monitor may run redundant and diverse software on hardware components to detect failures in perception and dynamic driving tasks. Output from ADAS system 438 may be provided to a supervisory MCU. When the outputs from the primary and secondary computers conflict, the supervisory MCU must determine how to reconcile the conflict to ensure safe operation.

[0118] In some examples, the primary computer can be configured to provide a confidence score to the monitor MCU, indicating the primary computer's reliability for the selected result. If the confidence score exceeds a threshold, the monitor MCU can follow the primary computer's instructions regardless of whether the secondary computer provides a conflicting or contradictory result. If the confidence score does not meet the threshold, and if the primary and secondary computers indicate different results (e.g., conflicting), the monitor MCU can arbitrate between the computers to determine the appropriate result.

[0119] The monitoring MCU may be configured to execute a neural network trained and configured to determine, based on outputs from the primary and secondary computers, conditions under which the secondary computer will provide a false alarm. Thus, the monitoring MCU's neural network can learn when the secondary computer's output can be trusted and when it cannot. For example, if the secondary computer is a radar-based FCW system, the monitoring MCU's neural network can learn when the FCW system is identifying a metal object that is not actually a hazard, such as a drain grate or manhole cover that triggers an alarm. Similarly, if the secondary computer is a camera-based LDW system, the monitoring MCU's neural network can learn to disable LDW when a bicyclist or pedestrian is present and lane departure is actually the safest maneuver. In embodiments including a neural network executing on the monitoring MCU, the monitoring MCU may include at least one of a DLA or GPU suitable for executing the neural network with associated memory. In preferred embodiments, the monitoring MCU may be provided and / or included as a component of the SoC 404.

[0120] In other examples, the ADAS system 438 may include a secondary computer that performs ADAS functions using traditional rules of computer vision. Thus, the secondary computer may use traditional computer vision rules, in which case the presence of a neural network in the supervisory MCU may improve reliability, safety, and performance. For example, diverse implementations and intentional non-identity may make the overall system more fault-tolerant, particularly to faults caused by software (or software-hardware interface) functions. For example, if software running on the primary computer has a software bug or error, and non-identical software code running on the secondary computer provides the same overall result, the supervisory MCU may have higher confidence that the overall result is correct and that a bug in the software or hardware used by the primary computer did not cause a critical error.

[0121] In some examples, the output of the ADAS system 438 may be provided to a perception block of the primary computer and / or a dynamic driving task block of the primary computer. For example, if the ADAS system 438 indicates a forward collision warning due to an immediately preceding object, the perception block can use this information when identifying the object. In other examples, the secondary computer may have its own neural network that is trained, as described herein, thus reducing the risk of false positives.

[0122] Vehicle 400 may further include an infotainment SoC 430 (e.g., an in-vehicle infotainment system (IVI)). While illustrated and described as an SoC, an infotainment system need not be an SoC and may include two or more separate components. Infotainment SoC 430 may include a combination of hardware and software that may be used to provide audio (e.g., music, personal digital assistants, navigation instructions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), telephony (e.g., hands-free calling, etc.), network connectivity (e.g., LTE, Wi-Fi, etc.), and / or information services (e.g., navigation systems, rear park assist, wireless data systems, and vehicle-related information such as fuel level, total mileage, brake fuel level, oil level, door opening / closing, air filter information, etc.) to vehicle 400. For example, the infotainment SoC 430 may include a radio, a disc player, a navigation system, a video player, USB and Bluetooth connectivity, a computer, in-car entertainment, Wi-Fi, steering wheel audio controls, hands-free voice control, a head-up display (HUD), an HMI display 434, telematics devices, a control panel (e.g., for controlling and / or interacting with various components, functions, and / or systems), and / or other components. The infotainment SoC 430 may further be used to provide information (e.g., visually and / or audibly) to a user of the vehicle, such as information from an ADAS system 438, automated driving information such as planned vehicle maneuvers, trajectories, surrounding environment information (e.g., intersection information, vehicle information, road information, etc.), and / or other information.

[0123] The infotainment SoC 430 may include GPU functionality. The infotainment SoC 430 may communicate with other devices, systems, and / or components of the vehicle 400 via the bus 402 (e.g., a CAN bus, Ethernet, etc.). In some examples, the infotainment SoC 430 may be coupled to a supervisory MCU so that the infotainment system's GPU can perform some self-driving functions even if the primary controller 436 (e.g., the vehicle's 400 primary and / or backup computer) fails. In such examples, the infotainment SoC 430 may put the vehicle 400 into a safe shutdown mode from the driver, as described herein.

