Gaze determination using glare as input

Abstract

A machine learning system and method that learns about glare, thereby determining gaze direction in a manner that is more resilient to the effects of glare on an input image. The machine learning system has an independent representation of glare, e.g., information about the locations of glare points in the image, as an explicit input in addition to the image itself. In this way, the machine learning system explicitly takes glare into account when determining gaze direction, thereby producing more accurate results for images containing glare.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. non-provisional patent application No. 16 / 902,737, filed on June 16, 2020, which claims the benefit of U.S. provisional patent application No. 62 / 948,793, filed on December 16, 2019, the disclosure of each of which is incorporated herein by reference in its entirety. Background Art

[0003] Embodiments of the present disclosure generally relate to machine learning systems. More specifically, embodiments of the present disclosure relate to gaze determination performed using a machine learning system with glare as input. Summary of the invention

[0004] Recently, convolutional neural networks (CNNs) have been developed to estimate the gaze direction of objects represented in images. For example, such CNNs can infer the gaze direction of an object from an input image of the object by determining certain features about the object's eyes. This enables systems using such CNNs to automatically determine the direction in which the object is looking and react accordingly in real time.

[0005] However, conventional gaze determination systems are not without their shortcomings. Such systems often have difficulty determining the direction of a subject's gaze when one or both eyes are occluded or not clearly visible in the input image. Conventional CNN systems often have particular difficulty determining gaze direction in the presence of glare. Light from various sources often reflects off the eyes, glasses, or other nearby surfaces, resulting in bright spots in the image of a subject whose distance from the image sensor (e.g., camera) may at least partially occlude one or her eyes, thereby reducing the accuracy of conventional gaze determination systems.

[0006] Existing systems have attempted to compensate or reduce the effects of glare using a variety of methods, such as by modifying the light source used in the lighting, using various polarization techniques, identifying and removing glare pixels, or using other information (such as head pose) to compensate for the lack of eye information. However, the effectiveness of each of these methods has been shown to be limited. Therefore, this article describes systems and methods for a more robust gaze estimation system that incorporates glare points as explicit inputs to a gaze estimation machine learning network architecture. An exemplary gaze estimation system can utilize a camera or other image determination device, and a processor (such as a parallel processor) capable of performing inference operations of a machine learning network (such as a CNN). In certain embodiments of the present disclosure, the system can receive an image of an object captured by a camera. One or more machine learning networks are constructed to receive inputs: an isolated representation of glare in the image of the object, one or more representations of at least a portion of the face of the object, and a portion of the image corresponding to at least one eye of the object. From these inputs, the machine learning network determines and outputs an estimate of the gaze direction of the object as shown in the image. The gaze direction can be transmitted to various systems that initiate operations based on the determined gaze.

[0007] The representation of the glare, the representation of the subject's face, and the portion of the image corresponding to the subject's eyes may be determined in any manner. For example, the system may determine each of the representations of the glare, the subject's face, and the portion of the image corresponding to the subject's eyes, respectively, prior to processing through the above-described machine learning network. The representation of the glare may be a binary mask of glare points in the input image determined in any manner, and the representation of the subject's face may be a similarly determined binary mask of the portion of the subject's face including at least his or her eyes. The portion of the image corresponding to the eyes may be a crop of the eyes, where any shape or object recognition technique is used to identify the eyes, whether based on machine learning or otherwise.

[0008] In one embodiment of the present disclosure, a machine learning model may take as input an isolated representation of glare (e.g., a binary mask of glare points) and a representation of at least a portion of a subject's face (e.g., a binary mask of a portion of the subject's face including at least his or her eyes).

[0009] In another embodiment of the present disclosure, an isolated representation of glare can be input to one machine learning model, and a representation of at least a portion of the subject's face can be input to another machine learning model. In particular, the representation of glare can be input to a set of fully connected layers, and the representation of at least a portion of the face can be input to a feature extraction model.

[0010] In one embodiment of the present disclosure, one machine learning model takes as input a portion of an image corresponding to one eye of a subject, while another machine learning model takes as input another portion of an image corresponding to another eye of the subject. That is, two different machine learning models take as input two eye crops taken from an image of a subject.

[0011] In another embodiment of the present disclosure, one machine learning model has as input a representation of an object's face, e.g., the above-mentioned binary code of the object's face determined from an input image, and another machine learning model has as input another representation of the object's face, e.g., a coarse facial mesh that can also be determined from the input image.

[0012] The final output of the machine learning model is an estimated gaze direction or focus. Any action can be initiated in response to this output. For example, the output of the machine learning model can be the gaze direction of the driver of the vehicle, and the initiated action can be a vehicle operation performed in response to the determined gaze direction. For example, upon determining that the driver is not looking in the direction the vehicle is traveling, the vehicle can issue a warning to the driver to redirect his or her attention back to the road.

[0013] Therefore, the embodiments of the present disclosure can be viewed as providing a machine learning system that determines the gaze direction of an object by explicitly learning glare. More specifically, the location of the glare points in the image is used as a separate input to the machine learning model. Therefore, the machine learning model of the embodiments of the present disclosure takes as input the image of the object and the explicit representation of the glare points extracted from the image. Based on these (and optionally other) inputs, the gaze direction is determined. Various operations can then be initiated based on the determined gaze direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other objects and advantages of the present disclosure will become apparent upon consideration of the following detailed description in conjunction with the accompanying drawings, wherein like reference numerals refer to like parts throughout, and wherein:

[0015] FIG1A conceptually illustrates the functioning of a conventional gaze determination process in the presence of glare;

[0016] Figure 1B conceptually illustrates the functionality of the anti-glare gaze determination process of an embodiment of the present disclosure;

[0017] Figure 2 is a block diagram representation of an exemplary anti-glare machine learning model of an embodiment of the present disclosure;

[0018] Figure 3 is a block diagram representation of a gaze determination system of an embodiment of the present disclosure;

[0019] Figure 4Ais an illustration of an example autonomous vehicle according to some embodiments of the present disclosure;

[0020] Figure 4B According to some embodiments of the present disclosure Figure 4A Examples of camera locations and fields of view for exemplary autonomous vehicles;

[0021] Figure 4C According to the following embodiment of the present disclosure Figure 4A A block diagram of an exemplary system architecture of an exemplary autonomous vehicle;

[0022] Figure 4D A cloud-based server and Figure 4A A system diagram of an exemplary autonomous vehicle-to-vehicle communication system;

[0023] Figure 5 is a block diagram of an example computing device suitable for implementing some embodiments of the present disclosure;

[0024] Figure 6 is a flow chart illustrating process steps for determining gaze direction in the presence of glare according to an embodiment of the present disclosure; and

[0025] Figure 7 is a block diagram representation of a further exemplary anti-glare machine learning model of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] In one embodiment, the present disclosure relates to machine learning systems and methods that learn about glare and thus determine gaze direction in a manner that is resistant to the effects of glare. In addition to the image itself, the machine learning system has an isolated glare representation, e.g., information about the location of glare points in the image, as an explicit input. In this way, the machine learning system explicitly considers glare when determining gaze direction, thereby producing more accurate results for images containing glare.

[0027] FIG. 1A conceptually illustrates the functionality of a conventional gaze determination process in the presence of glare. Typically, conceptually, an image 100 is input to a CNN 110 that is trained to estimate the gaze direction of an object in the input image 100. The presence of glare typically prevents this process from producing satisfactory results. Here, for example, glare spots 130 present on the subject's glasses may obscure the subject's eyes, making it difficult for the CNN 110 to detect enough of the subject's eyes to determine his gaze direction with a high degree of confidence. Since the CNN 110 does not have enough information to determine the gaze, the output of the CNN 110 may be inaccurate, meaningless, or the affected input image may be completely discarded.

[0028] Figure 1BThe functionality of the anti-glare gaze determination process of an embodiment of the present disclosure is conceptually illustrated. Compared to the traditional CNN 110 of FIG. 1A , the embodiment of the present disclosure describes a machine learning model that specifically learns about glare and therefore counteracts its effects. Therefore, when the image 100 and the isolated representation of the glare point 120 extracted from the image 100 are input to the machine learning model 140 of the embodiment of the present disclosure, the model 140 is still able to infer the gaze direction of the subject with sufficient confidence, resulting in an accurate output of the gaze vector even in the presence of the glare point 130 that may partially occlude the subject's eyes. FIG. 1A and Figure 1B An image 100 is shown as input to the machine learning model 140 to conceptually illustrate that the machine learning model 140 takes as input information extracted from or related to an image captured of a subject, noting that in practice, the input to the machine learning model 140 may be a different representation or set of information other than an image. For example, the input to the model 140 may be a facial mesh containing glare points, as further described below.

[0029] Figure 2 is a block diagram representation of an exemplary anti-glare machine learning model of an embodiment of the present disclosure. Figure 2 The machine learning model includes feature extraction layers 230, 235 and 240 and fully connected layers 245 and 255 arranged as shown. In particular, the feature extraction layer 230 has two different inputs, the glare representation 200 and the face representation 205, which are connected by the connection block 210. The connected inputs are then transmitted to the feature extraction layer 230. The feature extraction block 230 can extract features of the input image 100 according to any method. For example, the feature extraction block 230 can be a feature learning portion of a CNN, constructed in any suitable manner and using any convolution kernel and pooling layer suitable for extracting features of the object expected to be captured in the input image. Therefore, the feature extraction block 230 outputs a feature that can represent the orientation of the face of the object represented in the input image.

[0030] The feature extraction layers 235 and 240 may have a single input, a left eye representation 215 and a right eye representation 220, respectively. For example, the left eye representation 215 may be a crop of the left eye of the subject in the input image 100, and the right eye representation 220 may be a crop of the right eye of the subject in the input image 100. Similar to the feature extraction layer 230, each of the feature extraction layers 235 and 240 may be any feature learning portion of a CNN, constructed in any suitable manner and using any convolution kernel and pooling layer suitable for extracting features of the subject's eyes. The output of each feature extraction layer 235, 240 is a set of eye features, which may represent the gaze direction of the pupil of the input 215 and 220, respectively. Since the feature extraction layers 235 and 240 are each configured to analyze the eyes as input, for efficiency, their corresponding weight values ​​may be shared or the same. However, embodiments of the present disclosure contemplate any weight value for each of the feature extraction layers 235 and 240, whether shared or otherwise.

[0031] The fully connected layers 245 have the face mesh 225 as their single input. The fully connected layers 245 can be any classifier suitable for classifying an input image of a face into a location classification that indicates the location and / or position of the face of the object in the input image 100. For example, the fully connected layers 245 can be a multi-layer perceptron or any other fully connected layer of a CNN that is configured and trained to classify the input object into one of a plurality of discrete locations. Thus, the fully connected layers 245 output the likelihood of the location of the face in the input image 100.

[0032] The outputs of the feature extraction layers 230, 235, and 240 and the output of the fully connected layer 245 are connected together by the connection block 250 and input to the fully connected layer 255, which in turn outputs the gaze direction of the object in the input image 100. The fully connected layer 255 can be any classifier suitable for classifying the input features of the face of the object into a direction classification indicating the direction the object is looking at. For example, the fully connected layer 255 can be a multi-layer perceptron or any other fully connected layer of a CNN, which is configured and trained to classify the input facial and eye features and positions into one of a plurality of discrete gaze directions. The output classification of the fully connected layer 255 can be any representation of the gaze direction, such as a vector, a focal point on any predetermined imaginary plane, etc.

[0033] The glare representation 200 may be any isolated representation of glare in the input image 100, such as the glare representation 120. That is, the glare representation 200 may be any input that conveys the location and magnitude of any glare present in the input image 100. As one example, the glare representation 200 may be a binary mask of glare points extracted from the input image 100, i.e., an image containing only those pixels in the input image 100 that are determined to represent glare. Thus, each pixel of the binary mask is a black pixel (e.g., a pixel with a value of 0 in the binary representation) except for those pixels whose corresponding pixels in the input image 100 are determined to be glare pixels. As another example, the glare representation 200 may be a vector of the locations of those pixels in the input image 100 that are determined to contain glare.

