Intersection area detection and classification for autonomous machine applications

Abstract

In various instances, live perception from a vehicle's sensors can be leveraged to detect and classify intersection conflict areas within the vehicle's environment in real time or near real time. For example, a deep neural network (DNN) can be trained to calculate an output—e.g., a signed distance function—that can correspond to the location of a boundary outlining the intersection conflict area. The signed distance function can be decoded and / or post-processed to determine an instance classification mask that represents the location and classification of the intersection area or region. The location of the intersection area or region can be generated in image space and converted to world space coordinates to aid an autonomous or semi-autonomous vehicle in navigating the intersection in accordance with road rules, traffic priority considerations, and / or the like.

Description

Technical Field

[0001] The present invention relates to intersection area detection and classification for autonomous machine applications.

Background Art

[0002] Autonomous driving systems and advanced driver assistance systems (ADAS) can use various sensors, such as cameras, LIDAR sensors, RADAR sensors, etc., to perform various tasks - for example, lane keeping, lane changing, lane designation, camera calibration, turning, route planning, and localization, etc. For example, in order for autonomous and ADAS systems to operate independently and efficiently, an understanding of the vehicle's surrounding environment in real-time or near real-time can be generated. This understanding can include information about the position and type of objects, obstacles, lanes, and / or intersections in the environment with respect to various demarcations, such as lanes, road boundaries, intersections, and / or the like. Information about the surrounding environment can be used by the vehicle when making decisions such as which route or trajectory to follow, how to apply the rules of the road, when, where, and / or how much to stop.

[0003] As an example, information about the location, attributes, and layout of intersections in an autonomous or semi-autonomous vehicle's environment—for example, where other vehicles or pedestrians might be located, whether other vehicles have priority to cross the intersection, and / or similar—may prove useful when making route planning, obstacle avoidance, and / or control decisions. This is particularly important when a vehicle is operating in an urban and / or semi-urban driving environment, where understanding intersections and route planning becomes critical due to the increased number of variables associated with a major highway driving environment. For example, if a vehicle must slow down and come to a complete stop to wait for other vehicles with priority to cross an intersection in a two-way, multi-lane driving environment, determining the location and class of the intersection area (e.g., pedestrian crossing, vehicle entry, vehicle exit, no lane, etc.) becomes critical for safe and effective autonomous and / or semi-autonomous driving.

[0004] In conventional systems, intersections can be interpreted by individually detecting and combining several features of the intersection and its surrounding environment. For example, to detect an intersection, the intersection area, multiple objects (e.g., traffic lights, stop signs), vehicle location, vehicle orientation, lanes, free-space boundaries, etc., can be individually detected—for example, using multiple separate deep neural networks (DNNs)—and then stitched together to produce a combined understanding of a single intersection. However, such solutions require precise and detailed networks to recognize relevant features for intersection detection and to combine the various features of each intersection to safely cross the intersection while adhering to different traffic priority rules. As a result, the more complex the intersection, the more detailed annotations are required to train the DNN—thereby reducing the scalability of intersection detection while increasing the complexity of accurate detection and classification of intersection areas. Specifically, these conventional systems may not achieve a clearer and more complete understanding of vehicles that may be competitors with priority to cross the intersection and / or pedestrians crossing the intersection. In addition, many conventional systems also encounter obstructions in sensor data—for example, due to buildings or other vehicles or objects that obstruct part of the vehicle's sensor's perception field—and are therefore unable to robustly detect intersection areas.

[0005] Other conventional systems can analyze intersections by comparing individually detected features with features in pre-stored high-resolution (HD), three-dimensional (3D) maps of the vehicle's travel surface. However, such map-based solutions are heavily dependent on the accuracy and availability of the maps. Therefore, these conventional systems fail when maps for a particular area are outdated or unavailable. For example, the process can become logistically more complex when manual labeling of larger geographical areas (e.g., cities, states, countries) is required for vehicles to navigate independently and effectively at locations with various intersection types. These conventional systems also fail when temporary intersection conditions exist that cannot be reflected in the map (e.g., police traffic control or a stopped school bus). [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] U.S. Patent Application No. 16 / 101,232 [Overview of the Initiative] [Means for solving the problem]

[0007] Embodiments of this disclosure relate to intersection competition area detection for autonomous machine applications. A system and method are disclosed that uses a deep neural network (DNN) to process sensor data from one or more sensors of a vehicle in order to detect intersection areas—and their corresponding classifications—and to produce a holistic understanding of the intersection in real time or near real time.