[0124] The vehicle 400 may further include an instrument cluster 432 (e.g., a digital dashboard, an electronic instrument cluster, a digital instrument panel, etc.). The instrument cluster 432 may include a controller and / or a supercomputer (e.g., a separate controller or supercomputer). The instrument cluster 432 may include a set of instruments such as a speedometer, a fuel level, an oil pressure, a tachometer, an odometer, a turn signal, a gear shift position indicator, a seat belt warning light, a parking brake warning light, an engine malfunction light, an airbag (SRS) system information, lighting control, safety system control, navigation information, etc. In some examples, information may be displayed and / or shared between the infotainment SoC 430 and the instrument cluster 432. In other words, the instrument cluster 432 may be included as part of the infotainment SoC 430, or vice versa.

[0125] 4D is a system diagram for communication between a cloud-based server and the example autonomous vehicle 400 of FIG. 4A , according to some embodiments of the present disclosure. System 476 may include a server 478, a network 490, and a vehicle including vehicle 400. Server 478 may include multiple GPUs 484(A)-484(H) (collectively referred to herein as GPUs 484), PCIe switches 482(A)-482(H) (collectively referred to herein as PCIe switches 482), and / or CPUs 480(A)-480(B) (collectively referred to herein as CPUs 480). GPUs 484, CPUs 480, and PCIe switches may be interconnected with a high-speed interconnect, such as an NVLink interface 488 developed by NVIDIA and / or PCIe connections 486, by way of example and not limitation. In some examples, the GPUs 484 are connected via NVLink and / or NVSwitch SoCs, and the GPUs 484 and PCIe switches 482 are connected via PCIe interconnects. While eight GPUs 484, two CPUs 480, and two PCIe switches are shown, this is not intended to be limiting. Depending on the embodiment, each of the servers 478 may include any number of GPUs 484, CPUs 480, and / or PCIe switches. For example, the servers 478 may each include 8, 16, 32, and / or more GPUs 484.

[0126] Server 478 may receive image data from the vehicle via network 490, representing images showing unexpected or changed road conditions, such as recently started road construction. Server 478 may transmit neural network 492, updated neural network 492, and / or map information 494, including information about traffic and road conditions, to the vehicle via network 490. Updates to map information 494 may include updates to HD map 422, such as information about construction sites, potholes, detours, flooding, and / or other obstacles. In some examples, neural network 492, updated neural network 492, and / or map information 494 may result from new training and / or experience represented by data received from any number of vehicles in the environment and / or based on training performed at a data center (e.g., using server 478 and / or other servers).

[0127] The server 478 may be used to train a machine learning model (e.g., a neural network) based on the training data. The training data may be generated by the vehicle and / or generated in a simulation (e.g., using a game engine). In some examples, the training data is tagged (e.g., if the neural network benefits from supervised learning) and / or otherwise preprocessed, while in other examples, the training data is not tagged and / or preprocessed (e.g., if the neural network does not require supervised learning). The training may be performed according to one or more classes of machine learning techniques, including, but not limited to, classes such as supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, federated learning, transfer learning, feature learning (including principal component analysis and cluster analysis), multilinear subspace learning, manifold learning, representation learning (including preliminary dictionary learning), rule-based machine learning, anomaly detection, and variations or combinations thereof. Once the machine learning model is trained, it may be used in the vehicle (e.g., transmitted to the vehicle via the network 490) and / or the machine learning model may be used by the server 478 to remotely monitor the vehicle.

[0128] In some examples, server 478 can receive data from vehicles and apply the data to state-of-the-art real-time neural networks for real-time intelligent inference. Server 478 can include deep learning supercomputers and / or dedicated AI computers powered by GPUs 484, such as the DGX and DGX Station machines developed by NVIDIA. However, in some examples, server 478 can include deep learning infrastructure that uses only CPU-powered data centers.