[0034] The facial representation 205 may be any isolated representation of a face in the input image 100. That is, the facial representation 205 may be any input that conveys only the face of a subject present in the input image 100. For example, the facial representation 205 may be a binary mask of facial pixels extracted from the input image 100, wherein the facial pixels, or pixels corresponding to those pixels of the input image 100 determined to represent the face of the subject, are all one color (e.g., have one value), while all other pixels are black pixels (e.g., have a different value). As another example, the facial representation 200 may be a vector of the positions of those pixels in the input image 100 determined to represent the face of the subject.

[0035] The left eye representation 215 and the right eye representation 220 may each be any representation of a corresponding eye of a subject in the input image 100 that may indicate a gaze direction of the corresponding eye. For example, the left eye representation 215 may be a cropped portion of the input image 100 containing only the left eye of the subject, and the right eye representation 220 may be a cropped portion of the input image 100 containing only the right eye of the subject.

[0036] The face grid 225 may be any input that conveys the location of a face within the input image 100. As an example, the face grid 225 may be a binary mask of the subject's face, similar to the face representation 205. The face grid 225 may be of any resolution, such as the same resolution as the input image 100, or a coarser representation, such as a 25×25 square image binary mask, where each box corresponding to a face in the input image 100 is represented using one or more non-black colors, while the remaining boxes are black. The face grid 225 may be of any other resolution, such as 12×12, etc. The faces represented in the face grid 225 may be scaled to any particular size to aid in computation, or may be maintained at the same scale as the input image 100. Furthermore, the face grid 225 may include any information that conveys the spatial location of the subject's face in the input image 100, whether image or otherwise. For example, the face grid 225 may be a set of landmark points of the subject's face determined in any manner.

[0037] Figure 3 is a block diagram representation of an exemplary gaze determination system of an embodiment of the present disclosure. Here, computing device 300 may be any electronic computing device containing processing circuitry capable of performing gaze determination operations of embodiments of the present disclosure, in electronic communication with both camera 310 and gaze assistance system 320. In operation, camera 310, which may correspond to the following Figure 4A and Figure 4C The cabin camera 441 captures the image of the object and transmits it to the computing device 300, which then implements, for example Figure 2 The machine learning model is used to determine the image from the camera 310. Figure 2 The computing device 300 receives the input shown and calculates the output gaze direction of the object. The computing device 300 transmits this gaze direction to the gaze assistance system 320, and the gaze assistance system 320 takes an action or performs one or more operations in response. The computing device 300 can be any one or more electronic computing devices suitable for implementing the machine learning model of the embodiments of the present disclosure, such as the computing device 500 described in more detail below.

[0038] The gaze assist system 320 can be any system capable of performing one or more actions based on the gaze direction it receives from the computing device 300. Any configuration of the camera 310, the computing device 300, and the gaze assist system 320 can be envisioned. As an example, the gaze assist system 320 can be an autonomous vehicle capable of determining the gaze direction of a driver or another passenger and reacting to it, such as the autonomous vehicle 400 described in more detail below. In this example, the camera 310 and the computing device 300 can be located in the vehicle, and the gaze assist system 320 can represent the vehicle itself. The camera 310 can be positioned in any position in the vehicle that allows it to see the driver or passenger. Therefore, the camera 310 can capture images of the driver or passenger and transmit them to the computing device 300, which calculates the inputs 200, 205, 215, 220, and 225 and determines the gaze direction of the driver. The gaze direction can then be transmitted to another software module that determines, for example, the action that the vehicle can take in response. For example, the vehicle can determine that the gaze direction represents a distracted driver or a driver who is not paying attention to the road, and can initiate any type of operation in response. Such actions may include any type of warning to the driver (e.g., visual or audible warnings, warnings on a heads-up display, etc.), autonomous driving initiation, braking or turning maneuvers, or any other actions. The computing device 300 may be located within the vehicle of the gaze assist system 320 as a local processor, or may be a remote processor of the vehicle that receives images from the camera 310 and wirelessly transmits gaze directions or instructions to the gaze assist system 320.

[0039] As another example, the gaze assist system 320 can be a virtual reality or augmented reality system that can display images in response to the user's movement and gaze. In this example, the gaze assist system 320 includes a virtual reality or augmented reality display, such as a headset worn by the user and configured to project images to it. The camera 310 and the computing device 300 can be located in the headset, and the camera 310 captures images of the user's eyes and the computing device 300 determines his or her gaze direction. The gaze direction can then be transmitted to the virtual reality or augmented reality display, which can perform any action in response. For example, in order to save computing resources, the gaze assist system 320 can only render those virtual reality or augmented reality elements within the user's field of view determined using a determined gaze direction. Similarly, the gaze assist system 320 can remind the user to pay attention to objects or events that are determined to be outside the user's field of view but that the user may want to avoid or may be interested in. As with the autonomous vehicle example above, the computing device 300 of the virtual reality or augmented reality system may be located within the system 320, e.g., within the headset itself, or may be remotely located such that images are wirelessly transmitted to the computing device 300 and the calculated gaze direction may be wirelessly returned to the headset, which in turn may perform various actions in response.

[0040] As another example, the gaze assist system 320 can be a computer-based advertising system that determines the visual stimulus that the user is watching, such as but not limited to advertisements, warnings, objects, people, or other visible areas or points of interest. More specifically, the gaze assist system can be any electronic computing system or device, such as a desktop computer, a laptop computer, a smart phone, a server computer, etc. The camera 310 and the computing device 300 can be incorporated into the computing device to point to the user, such as entering or approaching the display of the computing device. The camera 310 can capture the image of the user and the computing device 300 can determine his or her gaze direction. The determined gaze direction can then be transmitted to the gaze assist system 320, such as a computing device, a remote computing device, etc. that displays advertisements for the user. The computing device can then use the calculated gaze direction to determine the subject of the user's attention, providing information about the effectiveness of various advertisements, warnings, or other visual stimuli.

[0041] Figure 4A 4 is a diagram of an example autonomous vehicle 400 according to some embodiments of the present disclosure. Autonomous vehicle 400 (also referred to herein as "vehicle 400") may include, but is not limited to, a passenger vehicle, such as a car, a truck, a bus, an emergency vehicle, a shuttle, an electric or motorized bicycle, a motorcycle, a fire truck, a police car, an ambulance, a boat, an engineering vehicle, an underwater vehicle, a drone, and / or other types of vehicles (e.g., unmanned and / or accommodating one or more passengers). Autonomous vehicles are generally described according to the automation levels defined by the National Highway Traffic Safety Administration (NHTSA), a division of the U.S. Department of Transportation, and the Society of Automotive Engineers (SAE) "Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles" (Standard No. J3016-201806 issued on June 15, 2018, Standard No. J3016-201609 issued on September 30, 2016, and previous and future versions of the standard). Vehicle 400 may be capable of implementing one or more functions that meet the autonomous driving level 3-5. For example, depending on the embodiment, vehicle 400 may be capable of conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5).

[0042] The vehicle 400 may include components such as a chassis, a body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other components of the vehicle. The 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 type of propulsion system. The propulsion system 450 may be connected to a drive train of the vehicle 400, which may include a transmission, to achieve propulsion of the vehicle 400. The propulsion system 450 may be controlled in response to receiving a signal from a throttle / accelerator 452.

[0043] A steering system 454, which may include a steering wheel, may be used to steer vehicle 400 (e.g., along a desired path or route) when propulsion system 450 is operating (e.g., when the vehicle is in motion). Steering system 454 may receive signals from steering actuator 456. For fully automated (Level 5) functionality, a steering wheel may be optional.

[0044] Brake sensor system 446 may be used to operate vehicle brakes in response to receiving signals from brake actuator 448 and / or brake sensors.

[0045] The system may include one or more CPUs, one or more system on chip (SoC) 404 ( Figure 4C ) and / or one or more controllers 436 of one or more GPUs may provide signals (e.g., representing commands) to one or more components and / or systems of the vehicle 400. For example, the one or more controllers may send signals to operate vehicle brakes via one or more brake actuators 448, to operate a steering system 454 via one or more steering actuators 456, and / or to operate a propulsion system 450 via one or more throttles / accelerators 452. The one or more controllers 436 may include one or more onboard (e.g., integrated) computing devices (e.g., supercomputers) that process sensor signals and output operating commands (e.g., signals representing commands) to enable autonomous driving and / or assist a human driver in driving the vehicle 400. The one or more 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 handle two or more of the above functions, two or more controllers 436 may handle a single function, and / or any combination thereof.

[0046] The one or more controllers 436 may provide signals for controlling one or more components and / or systems of the vehicle 400 in response to sensor data (e.g., sensor input) received from one or more sensors. The sensor data may be received from, for example and without limitation, a global navigation satellite system sensor 458 (e.g., a global positioning system sensor), a RADAR sensor 460, an ultrasonic sensor 462, a LIDAR sensor 464, an inertial measurement unit (IMU) sensor 466 (e.g., an accelerometer, a gyroscope, a magnetic compass, a magnetometer, etc.), a microphone 496, a stereo camera 468, a wide angle camera 470 (e.g., a fisheye camera), an infrared camera 472, a surround camera 474 (e.g., a 360 degree camera), a long-range and / or mid-range camera 498, a speed sensor 444 (e.g., for measuring the velocity of the vehicle 400), a vibration sensor 442, a steering sensor 440, a brake sensor (e.g., as part of a brake sensor system 446), and / or other sensor types.

[0047] One or more of the controllers 436 may receive inputs (e.g., represented by input data) from the instrument cluster 432 of the vehicle 400 and provide outputs (e.g., represented by output data, display data, etc.) via a human machine interface (HMI) display 434, an audible annunciator, a speaker, and / or via other components of the vehicle 400. These outputs may include information such as vehicle speed, velocity, time, map data (e.g., Figure 4C The HMI display 434 may include information such as the HD map 422 of the vehicle 400, location data (e.g., the location of the vehicle 400 on the map), directions, locations of other vehicles (e.g., occupancy grid), information about objects and states of objects as sensed by the controller 436, etc. For example, the HMI display 434 may display information about the presence of one or more objects (e.g., street signs, warning signs, traffic light changes, etc.) and / or information about driving maneuvers that the vehicle has made, is making, or will make (e.g., changing lanes now, leaving 34B in two miles, etc.).

[0048] The vehicle 400 further includes a network interface 424 that can communicate via one or more networks using one or more wireless antennas 426 and / or a modem. For example, the network interface 424 may be capable of communicating via LTE, WCDMA, UMTS, GSM, CDMA2000, etc. The one or more wireless antennas 426 may also enable communication between objects in the environment (e.g., vehicles, mobile devices, etc.) using one or more local area networks such as Bluetooth, Bluetooth LE, Z-wave, ZigBee, etc. and / or one or more low power wide area networks (LPWANs) such as LoRaWAN, SigFox, etc.

[0049] Figure 4BFor use according to some embodiments of the present disclosure Figure 4A An example of camera positions and fields of view for an example autonomous vehicle 400 is shown. The cameras and respective fields of view are an example embodiment and are not intended to be limiting. For example, additional and / or replaceable cameras may be included, and / or the cameras may be located at different locations on the vehicle 400.