[0008] In contrast to conventional systems, such as those mentioned above, this system can detect and classify intersection competition areas in the vehicle's environment using live vehicle perception. For example, a machine learning model—such as a DNN—can be trained to compute outputs corresponding to intersections—for example, signed distance functions corresponding to intersection regions or areas and their classes—and a post-processor can use this information to effectively and accurately determine the location and classification of the intersection regions. The outputs can be post-processed using various techniques—for example, smoothing, clustering, thresholding, upsampling, etc.—to determine the 3D world-space location of the detected intersection areas. These world-space locations and / or classifications of intersection areas can be used by the vehicle to safely and accurately navigate intersections. For example, the output of a DNN can be used directly or indirectly (for example, via decoding) to determine the location and corresponding classification of each intersection area.

[0009] As a result of using real-time or near-real-time holistic methods to generate an understanding of each intersection area, the system can learn to diagnose each intersection without requiring prior experience or knowledge of the intersection, and without the need to individually detect and then combine several features of the intersection. Therefore, the process of detecting and classifying intersection conflict areas can be relatively time-efficient, computationally intensive, and more scalable. Consequently, vehicles may have the ability to navigate streets, urban environments, and / or other locations more freely—without relying on HD maps, or in addition to relying on HD maps.

[0010] The system and method for detecting intersection conflict areas for autonomous machine applications are described in detail below with reference to the attached drawings. [Brief explanation of the drawing]

[0011] [Figure 1]This is an exemplary data flowchart illustrating an exemplary process for training a neural network to detect intersection areas in order to navigate intersections using outputs from vehicle sensors, according to some embodiments of the present disclosure. [Figure 2A] Illustrations illustrating illustrative annotations of images for generating ground truth data to train a neural network to detect intersection areas, according to some embodiments of the present disclosure. [Figure 2B] Illustrations illustrating illustrative annotations of images for generating ground truth data to train a neural network to detect intersection areas, according to some embodiments of the present disclosure. [Figure 3] Illustrations of exemplary ground truth signed distance functions for encoding intersection areas from annotations, according to some embodiments of the present disclosure. [Figure 4] This flowchart illustrates an exemplary method for training a neural network to detect intersection areas, according to some embodiments of the present disclosure. [Figure 5] This data flowchart illustrates an exemplary process for detecting intersection areas using real-time or near-real-time outputs from vehicle sensors, according to some embodiments of the present disclosure. [Figure 6] This figure shows a visualization of an exemplary signed distance function, which is converted to a corresponding instance partition mask using a post-processor, according to some embodiments of the present disclosure. [Figure 7A] This figure shows a visualization of exemplary intersection areas predicted in 3D space based on a signed distance function computed by a neural network, according to some embodiments of the present disclosure. [Figure 7B] This figure shows a visualization of exemplary intersection areas predicted in 3D space based on a signed distance function computed by a neural network, according to some embodiments of the present disclosure. [Figure 8]This flowchart illustrates an exemplary method for detecting an intersection area according to some embodiments of the present disclosure. [Figure 9A] Illustrations of exemplary autonomous vehicles according to some embodiments of the present disclosure. [Figure 9B] Figure 9A shows examples of camera positions and fields of view of an exemplary autonomous vehicle according to some embodiments of the present disclosure. [Figure 9C] Figure 9A is a block diagram of an exemplary system architecture of an exemplary autonomous vehicle according to some embodiments of the present disclosure. [Figure 9D] This is a system diagram of communication between a cloud-based server and the exemplary autonomous vehicle shown in Figure 9A, according to some embodiments of the present disclosure. [Figure 10] This is a block diagram of an exemplary computing device suitable for use in implementing some embodiments of the present disclosure. [Modes for carrying out the invention]

[0012] Systems and methods relating to intersection competition area detection for autonomous machine applications are disclosed. While this disclosure may be described in relation to an exemplary autonomous vehicle 900 (illustrated with respect to Figures 9A–9D and referred to herein as "vehicle 900" or "autonomous vehicle 900"), this is not intended to be limiting. For example, the systems and methods described herein may be used by non-autonomous vehicles, semi-autonomous vehicles (e.g., in one or more adaptive driver assistance systems (ADAS)), robots, warehouse vehicles, off-road vehicles, airships, boats, shuttles, emergency response vehicles, motorcycles, electric or motorized bicycles, aircraft, construction vehicles, submarines, drones, and / or other vehicle types. In addition, although this disclosure may be described as an intersection structure for vehicle applications, this is not intended to be limiting, and the systems and methods described herein may be used in augmented reality, virtual reality, robotics, security and surveillance, autonomous or semi-autonomous machine applications, and / or in any other technological space in which intersection or other environmental structure and / or attitude detection may be used.