[0129] The deep learning infrastructure of server 478 is capable of rapid real-time inference and can use that capability to evaluate and verify the state of processors, software, and / or associated hardware within vehicle 400. For example, the deep learning infrastructure can receive periodic updates from vehicle 400, such as a series of images and / or objects that vehicle 400 has located in the series of images (e.g., via computer vision and / or other machine learning object classification techniques). The deep learning infrastructure can run its own neural network to identify objects and compare those objects to those identified by vehicle 400; if the results do not match and the infrastructure concludes that vehicle 400's AI is malfunctioning, server 478 can send a signal to vehicle 400 instructing vehicle 400's failsafe computer to assume control, notify the occupants, and complete a safe parking maneuver.

[0130] For inference, server 478 may include a GPU 484 and one or more programmable inference accelerators (e.g., NVIDIA's TensorRT). Combining a GPU-powered server with inference acceleration can enable real-time responsiveness. In other instances, such as when performance is less critical, servers powered by CPUs, FPGAs, and other processors may be used for inference.

[0131] 5 is a block diagram of an exemplary computing device 500 suitable for use in implementing some embodiments of the present disclosure. Computing device 500 may include an interconnection system 502 that directly or indirectly couples the following devices: memory 504, one or more central processing units (CPUs) 506, one or more graphics processing units (GPUs) 508, a communications interface 510, I / O ports 512, input / output components 514, a power supply 516, one or more presentation components 518 (e.g., a display), and one or more logic units 520.

[0132] While the various blocks in FIG. 5 are shown connected by lines via interconnection system 502, this is not intended to be limiting and is for clarity only. For example, in some embodiments, a presentation component 518, such as a display device, may be considered an I / O component 514 (e.g., if the display is a touchscreen). As another example, CPU 506 and / or GPU 508 may include memory (e.g., memory 504 may represent storage in addition to memory for GPU 508, CPU 506, and / or other components). In other words, the computing devices of FIG. 5 are merely exemplary. Categories such as “workstation,” “server,” “laptop,” “desktop,” “tablet,” “client device,” “mobile device,” “handheld device,” “game console,” “electronic control unit (ECU),” “virtual reality system,” “augmented reality system,” and / or other device or system types are not distinguished, as they are all contemplated within the scope of the computing devices of FIG. 5.

[0133] Interconnect system 502 may represent one or more links or buses, such as an address bus, a data bus, a control bus, or a combination thereof. Interconnect system 502 may include an Industry Standard Architecture (ISA) bus, an Extended Industry Standard Architecture (EISA) bus, a Video Electronics Standards Association (VESA) bus, a Peripheral Component Interconnect (PCI) bus, a Peripheral Component Interconnect Express (PCIe) bus, and / or another type of bus or link. In some embodiments, there are direct connections between components. As an example, CPU 506 may be directly connected to memory 504. Further, CPU 506 may be directly connected to GPU 508. When there are direct or point-to-point connections between components, interconnect system 502 may include a PCIe link to effectuate the connections. In these examples, a PCI bus need not be included in computing device 500.

[0134] Memory 504 may include any of a variety of computer-readable media. Computer-readable media may be any available media that can be accessed by computing device 500. Computer-readable media may include both volatile and nonvolatile media, and removable and non-removable media. By way of example, and not limitation, computer-readable media may include computer storage media and communication media.

[0135] Computer storage media may include both 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 504 may store computer-readable instructions (e.g., representing programs and / or 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 (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by computing device 500. As used herein, computer storage media does not itself comprise signals.

[0136] Computer storage media may 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 may 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.

[0137] The CPUs 506 may be configured to execute at least some of the computer-readable instructions and to control one or more components of the computing device 500 to perform one or more of the methods and / or processes described herein. The CPUs 506 may each include one or more cores (e.g., 1, 2, 4, 8, 28, 72, etc.) capable of simultaneously processing multiple software threads. The CPUs 506 may include any type of processor, and may include different types of processors depending on the type of computing device 500 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 500, 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 500 may include one or more CPUs 506 in addition to one or more microprocessors or auxiliary coprocessors, such as mathematical coprocessors.