[0050] The camera type for the camera may include, but is not limited to, a digital camera that may be suitable for use with components and / or systems of the vehicle 400. The camera may operate at an automotive safety integrity level (ASIL) B and / or at another ASIL. The camera type may have any image capture rate, such as 60 frames per second (fps), 120fps, 240fps, 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-white-white-white (RCCC) color filter array, a red-white-white-blue (RCCB) color filter array, a red-blue-green-white (RBGC) color filter array, a Foveon X3 color filter array, a Bayer sensor (RGGB) color filter array, a monochrome sensor color filter array, and / or another type of color filter array. In some embodiments, a clear pixel camera such as a camera with an RCCC, RCCB, and / or RBGC color filter array may be used in an effort to improve light sensitivity.

[0051] In some examples, one or more of the cameras can be used to perform advanced driver assistance system (ADAS) functions (e.g., as part of a redundant or fail-safe design). For example, a multi-function monocular camera can be installed to provide functions including lane departure warning, traffic sign assistance, and intelligent headlight control. One or more of the cameras (e.g., all of the cameras) can simultaneously record and provide image data (e.g., video).

[0052] One or more of the cameras may be mounted in a mounting assembly such as a custom designed (3-D printed) assembly to cut off stray light and reflections from within the car (e.g., reflections from the dashboard reflected in the windshield mirror) that may interfere with the camera's image data capture capabilities. With respect to the wing mirror mounting assembly, the wing mirror assembly may be custom 3-D printed so that the camera mounting plate matches the shape of the wing mirror. In some examples, one or more cameras may be integrated into the wing mirror. For side view cameras, one or more cameras may also be integrated into the four pillars at each corner of the cab.

[0053] A camera (e.g., a front-facing camera) having a field of view that includes a portion of the environment in front of the vehicle 400 can be used for surround view to help identify the forward path and obstacles, as well as assist in providing information critical to generating an occupancy grid and / or determining a preferred vehicle path with the help of one or more controllers 436 and / or control SoCs. The front-facing camera can be used to perform many of the same ADAS functions as LIDAR, including emergency braking, pedestrian detection, and collision avoidance. The front-facing camera can also be used for ADAS functions and systems, including lane departure warning ("LDW"), autonomous cruise control ("ACC"), and / or other functions such as traffic sign recognition.

[0054] A variety of cameras may be used in the front-facing configuration, including, for example, a monocular camera platform including a CMOS (complementary metal oxide semiconductor) color imager. Another example may be a wide-angle camera 470, which may be used to sense objects (e.g., pedestrians, intersection traffic, or bicycles) entering the field of view from the periphery. Figure 4B Only one wide-angle camera is shown in the figure, but there may be any number of wide-angle cameras 470 on the vehicle 400. In addition, long-range cameras 498 (e.g., a long-view stereo camera pair) may be used for depth-based object detection, especially for objects for which a neural network has not been trained. Long-range cameras 498 may also be used for object detection and classification and basic object tracking.

[0055] One or more stereo cameras 468 may also be included in the front configuration. The stereo camera 468 may include an integrated control unit including an expandable processing unit that may provide a multi-core microprocessor and programmable logic (e.g., FPGA) with an integrated CAN or Ethernet interface on a single chip. Such a unit may be used to generate a 3-D map of the vehicle environment, including distance estimates for all points in the image. An alternative stereo camera 468 may include a compact stereo vision sensor that may include two camera lenses (one on each side) and an image processing chip that may measure the distance from the vehicle to the target object and use the generated information (e.g., metadata) to activate autonomous emergency braking and lane departure warning functions. Other types of stereo cameras 468 may be used in addition to or alternatively to those described herein.

[0056] A camera (e.g., a side view camera) having a field of view of a portion of the environment including the sides of the vehicle 400 can be used for surround viewing, providing information used to create and update the occupancy grid and generate side impact collision warnings. Figure 4BThe four surround cameras 474 shown in FIG. 4 can be placed around the vehicle 400. The surround cameras 474 can include wide-angle cameras 470, fisheye cameras, 360-degree cameras, and / or the like. For example, four fisheye cameras can be placed in front, behind, and on the sides of the vehicle. In an alternative arrangement, the vehicle can use three surround cameras 474 (e.g., left, right, and rear), and can utilize one or more other cameras (e.g., a forward-facing camera) as a fourth surround camera.

[0057] A camera having a field of view that includes a portion of the environment behind the vehicle 400 (e.g., a rear view camera) can be used to assist with parking, surround view, rear collision warning, and creating and updating an occupancy grid. A variety of cameras can be used, including but not limited to cameras that are also suitable as front-facing cameras as described herein (e.g., long-range and / or mid-range cameras 498, stereo cameras 468, infrared cameras 472, etc.).

[0058] A camera having a field of view including a portion of the interior or cabin of the vehicle 400 may be used to monitor one or more states of a driver, passenger, or object in the cabin. Any type of camera may be used, including but not limited to one or more cabin cameras 441, which may be any type of camera described herein and may be placed anywhere on or in the vehicle 400 to provide a view of the cabin or its interior. For example, one or more cabin cameras 441 may be placed in or on portions of the vehicle 400 dashboard, rearview mirrors, sideview mirrors, seats, or doors, and may be oriented to capture images of any driver, passenger, or any other object or portion of the vehicle 400.

[0059] Figure 4C For use according to some embodiments of the present disclosure Figure 4A Block diagram of an example system architecture for an example autonomous vehicle 400. It should be understood that this arrangement and other arrangements described herein are set forth merely as examples. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, functional groupings, etc.) may be used in addition to or in place of those shown, and some elements may be omitted entirely. Further, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in combination with other components, and in any appropriate combination and location. The various functions described herein as being performed by an entity may be implemented by hardware, firmware, and / or software. For example, the various functions may be implemented by a processor executing instructions stored in a memory.

[0060] Figure 4CEach of the components, features, and systems of vehicle 400 is illustrated as being connected via bus 402. Bus 402 may include a controller area network (CAN) data interface (alternatively, referred to herein as a "CAN bus"). CAN may be a network inside vehicle 400 that assists in controlling various features and functions of vehicle 400, such as actuation of brakes, acceleration, braking, steering, windshield wipers, and the like. The CAN bus may be configured to have tens or even hundreds of nodes, each with its own unique identifier (e.g., CAN ID). The CAN bus may be read to find steering wheel angle, ground speed, engine speed per minute (RPM), button position, and / or other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0061] Although bus 402 is described as a CAN bus here, this is not intended to be limiting. For example, in addition to or alternatively to the CAN bus, FlexRay and / or Ethernet can be used. In addition, although bus 402 is represented by a single line, this is not intended to be limiting. For example, there can be any number of buses 402, which may include one or more CAN buses, one or more FlexRay buses, one or more Ethernet buses, and / or one or more other types of buses using different protocols. In some examples, two or more buses 402 can be used to perform different functions, and / or can be used for redundancy. For example, a first bus 402 can be used for a collision avoidance function, and a second bus 402 can be used for drive control. In any example, each bus 402 can communicate with any component of the vehicle 400, and two or more buses 402 can communicate with the same component. In some examples, each SoC 404, each controller 436, and / or each computer in the vehicle can have access to the same input data (e.g., input from a sensor of the vehicle 400), and can be connected to a common bus such as a CAN bus.

[0062] The vehicle 400 may include one or more controllers 436, such as those described herein. Figure 4A Those controllers described. Controller 436 can be used for a variety of functions. Controller 436 can be coupled to any other different components and systems of vehicle 400, and can be used for control of vehicle 400, artificial intelligence of vehicle 400, infotainment for vehicle 400, and / or the like.

[0063] The vehicle 400 may include one or more system on chip (SoC) 404. The SoC 404 may include a CPU 406, a GPU 408, a processor 410, a cache 412, an accelerator 414, a data store 415, and / or other components and features not shown. The SoC 404 may be used to control the vehicle 400 in a variety of platforms and systems. For example, one or more SoCs 404 may be combined with an HD map 422 in a system (e.g., a system of the vehicle 400), and the HD map may be downloaded from one or more servers (e.g., a server) via a network interface 424. Figure 4D one or more servers 478) to obtain map refreshes and / or updates.

[0064] CPU 406 may include a CPU cluster or CPU complex (alternatively, referred to herein as "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 of which has a dedicated L2 cache (e.g., 2MB L2 cache). CPU 406 (e.g., CCPLEX) may be configured to support simultaneous cluster operations, so that any combination of clusters of CPU 406 can be active at any given time.

[0065] CPU 406 may implement power management capabilities including one or more of the following features: each hardware block may be automatically clock gated when idle to save dynamic power; each core clock may be gated when the core is not actively executing instructions due to the execution of WFI / WFE instructions; each core may be independently power gated; each core cluster may be independently clock gated when all cores are clock gated or power gated; and / or each core cluster may be independently power gated when all cores are power gated. CPU 406 may further implement an enhanced algorithm for managing power states, in which allowed power states and expected wake-up times are specified, and hardware / microcode determines the optimal power state to enter for the core, cluster, and CCPLEX. The processing core may support a simplified power state entry sequence in software, with the work being offloaded to the microcode.

[0066] The GPU 408 may include an integrated GPU (alternatively, referred to herein as an "iGPU"). The GPU 408 may be programmable and efficient for parallel workloads. In some examples, the GPU 408 may use an enhanced tensor instruction set. The GPU 408 may include one or more streaming microprocessors, each of which may include an L1 cache (e.g., an L1 cache with at least 96KB storage capacity), and two or more of these streaming microprocessors may share an L2 cache (e.g., an L2 cache with 512KB storage capacity). In some embodiments, the GPU 408 may include at least eight streaming microprocessors. The GPU 408 may use a computer-based application programming interface (API). In addition, the GPU 408 may use one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0067] In the case of automotive and embedded use, GPU 408 can be power optimized to achieve optimal performance. For example, GPU 408 can be manufactured on fin field effect transistors (FinFETs). However, this is not intended to be limiting, and GPU 408 can be manufactured using other semiconductor manufacturing processes. Each streaming microprocessor can merge several mixed precision processing cores divided into multiple blocks. For example and without limitation, 64 PF32 cores and 32 PF64 cores can be divided into four processing blocks. In such an example, each processing block can be allocated 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two mixed precision NVIDIA tensor cores for deep learning matrix arithmetic, L0 instruction cache, thread beam (warp) scheduler, dispatch unit and / or 64KB register file. In addition, the streaming microprocessor can include independent parallel integer and floating point data paths to provide efficient execution of workloads using a mix of calculations and addressing calculations. The streaming microprocessor may include independent thread scheduling capabilities to allow for finer-grained synchronization and collaboration between parallel threads. A streaming microprocessor may include a combined L1 data cache and shared memory unit to increase performance while simplifying programming.

[0068] GPU 408 can include high bandwidth memory (HBM) and / or a 16GB HBM2 memory subsystem that provides a peak memory bandwidth of approximately 900 GB / s in some examples. In some examples, synchronous graphics random access memory (SGRAM), such as fifth generation graphics double data rate synchronous random access memory (GDDR5), can be used in addition to or in lieu of HBM memory.

[0069] The GPU 408 may include unified memory technology that includes access counters to allow memory pages to be more accurately migrated to the processor that accesses them most frequently, thereby improving the efficiency of memory ranges shared between processors. In some examples, address translation service (ATS) support may be used to allow the GPU 408 to directly access the CPU 406 page table. In such an example, when the GPU 408 memory management unit (MMU) experiences a miss, an address translation request may be transmitted to the CPU 406. In response, the CPU 406 may look up the virtual-physical mapping for the address in its page table and transmit the translation back to the GPU 408. In this way, the unified memory technology may allow a single unified virtual address space to be used for memory of both the CPU 406 and the GPU 408, thereby simplifying GPU 408 programming and porting applications to the GPU 408.

[0070] In addition, GPU 408 may include access counters that can track how often GPU 408 accesses the memory of other processors. The access counters can help ensure that memory pages are moved to the physical memory of the processor that accesses them most frequently.