[0013] As described herein, in contrast to conventional approaches for intersection conflict area detection that use high-definition (HD) maps, the present system and method provide a technique for detecting and classifying intersection areas using the output from vehicle sensors (e.g., cameras, RADAR sensors, LIDAR sensors, etc.) in real-time or near real-time. As such, for each intersection, the live perception of the vehicle can be used to detect the location and / or attributes or classification of the intersection area corresponding to the intersection. A computer vision and / or machine learning model (e.g., a deep neural network (DNN), such as a convolutional neural network (CNN)) can be trained to calculate an output that results in the detected intersection area and / or its classification or attributes - in an embodiment, after decoding - and the output can be used by the vehicle in conjunction with object and / or lane detection to effectively and accurately navigate the intersection while observing the relevant traffic priority rules. The output of the computer vision and / or machine learning model can, in some embodiments, include a signed distance function that represents the pixel-based location of a particular region within an image where various classified intersection areas were detected. Additionally, in some embodiments, post-processing can be performed on the signed distance function to generate an instance segmentation mask corresponding to each detected intersection area and / or type.

[0014] As such, the system can learn to diagnose each intersection area in real-time or near real-time by using live perception to generate an understanding of the intersection area, so the process of detecting and classifying intersection areas can be relatively quick, have a low computational load, and be more scalable. Additionally, in further contrast to conventional systems, prior knowledge or experience of the intersection is not required - for example, the system can operate accurately without the need for an up-to-date HD map being available.

[0015] During deployment, sensor data (e.g., images, videos, point clouds, depth maps, and other sensor data representations) may be received and / or generated using sensors (e.g., cameras, radar sensors, and LiDAR sensors) placed on or otherwise positioned within autonomous or semi-autonomous vehicles. The sensor data may be applied to a DNN trained to identify the area of ​​interest related to the intersection represented by the sensor data (e.g., the interior of the intersection, the entrance to the intersection, the exit to the intersection, areas without lanes, pedestrian crossings, obscured areas, etc.) and the semantic information relating thereto. More specifically, the DNN may be trained to compute a signed distance function corresponding to each detected intersection area instance. A signed distance function may correspond for each pixel or point in sensor data to the (pixel) distance to the nearest boundary corresponding to the intersection area, the position relative to the intersection area (e.g., negative for pixels outside the intersection area, positive for pixels inside the intersection area, or vice versa), semantic information corresponding to the intersection area (e.g., crosswalk, intersection entrance, intersection exit, unclear area, no lane, inside the intersection, partially visible, fully visible, or a combination thereof), and / or other information. In some examples, the calculated intersection area in the signed distance function may be represented by pixels or points represented by sensor data, where pixels on the boundary of the intersection area are represented by a value of zero, pixels inside the boundary are represented by a positive distance value corresponding to the distance to the nearest boundary of the area, and pixels outside the boundary are represented by a negative value corresponding to the distance to the nearest boundary of the area. One or more instances of each class of intersection areas may be represented by separate signed distance functions (e.g., one signed distance function for each class, a separate signed distance function for each instance of each class, etc.). As such, the location and class of an intersection area can be determined based on a signed distance function and related information—for example, using a post-processor. In non-limiting examples, an intersection area class may include pedestrian crossings, intersection entrances, intersection exits, obscured areas, no lanes, inside the intersection, partially visible, fully visible, and / or other information corresponding to the intersection.

[0016] The DNN can be trained to predict various types of information corresponding to intersection area classification - for example, via any number of channels. For example, one or more of the channels can represent a signed distance function of the class of the intersection area and / or of each instance thereof. During training, the DNN can be trained with images or other sensor data representations labeled or annotated with polygons representing sidewalk areas, intersection entrance areas, intersection exit areas, unclear areas, areas without lanes, intersection interior areas, partially visible areas, fully visible areas, etc. In addition to the boundary labels or annotations, the ground truth annotations can include semantic information - for example, classification - corresponding to each boundary. The labeled polygons and semantic information can be used by a ground truth encoder to generate a signed distance function corresponding to an intersection area type corresponding to the area of the intersection as determined from the annotations. In some examples, each pixel within the signed distance function can be encoded with a distance value corresponding to the distance to the closest boundary of the intersection area (e.g., along the annotated boundary shape of the training data). In such examples, pixels inside the intersection area can also be encoded with positive distance values, pixels outside the intersection area can be encoded with negative distance values, or vice versa. Pixels forming the boundary of the intersection area can be encoded with zero values to generate the ground truth data.