[0138] In addition to, or as an alternative to, the CPU 506, the GPU 508 may be configured to execute at least some of the computer-readable instructions and control one or more components of the computing device 500 to perform one or more of the methods and / or processes described herein. One or more of the GPUs 508 may be integrated GPUs (e.g., with one or more CPUs 506) and / or one or more of the GPUs 508 may be discrete GPUs. In an embodiment, one or more of the GPUs 508 may be coprocessors of one or more of the CPUs 506. The GPU 508 may be used by the computing device 500 to render graphics (e.g., 3D graphics) or perform general-purpose computations. For example, the GPU 508 may be used for GPU-generated general-purpose computing (GPGPU). The GPU 508 may include hundreds or thousands of cores capable of simultaneously processing hundreds or thousands of software threads. The GPU 508 may generate pixel data for an output image in response to a rendering command (e.g., a rendering command from the CPU 506 received via a host interface). GPU 508 may include graphics memory, such as display memory, for storing pixel data or any other suitable data, such as GPGPU data. The display memory may be included as part of memory 504. GPU 508 may include two or more GPUs operating in parallel (e.g., via links). The links may connect the GPUs directly (e.g., using NVLINK) or through a switch (e.g., using NVSwitch). When combined together, each GPU 508 may generate pixel data or GPGPU data for a different portion of the output or for a different output (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. In addition to, or as an alternative to, the CPU 506 and / or GPU 508, the logic unit 520 may be configured to execute at least some of the computer-readable instructions and control one or more components of the computing device 500 to perform one or more of the methods and / or processes described herein. In an embodiment, the CPU 506, the GPU 508, and / or the logic unit 520 may individually or jointly execute any combination of methods, processes, and / or portions thereof. One or more of the logic units 520 may be part of and / or integrated with one or more of the CPUs 506 and / or GPUs 508, and / or one or more of the logic units 520 may be separate components or otherwise external to the CPUs 506 and / or GPUs 508. In an embodiment, one or more of the logic units 520 may be a co-processor of one or more of the CPUs 506 and / or one or more of the GPUs 508.

[0139] Examples of logic unit 520 include one or more processing cores and / or components thereof, such as tensor cores (TCs), tensor processing units (TPUs), pixel visual cores (PVCs), vision processing units (VPUs), graphics processing clusters (GPCs), texture processing clusters (TPCs), streaming multiprocessors (SMs), tree traversal units (TTUs), artificial intelligence accelerators (AIAs), deep learning accelerators (DLAs), arithmetic logic units (ALUs), application specific integrated circuits (ASICs), floating point units (FPUs), I / O elements, peripheral component interconnect (PCI) or peripheral component interconnect express (PCIe) elements, etc.

[0140] The communications interface 510 may include one or more receivers, transmitters, and / or transceivers that enable the computing device 500 to communicate with other computing devices over electronic communications networks, including wired and / or wireless communications. The communications interface 510 may include components and functionality that enable communication over any of several different networks, such as a wireless network (e.g., Wi-Fi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), a wired network (e.g., communication over Ethernet or InfiniBand), a low-power wide area network (e.g., LoRaWAN, SigFox, etc.), and / or the Internet.

[0141] The I / O ports 512 may enable the computing device 500 to be logically coupled to other devices, including an I / O component 514, a presentation component 518, and / or other I / O components, some of which may be built-in (e.g., integrated) to the computing device 500. Exemplary I / O components 514 include a microphone, mouse, keyboard, joystick, gamepad, game controller, satellite dish, scanner, printer, wireless device, etc. The I / O component 514 may provide a natural user interface (NUI) that processes air gestures, voice, or other physiological input generated by the user. In some cases, the input may be sent to an appropriate network element for further processing. The NUI may implement any combination of speech recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition both on-screen and adjacent to the screen, air gestures, head and eye tracking, and touch recognition (described in more detail below) associated with the display of the computing device 500. Computing device 500 may include a depth camera, such as a stereoscopic camera system, an infrared camera system, an RGB camera system, touchscreen technology, and combinations thereof, for gesture detection and recognition. Additionally, computing device 500 may include an accelerometer or gyroscope (e.g., as part of an inertial measurement unit (IMU)) that enables detection of movement. In some examples, the output of the accelerometer or gyroscope may be used by computing device 500 to render immersive augmented or virtual reality.

[0142] The power supply 516 may include a hardwired power supply, a battery power supply, or a combination thereof. The power supply 516 may provide power to the computing device 500 to enable the components of the computing device 500 to operate. The presentation component 518 may include a display (e.g., a monitor, a touch screen, a television screen, a heads-up display (HUD), other display types, or a combination thereof), speakers, and / or other presentation components. The presentation component 518 can receive data from other components (e.g., the GPU 508, the CPU 506, etc.) and output data (e.g., as images, video, sound, etc.).