[0071] SoC 404 may include any number of caches 412, including those described herein. For example, cache 412 may include an L3 cache available to both CPU 406 and GPU 408 (e.g., connected to both CPU 406 and GPU 408). Cache 412 may include a write-back cache that may track the state of a line, for example, by using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). Depending on the embodiment, the L3 cache may include 4MB or more, although smaller cache sizes may also be used.

[0072] The SoC 404 may include an arithmetic logic unit (ALU) that may be used to perform processing for any of a variety of tasks or operations with respect to the vehicle 400 - such as processing a DNN. In addition, the SoC 404 may include a floating point unit (FPU) - or other math coprocessor or digital coprocessor type - for performing mathematical operations within the system. For example, the SoC 104 may include one or more FPUs integrated as execution units within the CPU 406 and / or GPU 408.

[0073] SoC 404 may include one or more accelerators 414 (e.g., hardware accelerators, software accelerators, or combinations thereof). For example, SoC 404 may include a hardware acceleration cluster, which may include optimized hardware accelerators and / or large on-chip memory. The large on-chip memory (e.g., 4MB SRAM) may enable the hardware acceleration cluster to accelerate neural networks and other calculations. The hardware acceleration cluster may be used to supplement GPU 408 and offload some tasks of GPU 408 (e.g., free up more cycles of GPU 408 for performing other tasks). As an example, accelerator 414 may be used for targeted workloads (e.g., perception, convolutional neural networks (CNNs), etc.) that are sufficiently stable to be easily controlled for acceleration. When used herein, the term "CNN" may include all types of CNNs, including region-based or regional convolutional neural networks (RCNNs) and fast RCNNs (e.g., for object detection).

[0074] Accelerator 414 (e.g., hardware acceleration cluster) may include a deep learning accelerator (DLA). DLA may include one or more tensor processing units (TPUs) that may be configured to provide an additional 10 trillion operations per second for deep learning applications and reasoning. TPU may be an accelerator configured to perform image processing functions (e.g., for CNN, RCNN, etc.) and optimized for performing image processing functions. DLA may be further optimized for a specific set of neural network types and floating point operations and reasoning. The design of DLA may provide higher performance per millimeter than a general-purpose GPU, and far exceeds the performance of a CPU. TPU may perform several functions, including a single instance convolution function, support, for example, INT8, INT16, and FP16 data types for both features and weights, and post-processor functions.

[0075] The DLA can quickly and efficiently execute neural networks, particularly CNNs, on processed or unprocessed data for any of a wide variety of functions, such as, but not limited to: a CNN for object recognition and detection using data from a camera sensor; a CNN for distance estimation using data from a camera sensor; a CNN for emergency vehicle detection and identification and detection using data from a microphone; a CNN for facial recognition and vehicle owner identification using data from a camera sensor; and / or a CNN for safety and / or security-related events.

[0076] The DLA can perform any function of the GPU 408, and by using an inference accelerator, for example, the designer can target any function to either the DLA or the GPU 408. For example, the designer can focus the processing of CNNs and floating point operations on the DLA, and leave other functions to the GPU 408 and / or other accelerators 414.

[0077] The accelerator 414 (e.g., a hardware acceleration cluster) may include a programmable vision accelerator (PVA), which may be alternatively referred to herein as a computer vision accelerator. The PVA may be designed and configured to accelerate computer vision algorithms for advanced driver assistance systems (ADAS), autonomous driving, and / or augmented reality (AR) and / or virtual reality (VR) applications. The PVA may provide a balance between performance and flexibility. For example, each PVA may include, for example and without limitation, any number of reduced instruction set computer (RISC) cores, direct memory access (DMA), and / or any number of vector processors.

[0078] The RISC core can interact with an image sensor (e.g., an image sensor of any camera described herein), an image signal processor, and / or the like. Each of these RISC cores can include any number of memories. Depending on the embodiment, the RISC core can use any of a number of protocols. In some examples, the RISC core can execute a real-time operating system (RTOS). The RISC core can be implemented using one or more integrated circuit devices, application specific integrated circuits (ASICs), and / or storage devices. For example, the RISC core can include an instruction cache and / or a tightly coupled RAM.

[0079] The DMA may enable components of the PVA to access system memory independently of the CPU 406. The DMA may support any number of features used to provide optimizations for the PVA, including, but not limited to, support for multi-dimensional addressing and / or circular addressing. In some examples, the DMA may support addressing in up to six or more dimensions, which may include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0080] The vector processor can be a programmable processor that can be designed to efficiently and flexibly perform programming for computer vision algorithms and provide signal processing capabilities. In some examples, the PVA can include a PVA core and two vector processing subsystem partitions. The PVA core can include a processor subsystem, one or more DMA engines (e.g., two DMA engines), and / or other peripherals. The vector processing subsystem can operate as the main processing engine of the PVA, and can include a vector processing unit (VPU), an instruction cache, and / or a vector memory (e.g., VMEM). The VPU core can include a digital signal processor, such as, for example, a single instruction multiple data (SIMD), a very long instruction word (VLIW) digital signal processor. The combination of SIMD and VLIW can enhance throughput and rate.

[0081] Each of the vector processors may include an instruction cache and may be coupled to a 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 employ data parallelism. For example, in some embodiments, multiple vector processors included in a single PVA may execute the same computer vision algorithm, but 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 sequence images or portions of an image. Among other things, any number of PVAs may be included in a hardware acceleration cluster, and any number of vector processors may be included in each of these PVAs. In addition, the PVA may include additional error correction code (ECC) memory to enhance overall system security.

[0082] The accelerator 414 (e.g., a hardware acceleration cluster) may include an on-chip computer vision network and SRAM to provide high bandwidth, low latency SRAM for the accelerator 414. In some examples, the on-chip memory may include at least 4MB of SRAM consisting of, for example and not limited to, eight field-configurable memory blocks that can be accessed by both the PVA and the DLA. Each pair of memory blocks may include an advanced peripheral bus (APB) interface, configuration circuitry, a controller, and a multiplexer. Any type of memory may be used. The PVA and DLA may access the memory via a backbone that provides high-speed memory access to the PVA and DLA. The backbone may include an on-chip computer vision network that interconnects the PVA and DLA to the memory (e.g., using APB).

[0083] The on-chip computer vision network may include an interface that determines that both the PVA and DLA provide ready and valid signals before transmitting any control signals / addresses / data. Such an interface may provide separate phases and separate channels for transmitting control signals / addresses / data, as well as burst communication for continuous data transmission. This type of interface may comply with ISO 26262 or IEC 61508 standards, but other standards and protocols may also be used.

[0084] In some examples, SoC 404 may include a real-time ray tracing hardware accelerator such as described in U.S. Patent Application No. 16 / 101,232 filed on August 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the position and range of objects (e.g., within a world model) in order to generate real-time visualization simulations for RADAR signal interpretation, for sound propagation synthesis and / or analysis, for SONAR system simulations, for general wave propagation simulations, for comparison with LIDAR data for the purpose of positioning and / or other functions, and / or for other purposes. In some embodiments, one or more tree traversal units (TTUs) may be used to perform one or more ray tracing related operations.

[0085] The accelerator 414 (e.g., a hardware accelerator cluster) has a wide range of uses in autonomous driving. The PVA can be a programmable visual accelerator that can be used in key processing stages in ADAS and autonomous vehicles. The capabilities of the PVA are a good match for algorithmic domains that require predictable processing, low power, and low latency. In other words, the PVA performs well on semi-dense or dense rule computations, and even on small data sets that require predictable runtimes with low latency and low power. Therefore, in the context of a platform for autonomous vehicles, the PVA is designed to run classic computer vision algorithms because they are efficient at object detection and integer math.

[0086] For example, according to one embodiment of the technology, PVA is used to perform computer stereo vision. In some examples, algorithms based on semi-global matching can be used, but this is not intended to be limiting. Many applications for level 3-5 autonomous driving require instant motion estimation / stereo matching (e.g., structure from motion, pedestrian recognition, lane detection, etc.). PVA can perform computer stereo vision functions on input from two monocular cameras.

[0087] In some examples, PVA can be used to perform dense optical flow. For example, PVA can be used to process raw RADAR data (e.g., using a 4D fast Fourier transform) to provide a processed RADAR signal before transmitting the next RADAR pulse. In other examples, PVA is used for time-of-flight depth processing, for example by processing raw time-of-flight data to provide processed time-of-flight data.

[0088] DLA can be used to run any type of network to enhance control and driving safety, including, for example, a neural network that outputs a confidence measure for each object detection. Such a confidence value can be interpreted as a probability, or as providing a relative "weight" of each detection compared to other detections. This confidence value enables the system to make further decisions about which detections should be considered true positive detections rather than false positive detections. For example, the system can set a threshold for confidence and only consider detections that exceed the threshold as true positive detections. In an automatic emergency braking (AEB) system, a false positive detection will cause the vehicle to automatically perform emergency braking, which is obviously undesirable. Therefore, only the most confident detection should be considered a trigger for AEB. DLA can run a neural network for regressing confidence values. The neural network can take at least some subset of parameters as its input, such as bounding box dimensions, ground plane estimates obtained (e.g., from another subsystem), inertial measurement unit (IMU) sensor 466 output related to vehicle 400 orientation and distance, 3D position estimates of objects obtained from neural networks and / or other sensors (e.g., LIDAR sensor 464 or RADAR sensor 460), etc.

[0089] SoC 404 may include one or more data stores 416 (e.g., memory). Data store 416 may be on-chip memory of SoC 404 that may store neural networks to be executed on the GPU and / or DLA. In some examples, data store 416 may be large enough to store multiple instances of a neural network for redundancy and safety. Data store 412 may include an L2 or L3 cache 412. References to data store 416 may include references to memory associated with a PVA, DLA, and / or other accelerator 414 as described herein.

[0090] SoC 404 may include one or more processors 410 (e.g., embedded processors). Processor 410 may include a boot and power management processor, which may be a dedicated processor and subsystem for handling boot power and management functions and related safety implementations. The boot and power management processor may be part of the SoC 404 boot sequence and may provide runtime power management services. The boot power and management processor may provide clock and voltage programming, auxiliary system low power state transitions, SoC 404 thermal and temperature sensor management, and / or SoC 404 power state management. Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to the temperature, and SoC 404 may use the ring oscillator to detect the temperature of CPU 406, GPU 408, and / or accelerator 414. If it is determined that the temperature exceeds a threshold, the boot and power management processor may enter a temperature fault routine and place SoC 404 in a lower power state and / or place vehicle 400 in a driver safety parking mode (e.g., parking vehicle 400 safely).

[0091] Processor 410 may further include a set of embedded processors that may be used as an audio processing engine. The audio processing engine may be an audio subsystem that allows for full hardware support for multi-channel audio through multiple interfaces and a wide range of flexible audio I / O interfaces. In some examples, the audio processing engine is a dedicated processor core having a digital signal processor with dedicated RAM.

[0092] The processor 410 may further include an always-on processor engine that may provide the necessary hardware features to support low-power sensor management and wake-up use cases. The always-on processor engine may include a processor core, tightly coupled RAM, supporting peripherals (e.g., timers and interrupt controllers), various I / O controller peripherals, and routing logic.

[0093] The processor 410 may further include a safety cluster engine, which includes a dedicated processor subsystem that handles safety management of automotive applications. The safety cluster engine may include two or more processor cores, tightly coupled RAM, supporting peripherals (e.g., timers, interrupt controllers, etc.), and / or routing logic. In safety mode, the two or more cores may operate in lockstep mode and act as a single core with comparison logic to detect any differences between their operations.

[0094] Processor 410 may further include a real-time camera engine, which may include a dedicated processor subsystem for handling real-time camera management.

[0095] 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.