[0017] In some examples, in addition to or instead of using a signed distance function, labeled polygons and semantic information may be used by a ground truth encoder to generate a one-hot encoding map for each intersection area type, where pixels inside the intersection area boundary are encoded with a value of 1, and pixels outside the intersection area boundary are encoded with a value of 0. However, when using one-hot encoding, adjacent areas of the same type (e.g., those with a shared boundary) may not be individually identifiable, resulting in reduced accuracy compared to embodiments using a signed distance function. Finally, since information can be determined using polygon annotations and semantic information, intersection areas may be encoded using a signed distance function and / or a one-hot encoding map that requires limited labeling.

[0018] After the DNN has been trained, the DNN may regress on its outputs in the form of signed distance functions, one-hot encoding maps, and / or other outputs corresponding to encoded ground truth channels trained to predict the network, as described herein. Where the outputs correspond to signed distance functions, pixel distances corresponding to different classification types, and / or instances thereof, may be post-processed into two-dimensional (2D) coordinates representing the 2D location of the intersection area (e.g., each pixel in the area, a proper boundary pixel, or its vertex). In some embodiments, the signed distance function may undergo Gaussian smoothing on a pixel-by-pixel basis to smooth out sharp peaks and fluctuations in pixel values. The smoothed signed distance function may then be mapped to polygon areas representing the intersection area based on the negative and positive pixel values ​​associated with each pixel in the corresponding signed distance function. In some embodiments, a watershed clustering algorithm may be used to generate instance partition maps corresponding to each detected intersection area or region for each signed distance function. In such embodiments, for each intersection area, pixels with excessively high and excessively low distance values ​​may be filtered out using a predetermined distance threshold—thereby reducing the computational resources required. The remaining pixels may be used to determine the boundaries of the associated intersection areas and / or to determine each pixel associated with a given region of an intersection having an associated classification type. Each pixel within the boundary of the polygon representing the intersection area may then be encoded with a pixel value indicating the division of the corresponding intersection area. The output of the post-processing may also be an instance partition mask for each classification type—or its instances—corresponding to the locations of different classification type regions of the intersection.

[0019] In some embodiments, 2D image space position may be used to determine the three-dimensional (3D) world space coordinates of an intersection area to assist a vehicle in navigating a world space environment. For example, sensor-specific parameters and / or external parameters may be used to map the 2D image space position to a known 3D world space position. However, in some embodiments, a DNN may be trained on 3D data to predict the position in 3D world space.

[0020] Referring here to Figure 1, Figure 1 is an illustrative data flowchart showing an exemplary process 100 for training a neural network to detect an intersection area or region to assist a vehicle in intersection navigation, according to some embodiments of the present disclosure. It should be understood that this and other configurations described herein are merely illustrative. At a higher level, process 100 may include one or more machine learning models 104 receiving one or more inputs, e.g., sensor data 102, and generating one or more outputs, e.g., one or more signed distance functions 108. In some examples, when used for training, sensor data 102 may be referred to as training data. While sensor data 102 is primarily discussed in relation to image data representing an image, this is not intended to be limiting, and sensor data 102 may also include other types of sensor data used for intersection area detection, such as LIDAR data, SONAR data, RADAR data, and / or similar data—for example, as generated by one or more sensors of vehicle 900 (Figures 9A-9D).