[0143] The present disclosure may be described in the general context of computer code or machine-usable instructions executed by a computer or other machine, such as a personal data 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 particular tasks or implements particular abstract data types. The present disclosure may be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, more specialized computing devices, etc. The present disclosure may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.

[0144] As used herein, the listing of "and / or" with respect to two or more elements should be interpreted to mean only one element or a combination of elements. For example, "element A, element B, and / or element C" can include element A only, element B only, element C only, elements A and element B, elements A and element C, elements B and element C, or elements A, B, and C. Furthermore, "at least one of element A or element B" can 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. Furthermore, "at least one of element A and element B" can 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.

[0145] The subject matter of the present disclosure is described with specificity herein to satisfy statutory requirements. However, the description itself is not intended to limit the scope of this disclosure. Rather, the inventors contemplate that the claimed subject matter may be embodied in other ways and may include different steps or combinations of steps similar to those described in this specification, in conjunction with other current or future technologies. Furthermore, although the terms "step" and / or "block" may be used herein to connote different elements of a method used, these terms should not be construed as implying a particular order among the various steps disclosed herein unless and until the order of individual steps is explicitly stated.

[0146] Referring now to FIG. 6, each block of the methods depicted in FIG. 6 described herein includes computing processes that may be implemented using any combination of hardware, firmware, and / or software. For example, various functions may be performed by a processor executing instructions stored in a memory. These methods may also be embodied as computer-usable instructions stored on a computer storage medium. These methods may be provided, for example, by a standalone application, a service, or a hosted service (standalone or in combination with another hosted service), or as a plug-in to another product. Additionally, the method of FIG. 6 is described with respect to the exemplary autonomous vehicle system of FIGS. 4A-4D by way of example. However, these methods may additionally or alternatively be implemented by any system or combination of systems, including, but not limited to, those described herein.

[0147] FIG. 6 is a flowchart illustrating process steps for determining landmarks and confidence values ​​according to an embodiment of the present disclosure. The process of FIG. 6 may begin when computing device 300 receives an image of a subject captured by camera 300 (step 600). Next, computing device 300 identifies the subject's face and eyes in the received image (step 610). The subject's face can be located in the image using any method or process, including edge detection methods, feature search methods, probabilistic face models, graph matching, and known computer vision-based face detection processes that detect faces without using neural networks, such as HOG fed into a classifier, such as a support vector machine or HaarCascade classifier. Face localization can also be performed using neural network-based face recognition methods, such as methods employing DNN face recognition schemes, as well as other methods. Embodiments of the present disclosure also contemplate locating the subject's eyes from the same image. Eye localization can be performed in any manner, such as known computer vision-based eye detection processes, including the non-neural network-based techniques described above, neural network-based eye recognition methods, etc.

[0148] Once the subject's face and eyes are located within the received image, the computing device 300 determines landmarks and confidence values ​​for the located face (step 620). Facial landmarks can be any points or locations corresponding to distinctive elements or representative parts of a face and can be identified using any method or process. Facial landmarks can be located within the image using edge detection methods, feature search methods, probabilistic face models, graph matching, and known computer vision-based landmark detection processes that identify landmarks without using neural networks, such as HOG, which is fed into a classifier, such as a support vector machine or the HaarCascade classifier. Landmark detection can also be performed using neural network-based face recognition methods, such as methods employing DNN-based face recognition schemes, as well as any other method. The confidence value can be determined in any manner, such as by a confidence value output that can correspond to or be a confidence score generated by a classifier when determining a landmark classification.

[0149] Once the landmarks and confidence values ​​are determined, the face identification of step 610 may be modified (step 630). The confidence values ​​can be used to improve face detection by indicating which portions of the subject's face are occluded. More specifically, computing device 300 can use the confidence values ​​to select only those portions of the face that are not sufficiently occluded. Computing device 300 can, for example, redraw a bounding box around only the unoccluded portions of the subject's face, such that the output of face detection module 210 of FIG. 2 is a crop of only the unoccluded portions of the subject's face. In this way, determination of the subject's state, such as gaze, can be focused only on facial features that are clearly visible in the subject's image, potentially producing more accurate results.