[0096] The processor 410 may include a video image compositer, which may be a processing block (e.g., implemented on a microprocessor) that implements video post-processing functions required for the video playback application to generate a final image for the player window. The video image compositer may perform lens distortion correction for the wide-angle camera 470, the surround camera 474, and / or for an in-cab surveillance camera sensor. The in-cab surveillance camera sensor is preferably monitored by a neural network running on another instance of the advanced SoC, configured to recognize in-cab events and respond accordingly. The in-cab system may perform lip reading to activate mobile phone service and place a call, dictate an email, change a vehicle destination, activate or change the vehicle's infotainment system and settings, or provide voice-activated web surfing. Certain functions are only available to the driver when the vehicle is operating in autonomous mode and are disabled in other circumstances.

[0097] The video image compositer may include enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, in the case of motion in the video, the noise reduction appropriately weights the spatial information and reduces the weight of information provided by adjacent frames. In the case where an image or portion of an image does not include motion, the temporal noise reduction performed by the video image compositer may use information from previous images to reduce noise in the current image.

[0098] The video image compositor may also be configured to perform stereoscopic rectification on the input stereoscopic footage 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 powered on and active, actively performing 3D rendering, the video image compositor may be used to offload the GPU 408 to improve performance and responsiveness.

[0099] SoC 404 may further include a mobile industry processor interface (MIPI) camera serial interface for receiving video and input from the camera, a high-speed interface, and / or a video input block that can be used for the camera and related pixel input functions. SoC 404 may further include an input / output controller that can be controlled by software and can be used to receive I / O signals that are not submitted to a specific role. SoC 404 may further include a wide range of peripheral device interfaces to enable communication with peripheral devices, audio codecs, power management, and / or other devices. SoC 404 can be used to process data from cameras (connected via Gigabit multimedia serial links and Ethernet), sensors (e.g., LIDAR sensors 464, RADAR sensors 460, etc. that can be connected via Ethernet), data from bus 402 (e.g., the speed of vehicle 400, steering wheel position, etc.), data from GNSS sensors 458 (connected via Ethernet or CAN bus). SoC 404 may further include a dedicated high-performance mass storage controller that may include its own DMA engine and that can be used to release CPU 406 from routine data management tasks.

[0100] SoC 404 can be an end-to-end platform with a flexible architecture that spans automation levels 3-5, providing a comprehensive functional safety architecture that leverages and efficiently uses computer vision and ADAS technologies to achieve diversity and redundancy, together with deep learning tools to provide a platform for a flexible and reliable driving software stack. SoC 404 can be faster, more reliable, and even more energy efficient and space efficient than conventional systems. For example, when combined with CPU 406, GPU 408, and data storage 45, accelerator 414 can provide a fast and efficient platform for level 3-5 autonomous vehicles.

[0101] The technology thus provides capabilities and functionality that cannot be achieved by conventional systems. For example, computer vision algorithms can be executed on CPUs, which can be configured using high-level programming languages ​​such as the C programming language to execute a wide variety of processing algorithms across a wide variety of visual data. However, CPUs often fail to meet the performance requirements of many computer vision applications, such as those related to, for example, execution time and power consumption. In particular, many CPUs are unable to execute complex object detection algorithms in real time, which is a requirement for in-vehicle ADAS applications and a requirement for practical Level 3-5 autonomous vehicles.

[0102] In contrast to conventional systems, by providing a CPU complex, a GPU complex, and a hardware acceleration cluster, the techniques described herein allow multiple neural networks to be executed simultaneously and / or sequentially, and the results to be combined together to achieve Level 3-5 autonomous driving functions. For example, a CNN executed on a DLA or dGPU (e.g., GPU 420) may include text and word recognition, allowing the supercomputer to read and understand traffic signs, including signs for which a neural network has not been specifically trained. The DLA may further include a neural network capable of recognizing, interpreting, and providing semantic understanding of the signs, and passing that semantic understanding to a path planning module running on the CPU complex.

[0103] As another example, as required for Level 3, 4, or 5 driving, multiple neural networks may be running simultaneously. For example, a warning sign consisting of "Caution: Flashing Lights Indicate Icing Conditions" along with electric lights may be interpreted by several neural networks independently or collectively. The sign itself may be recognized as a traffic sign by a deployed first neural network (e.g., a trained neural network), and the text "Flashing Lights Indicate Icing Conditions" may be interpreted by a deployed second neural network that informs the vehicle's path planning software (preferably executing on a CPU complex) that icing conditions exist when the flashing lights are detected. The flashing lights may be identified by operating a deployed third neural network over multiple frames that informs the vehicle's path planning software of the presence (or absence) of the flashing lights. All three neural networks may be running simultaneously, for example, within the DLA and / or on the GPU 408.

[0104] In some examples, a CNN for facial recognition and owner recognition can use data from the camera sensor to identify the presence of an authorized driver and / or owner of the vehicle 400. The always-on sensor processing engine can be used to unlock the vehicle and turn on the lights when the owner approaches the driver's door, and in security mode, disable the vehicle when the owner leaves the vehicle. In this way, the SoC 404 provides security against theft and / or carjacking.

[0105] In another example, a CNN for emergency vehicle detection and identification can use data from microphone 496 to detect and identify emergency vehicle sirens. In contrast to conventional systems that use general classifiers to detect sirens and manually extract features, SoC 404 uses CNN to classify environmental and urban sounds and to classify visual data. In a preferred embodiment, the CNN running on the DLA is trained to identify the relative closing rate of emergency vehicles (e.g., by using the Doppler effect). The CNN can also be trained to identify emergency vehicles specific to the local area in which the vehicle is operating, as identified by the GNSS sensor 458. Thus, for example, when operating in Europe, the CNN will seek to detect European sirens, and when in the United States, the CNN will seek to identify only North American sirens. Once an emergency vehicle is detected, with the assistance of the ultrasonic sensor 462, the control program can be used to execute emergency vehicle safety routines to slow the vehicle, drive to the side of the road, stop the vehicle, and / or idle the vehicle until the emergency vehicle passes.

[0106] The vehicle may include a CPU 418 (e.g., a discrete CPU or dCPU) that 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, including, for example, arbitrating potentially inconsistent results between ADAS sensors and the SoC 404, and / or monitoring the status and health of the controller 436 and / or the infotainment SoC 430.

[0107] The vehicle 400 may include a GPU 420 (e.g., a discrete GPU or dGPU) that may be coupled to the SoC 404 via a high-speed interconnect (e.g., NVIDIA's NVLINK). The GPU 420 may provide additional artificial intelligence functionality, for example, by executing redundant and / or different neural networks, and may be used to train and / or update the neural network based on inputs (e.g., sensor data) from sensors of the vehicle 400.

[0108] The 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 cellular antennas, Bluetooth antennas, etc.). The network interface 424 can be used to enable wireless connections with the cloud (e.g., with a server 478 and / or other network devices), with other vehicles, and / or with computing devices (e.g., a passenger's client device) via the Internet. In order to communicate with other vehicles, a direct link can be established between the two vehicles, and / or an indirect link can be established (e.g., across a network and through the Internet). The direct link can be provided using a vehicle-to-vehicle communication link. The vehicle-to-vehicle communication link can provide the vehicle 400 with information about vehicles approaching the vehicle 400 (e.g., vehicles in front of, to the side of, and / or behind the vehicle 400). This functionality can be part of the cooperative adaptive cruise control functionality of the vehicle 400.

[0109] The network interface 424 may include a SoC that provides modulation and demodulation functions and enables the controller 436 to communicate over a wireless network. The network interface 424 may include a radio frequency front end for up-conversion from baseband to radio frequency and down-conversion from radio frequency to baseband. The frequency conversion may be performed by a known process and / or may be performed using a super-heterodyne process. In some examples, the radio frequency front end function may be provided by a separate chip. The network interface may include a wireless function for communicating via LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-wave, ZigBee, LoRaWAN, and / or other wireless protocols.

[0110] The vehicle 400 may further include a data store 428, which may include off-chip storage (e.g., outside the SoC 404). The data store 428 may include one or more storage elements, including RAM, SRAM, DRAM, VRAM, flash memory, a hard disk, and / or other components and / or devices that can store at least one bit of data.

[0111] The vehicle 400 may further include a GNSS sensor 458 (e.g., GPS and / or assisted GPS sensor) for assisting with mapping, perception, occupancy grid generation, and / or path planning functions. Any number of GNSS sensors 458 may be used, including, for example and without limitation, GPS using a USB connector with an Ethernet to serial (RS-232) bridge.

[0112] The vehicle 400 may further include a RADAR sensor 460. The RADAR sensor 460 may be used by the vehicle 400 for remote vehicle detection even in darkness and / or inclement weather conditions. The RADAR functional safety level may be ASIL B. The RADAR sensor 460 may use CAN and / or bus 402 (e.g., to transmit data generated by the RADAR sensor 460) for control and access to object tracking data, accessing Ethernet in some examples to access raw data. A variety of RADAR sensor types may be used. For example and without limitation, the RADAR sensor 460 may be suitable for front, rear, and side RADAR use. In some examples, a pulse Doppler RADAR sensor is used.

[0113] 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, short-range side coverage, and the like. In some examples, the long-range RADAR may be used for adaptive cruise control functions. The long-range RADAR system may provide a wide field of view (e.g., within a range of 250m) achieved by two or more independent scans. The RADAR sensor 460 may help distinguish between static and moving objects and may be used by the ADAS system for emergency braking assistance and forward collision warnings. The long-range RADAR sensor may include a single-station multimode RADAR with multiple (e.g., six or more) fixed RADAR antennas and high-speed CAN and FlexRay interfaces. In an example with six antennas, the central four antennas may create a focused beam pattern designed to record the surroundings of the vehicle 400 at a higher rate with minimal traffic interference from adjacent lanes. The other two antennas may expand the field of view, making it possible to quickly detect vehicles entering or leaving the lane of the vehicle 400.

[0114] As an example, a medium-range RADAR system may include a range of up to 460m (front) or 80m (rear) and a field of view of up to 42 degrees (front) or 450 degrees (rear). A short-range RADAR system may include, but is not limited to, a RADAR sensor designed to be mounted on both ends of a rear bumper. When mounted on both ends of a rear bumper, such a RADAR sensor system may create two beams that continuously monitor the blind spots behind and beside the vehicle.

[0115] Short-range RADAR systems can be used in ADAS systems for blind spot detection and / or lane change assistance.

[0116] The vehicle 400 may further include ultrasonic sensors 462. The ultrasonic sensors 462, which may be placed on the front, rear, and / or sides of the vehicle 400, may be used for parking assistance and / or creating and updating occupancy grids. 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). The ultrasonic sensors 462 may operate at a functional safety level of ASIL B.

[0117] 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 ASIL B for functional safety level. In some examples, the vehicle 400 may include multiple LIDAR sensors 464 (e.g., two, four, six, etc.) that may use Ethernet (e.g., to provide data to a Gigabit Ethernet switch).

[0118] In some examples, LIDAR sensor 464 may be able to provide a list of objects and their distances for a 360-degree field of view. Commercially available LIDAR sensor 464 may have, for example, an advertised range of approximately 100m, an accuracy of 2cm-3cm, and support for 100Mbps Ethernet connections. In some examples, one or more non-protruding LIDAR sensors 464 may be used. In such examples, LIDAR sensor 464 may be implemented as a small device that can be embedded in the front, back, side, and / or corner of vehicle 400. In such examples, LIDAR sensor 464 may provide a field of view of up to 120 degrees horizontally and 35 degrees vertically, with a range of 200m, even for low reflectivity objects. Front-mounted LIDAR sensor 464 may be configured for a horizontal field of view between 45 and 135 degrees.