[0021] Process 100 may include generating and / or receiving sensor data 102 from one or more sensors. Sensor data 102 may be received from one or more sensors of a vehicle (for example, the vehicle 900 in Figures 9A-9C and herein). Sensor data 102 may be used by the vehicle within process 100 to detect intersection areas in order to navigate the intersection in real time or near real time. Sensor data 102 may include, but is not limited to, sensor data from any of the vehicle's sensors, including, for example, as shown in Figures 9A-9C, global navigation satellite systems (GNSS) sensors 958 (e.g., global positioning system sensors), radar sensors 960, ultrasonic sensors 962, LiDAR sensors 964, inertial measurement unit (IMU) sensors 966 (e.g., accelerometers, gyroscopes, magnetic compasses, magnetometers, etc.), microphones 976, stereo cameras 968, wide-view cameras 970 (e.g., fisheye cameras), infrared cameras 972, surround cameras 974 (e.g., 360-degree cameras), long-range and / or medium-range cameras 978, speed sensors 944 (e.g., for measuring the speed of vehicle 900), and / or other sensor types. As another example, sensor data 102 may include virtual (e.g., simulated or augmented) sensor data generated from any number of sensors of a virtual vehicle or other virtual object in a virtual (e.g., test) environment. In such an example, the virtual sensors may correspond to a virtual vehicle or other virtual object in a simulated environment (e.g., used to test, train, and / or validate neural network performance), and the virtual sensor data may represent sensor data captured by the simulated or virtual sensors in the virtual environment.As such, by using virtual sensor data, the machine learning model 104 described herein may be tested, trained, and / or validated using simulation or augmented data in a simulation environment, which may enable testing of more extreme scenarios outside of the real world environment, where the safety of such tests may be lower.

[0022] In some embodiments, the sensor data 102 may include image data representing images, image data representing videos (e.g., video snapshots), and / or sensor data representing the sensor's perceptual field (e.g., depth diagrams for LiDAR sensors, value graphs for ultrasonic sensors, etc.). If the sensor data 102 includes image data, any type of image data format may be used, including but not limited to compressed images such as JPEG (Joint Photographic Experts Group) or Luminance / Chrominance (YUV) format, compressed images such as frames resulting from compressed video formats such as H.264 / AVC (Advanced Video Coding) or H.265 / HEVC (High Efficiency Video Coding), raw images such as those derived from RCCB (Red Clear Blue), RCCC (Red Clear), or other types of image sensors, and / or other formats. In addition, in some examples, the sensor data 102 may be used in process 100 without preprocessing (e.g., in its raw or captured format), while in other examples, the sensor data 102 may undergo preprocessing (e.g., noise balancing, demosaicing, scaling, cropping, expansion, white balancing, tone curve adjustment, etc., for example, using a sensor data preprocessor (not shown)). Herein, sensor data 102 may refer to raw sensor data, preprocessed sensor data, or a combination thereof.

[0023] The sensor data 102 used for training may include the original image (for example, as captured by one or more image sensors), downsampled images, upsampled images, cropped or region of interest (ROI) images, other augmented images, and / or combinations thereof. The machine learning model 104 may be trained using the images (and / or other sensor data 102) and corresponding ground truth data. The ground truth data may include annotations, labels, masks, maps, and / or similar elements. For example, in some embodiments, the ground truth data may include a ground truth (GT) signed distance function 116. The encoder 112 may use annotations 110 to encode the GT signed distance function 116. In some embodiments, annotations 110 may include intersection area 110A and classification 110B of intersection area 110A.

[0024] Referring to footnote 110, in some examples, footnote 110 may be generated within a drawing program (e.g., an annotation program), a computer-aided design (CAD) program, a labeling program, another type of program suitable for generating footnote 110 of the GT signed distance function 116 and / or others, and / or may be hand-drawn. In any example, footnote 110 and / or the GT signed distance function 116 may be manufactured synthetically (e.g., generated from a computer model or rendering), manufactured in reality (e.g., designed and manufactured from real-world data), machine-automated (e.g., using feature analysis and learning to extract features from data and then generate labels), annotated by a person (e.g., a labeler or annotation expert defines the location of the labels), and / or a combination thereof (e.g., a person identifies the center or origin and dimensions of an area, and a machine generates polygons and / or labels for the intersection area).

[0025] The intersection area 110A may include annotations or other label types corresponding to boundary shapes—for example, polygons—that indicate the contours of the area of ​​interest at the intersection. In some examples, the intersection area may be outlined in the sensor data 102 by one or more polygons—for example, within the sensor data representation of the sensor data 102—corresponding to crosswalk areas, intersection entrance areas, intersection exit areas, obscured areas, laneless areas, intersection interior areas, partially visible areas, fully visible areas, etc. Polygons may be generated as bounding boxes. Classifications 110B may be generated for each image (or other data representation) and / or for each of the polygons in the image represented by the sensor data 102 used to train the machine learning model 104. The number of classifications 110B may correspond to the number and / or types of features trained for the machine learning model 104 to predict, or to the number and / or types of features of intersection areas in each image.