[0150] The facial landmarks and confidence values ​​may also be sent to any other models from which further quantities may be determined (step 640). For example, the landmarks and confidence values ​​may be sent to any gaze assistance system 320 for determining gaze and other quantities as described above. Any action of any system may then be initiated in response to the quantities determined in step 640 (step 670).

[0151] Training of any model of embodiments of the present disclosure may be performed in any suitable manner. In certain embodiments, models of embodiments of the present disclosure may be trained in a supervised manner using images of subjects in various occlusion states, with corresponding facial features labeled as occluded or unoccluded, such as with a binary occluded / unoccluded flag or occlusion as a continuous quantity. In at least one embodiment, any backpropagation process using any loss function may be used to train models of embodiments of the present disclosure. Backpropagation may be performed using only unoccluded points, e.g., with a loss that does not account for occluded points. Alternatively, the loss function may take occluded points into account. In some embodiments, the facial landmark network 220 may be trained separately according to any loss function that may be a function of any one or more outputs of the network 220. In other embodiments, each of the networks 220, 230, 240, and 250, i.e., the facial landmark network 220 and any downstream networks, may be trained end-to-end.

[0152] The foregoing description, for purposes of explanation, used specific nomenclature to provide a thorough understanding of the present disclosure. However, it will be apparent to those skilled in the art that specific details are not required to practice the methods and systems of the present disclosure. Accordingly, the foregoing descriptions of specific embodiments of the present invention have been presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the present invention to the precise forms disclosed. Many modifications and variations are possible in light of the above teachings. For example, facial landmarks and confidence values ​​may be determined in any manner and used, along with any other relevant information, to determine any feature or attribute of a subject. The embodiments have been chosen and described to best explain the principles of the present invention and its practical applications, so as to enable others skilled in the art to optimally utilize the methods and systems of the present disclosure, as well as various embodiments with various modifications as may be appropriate for the particular applications contemplated. Furthermore, different features of the various embodiments, whether disclosed or not, may be mixed and matched or otherwise combined to create further embodiments contemplated by the present disclosure.

Claims

1. 1. A method for selecting landmarks of a subject in an image, comprising: using parallel processing circuitry to determine landmark points and corresponding confidence values ​​for the subject, the landmark points and confidence values ​​being determined according to the output of one or more machine learning models having image data representing the subject as input; using the landmark points and the confidence values ​​to monitor the subject's condition; Including, The method, wherein the confidence value corresponds to an amount of occlusion of a portion of the subject image data corresponding to each of the landmark points.

2. the one or more machine learning models are a first set of one or more machine learning models, and the using The subject's gaze direction the subject's head posture, or the subject's state of drowsiness 10. The method of claim 1, further comprising transmitting the landmark points and the confidence values ​​for use by a second set of one or more machine learning models configured to generate as an output at least one of:

3. 2. The method of claim 1, wherein the one or more machine learning models are a first set of one or more machine learning models, and wherein the using further comprises transmitting the landmark points and the confidence values ​​for use with a second set of one or more machine learning models configured to generate as an output one or both of the subject's cognitive load or the subject's distraction state.

4. 2. The method of claim 1, wherein the input image data includes first input image data corresponding to a first portion of the image, the method further comprising selecting second input image data from a second portion of the image selected according to the determined confidence value.

5. 5. The method of claim 4, wherein the one or more machine learning models are a first set of one or more machine learning models, the method further comprising transmitting only the second input image data for use by a second set of one or more machine learning models.

6. The method of claim 1 , further comprising selecting a subset of the landmark points having respective confidence values ​​that correspond to unoccluded portions of the subject.

7. 7. The method of claim 6, wherein the one or more machine learning models are a first set of one or more machine learning models, the method further comprising transmitting the selected subset of the landmark points for use by a second set of one or more machine learning models.

8. The method of claim 1 , further comprising assigning a weight value to each of the landmark points according to the corresponding confidence value.

9. The method of claim 8 , wherein the weight value is inversely proportional to the corresponding confidence value.

10. The method of claim 1 , wherein said using further comprises initiating an action based on said condition of said subject.

11. The method of claim 10 , wherein the movement comprises a movement of a vehicle.

12. 1. A system for selecting landmarks of a subject in an image, comprising: Memory and 1. A parallel processing circuit, comprising: determining landmark points and corresponding confidence values ​​for the subject, the landmark points and confidence values ​​being determined according to the output of one or more machine learning models having image data representing the subject as input; and a parallel processing circuit configured to use the landmark points and the confidence values ​​to monitor a condition of the subject; Including, The system, wherein the confidence value corresponds to an amount of occlusion of a portion of the subject image data corresponding to each of the landmark points.