[0119] In some examples, LIDAR technologies such as 3D flash LIDAR may also be used. 3D flash LIDAR uses a flash of laser as an emission source to illuminate the vehicle's surroundings up to about 200m. The flash LIDAR unit includes a receiver that records the laser pulse transmission time and the reflected light on each pixel, which in turn corresponds to the range from the vehicle to the object. Flash LIDAR can allow highly accurate and distortion-free images of the surrounding environment to be generated with each laser flash. In some examples, four flash LIDAR sensors can be deployed, one on each side of the vehicle 400. Available 3D flash LIDAR systems include solid-state 3D gaze array LIDAR cameras (e.g., non-scanning LIDAR devices) with no moving parts other than fans. The flash LIDAR device can use 5 nanosecond Class I (eye-safe) laser pulses per frame, and can capture the reflected laser light in the form of a 3D range point cloud and co-registered intensity data. By using flash LIDAR, and because flash LIDAR is a solid-state device with no moving components, the LIDAR sensor 464 may be less susceptible to motion blur, vibration, and / or shock.

[0120] 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, for example and without limitation, an accelerometer, a magnetometer, a gyroscope, a magnetic compass, and / or other sensor types. In some examples, for example, in a six-axis application, the IMU sensor 466 may include an accelerometer and a gyroscope, and in a nine-axis application, the IMU sensor 466 may include an accelerometer, a gyroscope, and a magnetometer.

[0121] In some embodiments, IMU sensor 466 can be implemented as a miniature high-performance GPS-assisted inertial navigation system (GPS / INS) that combines micro-electromechanical system (MEMS) inertial sensors, high-sensitivity GPS receivers, and advanced Kalman filter algorithms to provide estimates of position, velocity, and attitude. Thus, in some examples, IMU sensor 466 can enable vehicle 400 to estimate heading by directly observing and correlating velocity changes from GPS to IMU sensor 466 without input from a magnetic sensor. In some examples, IMU sensor 466 and GNSS sensor 458 can be combined into a single integrated unit.

[0122] The vehicle may include microphones 496 positioned in and / or around the vehicle 400. The microphones 496 may be used for, among other things, emergency vehicle detection and identification.

[0123] The vehicle may further include any number of camera types, including a stereo camera 468, a wide angle camera 470, an infrared camera 472, a surround camera 474, a long-range and / or mid-range camera 498, and / or other camera types. These cameras can be used to capture image data around the entire periphery of the vehicle 400. The type of camera used depends on the embodiment and the requirements of the vehicle 400, and any combination of camera types can be used to provide the necessary coverage around the vehicle 400. In addition, the number of cameras can vary depending on the embodiment. For example, the vehicle may include six cameras, seven cameras, ten cameras, twelve cameras, and / or another number of cameras. As an example and not limitation, the cameras can support Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet. Each of the cameras described herein may be capable of supporting Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet. Figure 4A and Figure 4B Described in more detail.

[0124] The vehicle 400 may further include a vibration sensor 442. The vibration sensor 442 may measure vibrations of components of the vehicle, such as an axle. 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 may be used to determine friction or slip of the road surface (e.g., when there is a vibration difference between a powered drive shaft and a free-spinning shaft).

[0125] The vehicle 400 may include an ADAS system 438. In some examples, the ADAS system 438 may include a 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.

[0126] The ACC system may use a RADAR sensor 460, a LIDAR sensor 464, and / or a camera. The ACC system may include a longitudinal ACC and / or a lateral ACC. The longitudinal ACC monitors and controls the distance to the vehicle immediately in front of the vehicle 400, and automatically adjusts the vehicle speed to maintain a safe distance from the vehicle in front. The lateral ACC performs distance keeping and suggests that the vehicle 400 change lanes when necessary. The lateral ACC is related to other ADAS applications such as LC and CWS.

[0127] CACC uses information from other vehicles, which can be received indirectly from other vehicles via a wireless link or through a network connection (e.g., through the Internet) via a network interface 424 and / or a wireless antenna 426. A direct link can be provided by a vehicle-to-vehicle (V2V) communication link, while an indirect link can be an infrastructure-to-vehicle (I2V) communication link. Typically, the V2V communication concept provides information about the vehicle immediately ahead (e.g., the vehicle immediately ahead of the vehicle 400 and in the same lane as it), while the I2V communication concept provides information about traffic farther ahead. The CACC system may include either or both of the I2V and V2V information sources. Given information about the vehicle ahead of the vehicle 400, CACC can be more reliable, and it is possible to improve the smoothness of traffic flow and reduce road congestion.

[0128] The FCW system is designed to alert the driver to hazards so that the driver can take corrective action. The FCW system uses a front 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. The FCW system can provide warnings in the form of, for example, sound, visual warnings, vibrations and / or rapid brake pulses.

[0129] The AEB system detects an impending front collision with another vehicle or other object and can automatically apply the brakes if the driver does not take corrective action within specified time or distance parameters. The AEB system can use a front 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, and if the driver does not take corrective action, the AEB system can automatically apply the brakes in an effort to prevent or at least mitigate the effects of the predicted collision. The AEB system may include technologies such as dynamic brake support and / or collision approach braking.

[0130] The LDW system provides visual, audible, and / or tactile warnings, such as steering wheel or seat vibrations, to alert the driver when the vehicle 400 crosses a lane marking. When the driver indicates an intention to leave the lane, by activating a turn signal, the LDW system is not activated. The LDW system may use a front-facing camera coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration assembly.

[0131] The LKA system is a variation of the LDW system. If the vehicle 400 begins to leave the lane, the LKA system provides steering input or braking to correct the vehicle 400. The BSW system detects and warns the driver of vehicles in the car's blind spot. The BSW system can provide visual, auditory and / or tactile alerts to indicate that it is unsafe to merge or change lanes. The system can provide additional warnings when the driver uses a turn signal. The BSW system can use a rear-facing camera and / or RADAR sensor 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.

[0132] The RCTW system may provide visual, audible, and / or tactile notifications when an object is detected outside the range of the rear camera while the vehicle 400 is in reverse. Some RCTW systems include AEB to ensure that the vehicle brakes are applied to avoid a crash. The RCTW system may use one or more rear RADAR sensors 460 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration assembly.

[0133] Conventional ADAS systems may be prone to false positive results, which may annoy and distract the driver, but are typically not catastrophic because the ADAS system alerts the driver and allows the driver to decide whether the safety condition really exists and take action accordingly. However, in the autonomous vehicle 400, in the case of conflicting results, the vehicle 400 itself must decide whether to pay attention to the results from the main computer or the auxiliary computer (e.g., the first controller 436 or the second controller 436). For example, in some embodiments, the ADAS system 438 can be a backup and / or auxiliary computer for providing perception information to the backup computer rationality module. The backup computer rationality monitor can run redundant and diverse software on hardware components to detect faults in perception and dynamic driving tasks. The output from the ADAS system 438 can be provided to the supervisory MCU. If the outputs from the main computer and the auxiliary computer conflict, the supervisory MCU must determine how to coordinate the conflict to ensure safe operation.

[0134] In some examples, the master computer can be configured to provide a confidence score to the supervisory MCU, indicating the master computer's confidence in the selected result. If the confidence score exceeds a threshold, the supervisory MCU can follow the direction of the master computer, regardless of whether the auxiliary computer provides conflicting or inconsistent results. In the case where the confidence score does not meet the threshold and the master computer and the auxiliary computer indicate different results (e.g., conflicts), the supervisory MCU can arbitrate between these computers to determine the appropriate result.

[0135] The supervisory MCU may be configured to run a neural network that is trained and configured to determine conditions under which the auxiliary computer provides a false alarm based on outputs from the primary computer and the auxiliary computer. Thus, the neural network in the supervisory MCU may learn when the output of the auxiliary computer may be trusted and when it may not. For example, when the auxiliary computer is a RADAR-based FCW system, the neural network in the supervisory MCU may learn when the FCW system is identifying a metal object that is not actually dangerous, such as a drainage grate or manhole cover that triggers an alarm. Similarly, when the auxiliary computer is a camera-based LDW system, the neural network in the supervisory MCU may learn to ignore the LDW when a cyclist or pedestrian is present and lane departure is actually the safest strategy. In an embodiment that includes a neural network running on the supervisory MCU, the supervisory MCU may include at least one of a DLA or a GPU suitable for running the neural network with associated memory. In a preferred embodiment, the supervisory MCU may include a component of SoC 404 and / or be included as a component of SoC 404.

[0136] In other examples, the ADAS system 438 may include an auxiliary computer that uses traditional computer vision rules to perform ADAS functions. In this way, the auxiliary computer can use classic computer vision rules (if-then), and the presence of a neural network in the supervisory MCU can improve reliability, safety, and performance. For example, diverse implementations and intentional non-identity make the entire system more fault-tolerant, especially for failures caused by software (or software-hardware interface) functions. For example, if there is a software vulnerability or error in the software running on the main computer and the non-identical software code running on the auxiliary computer provides the same overall result, the supervisory MCU can be more confident that the overall result is correct and that the vulnerability in the software or hardware used by the main computer does not cause a substantial error.

[0137] In some examples, the output of the ADAS system 438 can be fed to the perception block of the main computer and / or the dynamic driving task block of the main computer. For example, if the ADAS system 438 indicates a forward collision warning due to an object immediately ahead, the perception block can use this information in identifying the object. In other examples, the auxiliary computer can have its own neural network that is trained and thus reduces the risk of false positives as described herein.

[0138] The vehicle 400 may further include an infotainment SoC 430 (e.g., an in-vehicle infotainment system (IVI)). Although illustrated and described as an SoC, an infotainment system may not be an SoC and may include two or more discrete components. The infotainment SoC 430 may include a combination of hardware and software that may be used to provide audio (e.g., music, personal digital assistant, navigation instructions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), phone (e.g., hands-free calling), network connectivity (e.g., LTE, WiFi, etc.), and / or information services (e.g., navigation system, rear parking assistance, radio data system, vehicle-related information such as fuel level, total distance covered, brake fuel level, oil level, door open / closed, air filter information, etc.) to the vehicle 400. For example, the infotainment SoC 430 may include a radio, a disc player, a navigation system, a video player, USB and Bluetooth connections, an onboard computer, onboard entertainment, WiFi, steering wheel audio controls, hands-free voice controls, a head-up display (HUD), an HMI display 434, a telematics device, a control panel (e.g., for controlling and / or interacting with various components, features and / or systems), and / or other components. The infotainment SoC 430 may further be used to provide information (e.g., visual and / or auditory) to a user of the vehicle, such as information from an ADAS system 438, autonomous driving information such as planned vehicle maneuvers, trajectories, surrounding environment information (e.g., intersection information, vehicle information, road information, etc.), and / or other information.

[0139] 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 a 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 in the event of a failure of a master controller 436 (e.g., a main and / or backup computer of the vehicle 400), the GPU of the infotainment system may perform some self-driving functions. In such an example, the infotainment SoC 430 may place the vehicle 400 in a driver-safe parking mode as described herein.

[0140] The vehicle 400 may further include an instrument cluster 432 (e.g., a digital instrument panel, 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 a 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 indicator, a shift position indicator, a seat belt warning light, a parking brake warning light, an engine fault light, an airbag (SRS) system information, lighting controls, safety system controls, 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.

[0141] Figure 4D For a cloud-based server and Figure 4A 4. System diagram of communication between example autonomous vehicles 400. System 476 may include server 478, network 490, and vehicles including vehicle 400. Server 478 may include multiple GPUs 484 (A)-1284 (H) (collectively referred to as GPUs 484 here), PCIe switches 482 (A)-482 (H) (collectively referred to as PCIe switches 482 here), and / or CPUs 480 (A)-480 (B) (collectively referred to as CPUs 480 here). GPUs 484, CPUs 480, and PCIe switches may be interconnected with high-speed interconnects and / or PCIe connections 486 such as, for example and without limitation, NVLink interfaces 488 developed by NVIDIA. In some examples, GPUs 484 are connected via NVLink and / or NVSwitch SoCs, and GPUs 484 and PCIe switches 482 are connected via PCIe interconnects. Although eight GPUs 484, two CPUs 480, and two PCIe switches are illustrated, 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, each of the servers 478 may include eight, sixteen, thirty-two, and / or more GPUs 484.