[0026] Depending on the embodiment, classification 110B may correspond to classifications or tags corresponding to feature types or intersection area classes, including but not limited to, pedestrian crossing areas, intersection entrance areas, intersection exit areas, obscured areas, laneless areas, intersection interior areas, partially visible areas, fully visible areas, and / or similar. In some embodiments, classification 110B may first correspond to intersection interior areas and / or intersection exterior areas. Intersection interior area classification may refer to intersection area 110A, which includes areas inside the intersection that can be traversed by the paths of vehicles crossing the intersection in various directions. Intersection exterior area classification may refer to intersection area 110A, which includes areas outside the intersection interior area.

[0027] Intersection areas 110A, classified as external intersection areas, may be further labeled with classifications 110B corresponding to attributes corresponding to characteristic types of intersection exit areas, including crosswalk areas, intersection entrance areas, intersection exit areas, unclear areas, laneless areas, and / or similar. Specifically, the intersection entrance attribute may correspond to an intersection area 110A where one or more vehicles are entering the corresponding intersection from various different directions. The intersection exit area may correspond to an intersection area 110A where one or more vehicles that have recently exited the intersection in various directions may be located. It should be understood that information regarding the intersection exit area may be particularly important, as a vehicle 900 must safely cross the intersection exit area itself in order to safely cross the intersection. Similarly, a crosswalk area may refer to an intersection area 110A corresponding to a crosswalk located outside the internal intersection area. Areas classified as “laneless areas” may correspond to intersection areas 110A where vehicles are not permitted to cross, such as bicycle lanes, pedestrian walkways, and / or similar. The “unclear area” attribute may correspond to an intersection area where the direction of vehicle travel is unclear. In addition, the classification 110A of the intersection interior area and intersection exterior area classes may also include one of the fully visible area and / or partially visible area attributes. In an example where classification 110B includes a fully visible area attribute or class label, the corresponding intersection area 110A may include a fully visible surface, for example, an unobstructed surface. In contrast, where classification 110B includes a partially visible area attribute or class label, the corresponding intersection area 110A may include an obstruction, for example, an obstruction such that the driving surface within the area is only partially visible in the corresponding sensor data 102. The labeling ontology described herein is for illustrative purposes only, and additional and / or alternative class labels may be used without departing from the scope of this disclosure.

[0028] As a non-limiting example, with respect to Figures 2A-2B, exemplary annotations applied to sensor data for use in ground truth generation to train a machine learning model to detect intersection areas, according to some embodiments of the present disclosure. For example, Figure 2A shows exemplary labeling (e.g., corresponding to annotation 110) of image 200A, which may be used to generate ground truth data according to the training process 100 of Figure 1. Intersection areas or regions in the image may be annotated by intersection area 110A (e.g., areas 204A, 204B, 206, 208, 210A, and 210B) and corresponding classifications (e.g., “Intersection Interior”, “Partially Visible”, “Vehicle Exit”, “Vehicle Entrance”, “Partially Visible”, “Pedestrian Crossing”, etc.). For example, intersection area 204A may be labeled using polygons and classified to have one or more attributes such as “Intersection Entrance”, and “Partially Visible”. Similarly, intersection areas 204B, 206, 208, 210A, and 210B may be labeled using polygons, so that intersection 204B may be classified as having one or more attributes such as "vehicle entrance" and "partially visible," intersection area 206 may be classified as having one or more attributes such as "pedestrian crossing" and "partially visible," intersection area 208 may be classified as having one or more attributes such as "intersection interior" and "partially visible," intersection area 210A may be classified as having one or more attributes such as "vehicle exit" and "partially visible," and intersection area 210B may be classified as having one or more attributes such as "vehicle exit" and "fully visible." In some examples, each intersection area belonging to a common class or classification may also be annotated with a polygon of a matching color (or some other visibility indication of semantic information). For example, the polygons in intersection areas 204A and 204B are both classified as vehicle entrances, so they may be the same color and / or style. Similarly, the polygons in intersection areas 210A and 210B are both classified as vehicle exits, so they may be annotated using the same color and / or style.These labeling or annotation styles may be known to system 100 as corresponding to a particular class, and this information may be used to generate encoded ground truth data for training a machine learning model 104.