13. the one or more machine learning models are a first set of one or more machine learning models, and the using The subject's gaze direction the subject's head posture, or the subject's state of drowsiness 13. The system of claim 12, further comprising transmitting the landmark points and the confidence values ​​for use in a second set of one or more machine learning models configured to generate as an output at least one of:

14. 13. The system of claim 12, wherein the one or more machine learning models are a first set of one or more machine learning models, and wherein the using further comprises transmitting the landmark points and the confidence values ​​for use with a second set of one or more machine learning models configured to generate as an output one or both of the subject's cognitive load or the subject's distraction state.

15. 13. The system of claim 12, wherein the input image data includes first input image data corresponding to a first portion of the image, the method further comprising selecting second input image data from a second portion of the image selected according to the determined confidence value.

16. 16. The system of claim 15, wherein the one or more machine learning models are a first set of one or more machine learning models, the method further comprising transmitting only the second input image data for use by a second set of one or more machine learning models.

17. The system of claim 12 , further comprising selecting a subset of the landmark points having respective confidence values ​​that correspond to unoccluded portions of the subject.

18. 20. The system of claim 17, wherein the one or more machine learning models are a first set of one or more machine learning models, and the method further comprises transmitting the selected subset of the landmark points for use by a second set of one or more machine learning models.

19. The system of claim 12 , further comprising assigning a weight value to each of the landmark points according to the corresponding confidence value.

20. 20. The system of claim 19, wherein the weight value is inversely proportional to the corresponding confidence value.

21. The system of claim 12 , wherein said using further comprises initiating an action based on said condition of said subject.

22. The system of claim 21 , wherein the operation comprises a vehicle operation.

23. A non-transitory computer-readable medium, comprising: When executed by a parallel processing circuit, the parallel processing circuit: determining landmark points and corresponding confidence values ​​for the subject, the landmark points and confidence values ​​being determined according to the output of one or more machine learning models having image data representing the subject as input; and instructions encoded to use the landmark points and the confidence values ​​to monitor a condition of the subject; A non-transitory computer-readable medium, wherein the confidence values ​​correspond to an amount of occlusion of a portion of the subject image data corresponding to each of the landmark points.

24. the one or more machine learning models are a first set of one or more machine learning models, and the using The subject's gaze direction the subject's head posture, or the subject's state of drowsiness 24. The non-transitory computer-readable medium of claim 23, further comprising transmitting the landmark points and the confidence values ​​for use by a second set of one or more machine learning models configured to generate as an output at least one of:

25. 24. The non-transitory computer-readable medium of claim 23, wherein the one or more machine learning models are a first set of one or more machine learning models, and wherein the using further comprises transmitting the landmark points and the confidence values ​​for use with a second set of one or more machine learning models configured to generate as an output one or both of the subject's cognitive load or the subject's distraction state.

26. 24. The non-transitory computer-readable medium of claim 23, wherein the input image data includes first input image data corresponding to a first portion of the image, the method further comprising selecting second input image data from a second portion of the image selected according to the determined confidence value.

27. 27. The non-transitory computer-readable medium of claim 26, wherein the one or more machine learning models are a first set of one or more machine learning models, the method further comprising transmitting only the second input image data for use by a second set of one or more machine learning models.

28. 24. The non-transitory computer-readable medium of claim 23, further comprising selecting a subset of the landmark points having respective confidence values ​​that correspond to unoccluded portions of the subject.

29. 30. The non-transitory computer-readable medium of claim 28, wherein the one or more machine learning models are a first set of one or more machine learning models, and the method further comprises transmitting the selected subset of the landmark points for use by a second set of one or more machine learning models.

30. 24. The non-transitory computer-readable medium of claim 23, further comprising assigning a weight value to each of the landmark points according to the corresponding confidence value.

31. 31. The non-transitory computer-readable medium of claim 30, wherein the weight value is inversely proportional to the corresponding confidence value.

32. 24. The non-transitory computer-readable medium of claim 23, wherein said using further comprises initiating an action based on said condition of said subject.

33. The non-transitory computer-readable medium of claim 32 , wherein the operation comprises an operation of a vehicle.

Images