[0142] Server 478 may receive image data over network 490 and from a vehicle, the image data representing images showing unexpected or changed road conditions, such as recently begun road work. Server 478 may transmit neural network 492, updated neural network 492, and / or map information 494, including information about traffic and road conditions, over network 490 and to the vehicle. Updates to map information 494 may include updates to HD map 422, such as information about construction sites, potholes, curves, flooding, or other obstacles. In some examples, neural network 492, updated neural network 492, and / or map information 494 may have been generated from new training and / or data received from any number of vehicles in the environment and / or based on experience of training performed at a data center (e.g., using server 478 and / or other servers).

[0143] Server 478 can be used to train a machine learning model (e.g., a neural network) based on training data. The training data can be generated by the vehicle, and / or can be generated in a simulation (e.g., using a game engine). In some examples, the training data is labeled (e.g., when the neural network benefits from supervised learning) and / or undergoes other preprocessing, while in other examples, the training data is not labeled and / or preprocessed (e.g., when the neural network does not require supervised learning). Training can be performed according to any one or more classes of machine learning techniques, including but not limited to the following classes: supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, joint learning, transfer learning, feature learning (including principal components and cluster analysis), multilinear subspace learning, manifold learning, representation learning (including alternate dictionary learning), rule-based machine learning, anomaly detection, and any variants or combinations thereof. Once the machine learning model is trained, the machine learning model can be used by the vehicle (e.g., transmitted to the vehicle via network 490), and / or the machine learning model can be used by server 478 to remotely monitor the vehicle.

[0144] In some examples, server 478 can receive data from the vehicle and apply the data to the latest real-time neural network for real-time intelligent reasoning. Server 478 may include a deep learning supercomputer and / or a dedicated AI computer powered by GPU 484, such as DGX and DGX Station machines developed by NVIDIA. However, in some examples, server 478 may include a deep learning infrastructure of a data center powered only by CPUs.

[0145] The deep learning infrastructure of server 478 may be capable of rapid real-time inference, and may use this capability to assess and verify the health of the processors, software, and / or associated hardware in vehicle 400. For example, the deep learning infrastructure may receive periodic updates from vehicle 400, such as a sequence of images and / or objects located in the sequence of images that vehicle 400 has located (e.g., via computer vision and / or other machine learning object classification techniques). The deep learning infrastructure may run its own neural network to identify objects and compare them to the objects identified by vehicle 400, and if the results do not match and the infrastructure concludes that the AI ​​in vehicle 400 has failed, server 478 may transmit a signal to vehicle 400 instructing the fail-safe computer of vehicle 400 to take control, notify passengers, and complete a safe parking maneuver.

[0146] For reasoning, the server 478 may include a GPU 484 and one or more programmable reasoning accelerators (e.g., NVIDIA's TensorRT 3). The combination of GPU-powered servers and reasoning acceleration can make real-time responses possible. In other examples, such as when performance is not so important, CPU, FPGA, and other processor-powered servers can be used for reasoning.

[0147] Figure 5 5 is a block diagram of an example computing device 500 suitable for implementing some embodiments of the present disclosure. The computing device 500 may include an interconnect 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, communication interfaces 510, I / O ports 512, input / output components 514, power supplies 516, one or more presentation components 518 (e.g., displays), and one or more logic units 520.

[0148] although Figure 5 The various blocks of are shown as being connected via an interconnect system 502 having wires, but this is not intended to be limiting and is merely for clarity. 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 touch screen). As another example, CPU 506 and / or GPU 508 may include memory (e.g., memory 504 may represent a storage device in addition to the memory of GPU 508, CPU 506, and / or other components). In other words, Figure 5The computing devices in the description are merely illustrative. No distinction is made between categories such as "workstations," "servers," "laptops," "desktops," "tablets," "client devices," "mobile devices," "handheld devices," "game consoles," "electronic control units (ECUs)," "virtual reality systems," "augmented reality systems," and / or other device or system types, as all of these are considered Figure 5 within the range of computing devices.

[0149] The interconnection 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. The interconnection system 502 may include one or more bus types, such as an industry standard architecture (ISA) bus, an extended industry standard architecture (EISA) bus, a video electronics standard 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 is a direct connection between components. As an example, the CPU 506 may be directly connected to the memory 504. In addition, the CPU 506 may be directly connected to the GPU 508. In the case where there is a direct or point-to-point connection between components, the interconnection system 502 may include a PCIe link to perform the connection. In these examples, the computing device 500 does not need to include a PCI bus.

[0150] 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 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.

[0151] Computer storage media may include volatile and nonvolatile media and / or removable and non-removable media implemented in any method or technology for storing 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 are not limited to, RAM, ROM, EEPROM, flash memory or other storage technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, 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 can be accessed by computing device 500. As used herein, computer storage media does not include the signals themselves.

[0152] Computer storage media may contain computer readable instructions, data structures, program modules, and / or other data types in a modulated data signal such as a carrier wave or other transmission mechanism, and include 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 way that information is encoded into the signal. By way of example and not limitation, computer storage media may include wired media such as a wired network or a direct wired connection, and wireless media such as sound, RF, infrared, and other wireless media. Any combination of the above should also be included within the scope of computer readable media.

[0153] The CPU 506 may be configured to execute at least some computer-readable instructions in order to control one or more components of the computing device 500 to perform one or more of the methods and / or processes described herein. Each of the CPUs 506 may include one or more cores (e.g., one, two, four, eight, twenty-eight, seventy-two, etc.) capable of simultaneously processing a large number of software threads. The CPU 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 mobile devices and a processor with more cores for servers). 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 also include one or more CPUs 506 in addition to one or more microprocessors or supplementary coprocessors such as math coprocessors.

[0154] In addition to or in lieu of CPU 506 , GPU 508 may be configured to execute at least some computer-readable instructions to control one or more components of computing device 500 to perform one or more methods and / or processes described herein. One or more GPUs 508 may be an integrated GPU (e.g., with one or more CPUs 506) and / or one or more GPUs 508 may be a discrete GPU. In embodiments, one or more GPUs 508 may be a coprocessor of one or more CPUs 506. The computing device 500 may use the GPU 508 to render graphics (e.g., 3D graphics) or perform general-purpose computing. For example, the GPU 508 may be used for general-purpose computing on a GPU (GPGPU). The GPU 508 may include hundreds or thousands of cores capable of processing hundreds or thousands of software threads simultaneously. The GPU 508 may generate pixel data for outputting an image in response to a rendering command (e.g., a rendering command received from the CPU 506 via a host interface). The 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 the memory 504. The GPU 508 may include two or more GPUs operating in parallel (e.g., via a link). The link may connect the GPUs directly (e.g., using NVLINK) or may connect the GPUs through a switch (e.g., using NVSwitch). When combined together, each GPU 508 may generate pixel data or GPGPU data for a different portion of the output or for different outputs (e.g., a first GPU for a first image and a second GPU for a second image). Each GPU may include its own memory or may share memory with other GPUs.

[0155] In addition to or in lieu of the CPU 506 and / or the GPU 508, the logic unit 520 may be configured to execute at least some computer-readable instructions to control one or more components of the computing device 500 to perform one or more methods and / or processes described herein. In embodiments, the CPU 506, the GPU 508, and / or the logic unit 520 may perform any combination of methods, processes, and / or portions thereof, either separately or in conjunction. The one or more logic units 520 may be a part of and / or integrated with one or more of the CPU 506 and / or the GPU 508, and / or the one or more logic units 520 may be a discrete component or otherwise external to the CPU 506 and / or the GPU 508. In embodiments, the one or more logic units 520 may be a coprocessor to the one or more CPUs 506 and / or the one or more GPUs 508.

[0156] Examples of logic unit 520 include one or more processing cores and / or components thereof, such as a tensor core (TC), a tensor processing unit (TPU), a pixel vision core (PVC), a vision processing unit (VPU), a graphics processing cluster (GPC), a texture processing cluster (TPC), a streaming multiprocessor (SM), a tree traversal unit (TTU), an artificial intelligence accelerator (AIA), a deep learning accelerator (DLA), an arithmetic logic unit (ALU), an application specific integrated circuit (ASIC), a floating point unit (FPU), an I / O element, a peripheral component interconnect (PCI) or a peripheral component interconnect express (PCIe) element, etc.

[0157] The communication interface 510 may include one or more receivers, transmitters, and / or transceivers that enable the computing device 500 to communicate with other computing devices via an electronic communication network, including wired and / or wireless communications. The communication interface 510 may include components and functionality that enable communication over any of a number of different networks, such as a wireless network (e.g., Wi-Fi, Z-wave, Bluetooth, Bluetooth LE, ZigBee, etc.), a wired network (e.g., via Ethernet or Infiniband communications), a low power wide area network (e.g., LoRaWAN, SigFox, etc.), and / or the Internet.

[0158] The I / O ports 512 can enable the computing device 500 to be logically coupled to other devices including I / O components 514, presentation components 518, and / or other components, some of which can be built into (e.g., integrated into) the computing device 500. Illustrative I / O components 514 include microphones, mice, keyboards, joysticks, game pads, game controllers, satellite dishes, scanners, printers, wireless devices, and the like. The I / O components 514 can provide a natural user interface (NUI) that processes air gestures, voice, or other physiological inputs generated by a user. In some instances, the input can be transmitted to an appropriate network element for further processing. The NUI can implement any combination of voice recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition on and adjacent to the screen, air gestures, head and eye tracking, and touch recognition associated with the display of the computing device 500 (as described in more detail below). The computing device 500 may include a depth camera such as a stereo camera system, an infrared camera system, an RGB camera system, touch screen technology, and combinations of these for gesture detection and recognition. Additionally, computing device 500 may include an accelerometer or gyroscope to enable motion detection (e.g., as part of an inertial measurement unit (IMU)). In some examples, the output of the accelerometer or gyroscope may be used by computing device 500 to render immersive augmented reality or virtual reality.

[0159] 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 head-up display (HUD), other display types, or a combination thereof), a speaker, and / or other presentation components. The presentation component 518 may receive data from other components (e.g., GPU 508, CPU 506, etc.) and output the data (e.g., as an image, video, sound, etc.).

[0160] The present disclosure may be described in the general context of machine-usable instructions or computer codes executed by a computer or other machine such as a personal digital assistant or other handheld device, including computer-executable instructions such as program modules. Typically, program modules including routines, programs, objects, components, data structures, etc. refer to codes that perform specific tasks or implement specific 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 a distributed computing environment in which tasks are performed by remote processing devices linked through a communication network.

[0161] As used herein, the description of "and / or" about two or more elements should be interpreted as referring to only one element or a combination of elements. For example, "element A, element B and / or element C" may include only element A, only element B, only element C, element A and element B, element A and element C, element B and element C, or elements A, B and C. In addition, "at least one of element A or element B" may include at least one of element A, at least one of element B, or at least one of element A and at least one of element B. Further, "at least one of element A and element B" may include at least one of element A, at least one of element B, or at least one of element A and at least one of element B.

[0162] The subject matter of the present disclosure is described in detail herein to meet statutory requirements. However, the description itself is not intended to limit the scope of the present disclosure. On the contrary, the inventors have contemplated that the claimed subject matter may also be embodied in other ways to include steps different from the steps described herein in conjunction with other current or future technologies or combinations of similar steps. Moreover, although the terms "step" and / or "block" may be used herein to imply different elements of the method employed, these terms should not be interpreted as implying any particular order among or between the various steps disclosed herein, unless the order of the steps is explicitly described.