[0029] Referring here to Figure 2B, Figure 2B shows another example of annotations applied to sensor data to train a machine learning model to detect intersection areas, according to several embodiments of the present invention. As shown here, intersection areas 222A-222C, 224A-224C, 226A-226B, 228A-228B, and 230 can be annotated with polygons and corresponding classifications (e.g., “Intersection Interior”, “Partially Visible”, “Vehicle Exit”, “Vehicle Entrance”, “Partially Visible”, “Crosswalk”, etc.). For example, intersection areas 222A, 222B, and 222C can be labeled using similar color and / or style polygons and can be classified as one or more of “Vehicle Entrance”, and “Partially Visible”. Similarly, intersection areas 224A, 224B, and 224C can be labeled using similar color and / or style polygons and can be classified as one or more of “Crosswalk”, “Fully Visible”, and “Partially Visible”. Intersection areas 226A and 226B may be labeled using similar color and / or style polygons and may be classified as one or more of “No Lane,” “Fully Visible,” and “Partially Visible.” Intersection areas 228A and 228B may be labeled using similar color and / or style polygons and may be classified as one or more of “Vehicle Exit,” “Fully Visible,” and “Partially Visible.” Intersection area 230 may be labeled using polygons and may be classified as one or more of “Intersection Interior” and “Partially Visible.”

[0030] Annotations may be similar visual representations of the same classification. As shown in the figure, intersection areas 222A, 222B, and 222C may be classified as vehicle exit areas. In this way, similarly classified features in the image may be annotated in a similar manner. Furthermore, note that classification 110B may also be a compound noun. Different classification labels may be represented in Figure 2B by solid lines, dashed lines, etc., to represent different classifications. Furthermore, different classification labels may be nouns and / or compound nouns. This is not intended as an limitation, and any naming convention for classification may be used to indicate differences in the classification labels (e.g., intersection areas) of features in the image.

[0031] Referring again to Figure 1, the encoder 112 may be configured to encode ground truth information corresponding to the intersection area using annotation 110. Annotation 110 may be encoded by the encoder 112 to generate ground truth data, for example, a ground truth (GT) signed distance function 116. In addition, since the spatial resolution of the sensor data 102 applied to the machine learning model 104 may differ from the output spatial resolution corresponding to the predictions of the machine learning model 104—for example, it may be greater, smaller, etc.—the ground truth information may be encoded in various formats.

[0032] For example, annotation 110 may be encoded in the form of a GT signed distance function 116, such that each intersection area class (e.g., intersection area type) is encoded in a separate GT signed distance function. Intersection areas 110A and the corresponding classifications 110B may be used by encoder 112 to generate a GT signed distance function 116, each corresponding to an intersection area type that corresponds to the area of ​​the intersection as determined from annotation 110. In some examples, each pixel in the signed distance function 116 may be encoded with a distance value corresponding to the distance of that pixel to the nearest boundary of intersection area 110A of the corresponding intersection area type (e.g., along the annotated boundary shape of the training data). In such examples, pixels inside intersection area 110A (e.g., polygons) may also be encoded with a positive distance value, and pixels outside intersection area 110A may be encoded with a negative distance value, and vice versa. Pixels forming the boundary of intersection area 110A may be encoded with a zero value for generating the GT signed distance function 116. This allows ground truth to provide a continuous-value representation of each pixel in the image, enabling efficient and accurate upsampling of the resulting detection during deployment. Furthermore, encoding each intersection area class into its own signed distance function may allow for the recall and enhancement of individual instances of intersection areas within each class with accuracy and precision.

[0033] In some examples, in addition to or instead of encoding the GT signed distance function 116, labeled intersection areas 110A and corresponding classifications 110B may be used by encoder 112 to generate a one-hot encoding map for each intersection area type. In such examples, encoder 112 may encode the intersection areas 110A using a binary representation, where pixels inside the intersection area boundary (e.g., polygon boundary) may be encoded with a value of 1, and pixels outside the intersection area boundary may be encoded with a value of 0. In some examples, the one-hot encoding map may contain multiple heat maps, each heat map corresponding to a different type of classification 110B of the intersection area determined from annotation 118. However, when using one-hot encoding maps, adjacent regions of the same intersection area type (e.g., polygons with a shared boundary) may not be individually identifiable, which may result in reduced accuracy compared to embodiments using the GT signed distance function 116. In this way, since the information can be determined using polygon annotations and semantic information, the intersection area 110A can be encoded using the GT signed distance function 116 and / or a one-hot encoding map, requiring limited labeling.