[0163] Combine the following Figure 6Each block of the described methods includes a computing process that can be performed using any combination of hardware, firmware, and / or software. For example, the various functions can be performed by a processor executing instructions stored in a memory. The methods can also be implemented as computer-usable instructions stored on a computer storage medium. The methods can be provided by a stand-alone application, a service, or a hosted service (standalone or in combination with another hosted service), or a plug-in for another product. In addition, as an example, for Figures 4A-4D An exemplary autonomous vehicle system is described in Figure 6 However, these methods may additionally or alternatively be performed by any one system or any combination of systems, including but not limited to those described herein.

[0164] Figure 6 A flow chart illustrating process steps for determining gaze direction in the presence of glare according to an embodiment of the present disclosure. Figure 6 The processing begins when the computing device 300 receives an image of the subject captured by the camera 310 (step 600). The computing device then identifies the face and eyes of the subject in the received image (step 610). Any method or process can be used to locate the face of the subject within the image, including known computer vision-based face detection processes that do not use neural networks to detect faces, such as edge detection methods, feature search methods, probabilistic face models, graphic matching, histograms of oriented gradients (HOG), which are fed to input classifiers such as support vector machines, HaarCascade classifiers, etc. Neural network-based facial recognition methods can also be used to determine the facial position, such as methods using deep neural network (DNN) facial recognition schemes and any other methods. Embodiments of the present disclosure also consider the location of the eyes of the subject from the same image. Eye positioning can be performed in any manner, such as by a known computer vision-based eye detection process, including any of the above-mentioned non-neural network-based techniques, neural network-based eye recognition methods, etc.

[0165] Once the face and eyes of the subject are within the received image, computing device 300 extracts glare points from the received image and generates a facial representation (step 620). Any method or process may be used to identify glare. For example, glare pixels may be identified as those pixels in the received image having a brightness value above a predetermined threshold. As another example, glare pixels may be identified based on known or characteristic glare shapes (e.g., circular or star-shaped spots, etc.) in the captured image. For example, a glare mask may be formed by setting glare pixels to white or saturated pixels and setting the remaining pixels to black pixels.

[0166] Any method or process may be used to determine a facial representation, such as a binary mask of an identified face. For example, such a binary mask may be formed by setting pixels corresponding to an identified face to white or another uniform color and the remaining pixels to black pixels.

[0167] The eye crop is then selected (step 630), for example by drawing a bounding box around each eye identified in step 610 in a known manner, and cropping the image accordingly. Similarly, the face mesh is then calculated (step 640). The face mesh 225 can be formed, for example, by cropping the identified face according to a bounding box that can be drawn in any known manner and projecting the cropped image onto a predetermined coarse grid, containing grid elements that assign one or more facial pixels a color (or a first value), and grid elements that do not contain facial pixels that are set to black (or a value different from the first value).

[0168] Then, the mask of step 620, the eye cropping of step 630, and the facial mesh of step 640 are input to the machine learning model of the embodiment of the present disclosure (e.g., feature extraction layers 230, 235, and 240, and fully connected layers 245 and 255) to calculate the gaze direction as described above (step 650). The gaze direction is then output to the gaze assistance system 320 (step 660), which can perform any one or more actions based on the gaze direction it receives (step 670).

[0169] The embodiments of the present disclosure consider Figure 2 A machine learning model that is configured in a way other than that shown in . Figure 7 is a block diagram representation of one such alternative example. Here, Figure 7 The machine learning model includes fully connected layers 710, 740 and 755, and feature extraction layers 715, 745 and 750. The feature extraction layers 745 and 750 and the fully connected layer 755 have the left eye representation 725, the right eye representation 730 and the face representation 735 as inputs. Figure 7 The three rightmost branches of the model are basically the same as Figure 2 That is, feature extraction layers 745 and 750 and fully connected layer 755 may be substantially the same as feature extraction layers 235 and 240 and fully connected layer 245, respectively. Similarly, inputs 725, 730, and 735 may be substantially the same as inputs 215, 220, and 225, respectively.

[0170] The glare representation 700 and the face representation 705 may be associated with Figure 2The inputs 200 and 205 are substantially the same. Thus, the glare representation 700 may be a binary mask of glare points identified in the input image 100, and the face representation 705 may be a binary mask of facial pixels extracted from the input image 100. However, unlike Figure 2 Differently, the glare representation 700 is input to a fully connected layer 710, and the face representation 705 is input to a feature extraction layer 715. The fully connected layer 710 outputs features describing the spatial location of the glare in the input image 100, and the feature extraction layer 715 outputs features describing the orientation of the face in the input image 100. These two outputs are concatenated by a concatenation block 720 and input to a fully connected layer 740, which outputs a set of features describing the orientation of the face in the input image 100. Then, as described above, combined with Figure 2 As described above, the outputs of the fully connected layer 740, the feature extraction layers 745 and 750, and the fully connected layer 755 are processed. Specifically, the outputs are concatenated through the concatenation block 760 and input to the fully connected layer 765 which outputs the gaze direction. In this way, the spatial position of the glare is used to determine the facial direction, rather than the glare point itself.

[0171] The training of any fully connected layer and feature extraction layer of the embodiments of the present disclosure may be performed in any suitable manner. In some embodiments, the machine learning model of the embodiments of the present disclosure, for example, Figure 2 and 7 , end-to-end training can be performed. In some embodiments, the fully connected layers and feature extraction layers of the disclosed embodiments can be trained in a supervised manner using labeled input images of objects whose gaze directions are known. In some embodiments, glare points can be added to these input images to simulate the presence of glare. The added glare points can be added to their respective images in any manner, such as by random placement in various images, placement at predetermined locations in the input images, etc.

[0172] Those skilled in the art will note that embodiments of the present disclosure are not limited to the estimation of gaze. Specifically, the methods and systems of embodiments of the present disclosure may be used to account for glare when determining any oculomotor variable. For example, Figure 2 and / or Figure 7The machine learning model of the present disclosure can be constructed and trained so that the output of the FC layers 255, 765 is an estimate of any other oculomotor variable in addition to gaze, such as pupil size, pupil movement, cognitive load, fatigue, incapacity due to drugs or alcohol, for example, person or object recognition through the iris, any other physiological response, etc. This can be achieved by, for example, training the machine learning model of the present disclosure embodiment using an input data set labeled with the values ​​of any oculomotor variable of interest (e.g., glare masks, masks, eye clipping, facial meshes, etc.). The machine learning model can also be constructed to extract any features useful in determining such oculomotor variables. In this way, the machine learning model of the present disclosure embodiment can be constructed and trained to estimate any oculomotor variable in a glare-resistant manner, that is, taking glare into account when determining any oculomotor variable.

[0173] For the purpose of explanation, the above description uses specific terms to provide a thorough understanding of the present disclosure. However, it is obvious to those skilled in the art that the methods and systems of the present invention do not require specific details to implement them. Therefore, for the purpose of illustration and description, the foregoing description of specific embodiments of the present invention is given. They are not intended to be exhaustive or to limit the present invention to the precise form disclosed. In view of the above teachings, many modifications and variations are possible. For example, glare points can be input to various machine learning models in any manner, such as by inputting a binary mask of glare points, an input table, or a set of glare positions or any other representation of glare. The embodiments are selected and described in order to best explain the principles of the present invention and its practical application, so that other technicians in the field can best utilize the methods and systems of the present invention and various embodiments with various modifications to suit the intended specific purposes. In addition, different features of various embodiments disclosed or otherwise disclosed can be mixed and matched or otherwise combined to create other embodiments contemplated by the present invention.

Claims

1. A method for determining gaze direction in the presence of glare, the method comprising: determining a gaze direction of an object in an image using parallel processing circuitry, the gaze direction determined at least in part based on an output of one or more machine learning models that take as input an isolated representation of glare in the image, one or more representations of at least a portion of the object's face, and a portion of the image corresponding to at least one eye of the object, wherein the one or more representations of at least a portion of the object's face include a first representation and a second representation of the object's face, the first representation of the object's face including a mask of the object's face, the second representation of the object's face including a facial mesh of the object's face, and the one or more machine learning models include a first machine learning model that takes as input the first representation of the object's face, and a second machine learning model that takes as input the second representation of the object's face; and Initiate actions based on the determined gaze direction.

2. The method of claim 1 further comprising generating, using processing circuitry and based at least in part on the image, a representation of the glare, the one or more representations of at least a portion of the subject's face, and a portion of the image corresponding to at least one eye of the subject.

3. The method of claim 1, wherein: One of the machine learning models takes as input both an isolated representation of the glare and a representation of at least a portion of the subject's face.

4. The method of claim 1, wherein: One of the machine learning models takes as input an isolated representation of the glare and another of the machine learning models takes as input a representation of at least a portion of the subject's face.

5. The method of claim 4, wherein: One of the machine learning models includes a fully connected layer with an isolated representation of the glare as input.

6. The method of claim 1, wherein: The image is generated using sensor data obtained by a sensor device corresponding to a vehicle, wherein the initiating further comprises initiating operation of the vehicle based on the determined gaze direction.

7. The method according to claim 1: in, The portion of the image corresponding to at least one eye of the subject also includes a first portion of the image corresponding to a first eye of the subject, and a second portion of the image corresponding to a second eye of the subject.

8. A system for determining gaze direction in the presence of glare, the system comprising: Memory; and A parallel processing circuit configured to: determining a gaze direction of a subject in an image, the gaze direction determined at least in part based on an output of one or more machine learning models that take as input an isolated representation of glare in the image, one or more representations of at least a portion of the subject's face, and a portion of the image corresponding to at least one eye of the subject, wherein the one or more representations of at least a portion of the subject's face include a mask of the subject's face and a facial mesh of the subject's face, and wherein the one or more machine learning models include a first machine learning model that takes as input the mask of the subject's face and a second machine learning model that takes as input the facial mesh of the subject's face; and An operation is initiated according to the gaze direction.

9. The system according to claim 8, wherein: The parallel processing circuitry is further configured to generate an isolated representation of the glare, the one or more representations of at least a portion of the subject's face, and a portion of the image corresponding to the at least one eye of the subject, all based at least in part on the image.

10. The system of claim 8, wherein: One of the machine learning models takes as input both an isolated representation of the glare and a representation of at least a portion of the subject's face.

11. The system according to claim 8, wherein: One of the machine learning models takes as input an isolated representation of the glare and another of the machine learning models takes as input a representation of at least a portion of the subject's face.

12. The system of claim 11, wherein: One of the machine learning models includes a fully connected layer that takes as input the isolated representation of the glare.

13. The system of claim 8, wherein: The operations also include operation of a vehicle based at least in part on the gaze direction.

14. The system of claim 8: in, The portion of the image corresponding to at least one eye of the subject also includes a first portion of the image corresponding to a first eye of the subject, and a second portion of the image corresponding to a second eye of the subject.

15. A method for determining gaze direction in the presence of glare, the method comprising: receiving first input data indicating a glare location in an image, the first input data comprising: a mask of a face of an object in the image; receiving second input data representing at least a portion of the image corresponding to an object presented in the image, the second input data comprising a facial mesh of a face of the object; determining, using parallel processing circuitry, a gaze direction of the subject, the gaze direction being an output of one or more machine learning models using the first input data, the second input data, and a portion of the image corresponding to at least one eye of the subject; and An action is initiated based on the gaze direction.

16. The method of claim 15, wherein: One of the machine learning models has as input first input data and third input data comprising at least one position of the subject's eye in the image.

17. The method of claim 15, wherein: One of the machine learning models has as input the first input data and another of the machine learning models has as input at least one position of the subject's eye in the image.

18. The method of claim 15, wherein: The operations also include operation of a vehicle based at least in part on the gaze direction.

19. The method of claim 15, wherein: The one or more machine learning models also have third input data, which includes a portion of the image corresponding to the first eye of the subject, and wherein the one or more machine learning models also have fourth input data, which includes a portion of the image corresponding to the second eye of the subject.

Images