[0034] As an example, referring to Figure 3, Figure 3 shows ground truth signed distance functions 310B, 320B, 330B, 340B, and 350B for encoding intersection areas from annotations 310A, 320A, 330A, 340A, and 350A, respectively. Annotated image 310A contains intersection area 312A. In some examples, intersection area 312A may contain polygons representing the intersection area, and the corresponding signed distance function 310B may be used to encode intersection area 312A of a particular intersection area type. Intersection area 312A may be encoded in signed distance function 310B such that pixels inside intersection area 312B are encoded with positive distance values, pixels outside intersection area 312B are encoded with negative distance values, and / or pixels on the boundary of intersection area 312B are encoded with zero values. The distance value of a pixel may correspond to the distance of each pixel to the nearest boundary of an intersection area or region (for example, the intersection classification type to which the signed distance function corresponds).

[0035] Similarly, annotated image 320A includes intersection areas 322A, 324A, 326A, and 328A of a different intersection area type as represented by annotation 110. In some examples, intersection areas 322A, 324A, 326A, and 328A may include corresponding polygons representing their respective intersection areas. A corresponding signed distance function 320B may be used to encode intersection areas 322A, 324A, 326A, and 328A of a different intersection area type. Intersection areas 322A, 324A, 326A, and 328A can be encoded in the signed distance function 320B such that pixels within each intersection area 322B, 324B, 326B, and 328B are encoded with positive distance values, pixels outside intersection areas 322A, 324A, 326A, and 328A are encoded with negative distance values, and pixels on the boundaries of each intersection area 322A, 324A, 326A, and 328A are encoded with zero values. Signed distance functions 330B, 340B, and 350B can similarly be generated to represent their corresponding intersection area types by encoding the intersection areas within annotations 330A, 340A, and 350A, respectively.

[0036] After the GT signed distance function 116 has been generated for each instance of the sensor data 102 (for example, for each pixel in the sensor data 102 that contains image data), the machine learning model 104 can be trained using the GT signed distance function 116. For example, the machine learning model 104 can generate a signed distance function 108, which can be compared with the GT signed distance function 116 corresponding to each instance of the sensor data 102 using a loss function 120. As such, feedback from the loss function 120 can be used to update the parameters of the machine learning model 104 (e.g., weights and biases) taking the GT signed distance function 116 into account until the machine learning model 104 converges to an acceptable or desirable level of accuracy. Using process 100, the machine learning model 104 can be trained to accurately predict the signed distance function 108 (and / or related classifications) from the sensor data 102 using the loss function 120 and the GT signed distance function 116. In some examples, different loss functions 120 may be used to train a machine learning model 104 to predict different signed distance functions 108 for each semantic class type (e.g., intersection area type). For example, a first loss function 120 may be used to compare the signed distance function 108 with the GT signed distance function 116 for a first semantic class type, and a second loss function 120 may be used to compare the signed distance function 108 with the GT signed distance function 116 for a second semantic class type. As such, in a non-restrictive embodiment, one or more of the output channels may be trained using a different loss function 120 than that of another output channel.

[0037] The machine learning model 104 can use sensor data 102 to compute a signed distance function 108, which may be applied last to a decoder or one or more other post-processing components (at least as described further herein with respect to Figure 5) to generate a signed distance function, classification, and / or other information. Examples described herein (for example with respect to Figures 1 and 5) include the use of deep neural networks (DNNs) and, more specifically, convolutional neural networks (CNNs) as machine learning models 104, but this is not intended to be an limitation. For example, and without limitation, machine learning model 104 may include any type of machine learning model, such as linear regression, logistic regression, decision trees, support vector machines (SVMs), naive Bayes, k-nearest neighbors (Knn), K-means clustering, random forests, dimensionality reduction algorithms, gradient boosting algorithms, neural networks (e.g., autoencoders, convolutions, recurrences, perceptrons, long / short-term memory (LSTMs), Hopfield, Boltzmann, deep belief, deconvolution, adversarial generation, liquid state machines, etc.), target area detection algorithms, machine learning models using computer vision algorithms, and / or other types of machine learning models.

[0038] As an example, such as when the machine learning model 104 includes a CNN, the machine learning model 104 may include any number of layers. One or more of these layers may include an input layer. The input layer may hold values ​​related to the sensor data 102 (e.g., before or after post-processing). For example, when the sensor data 102 is an image, the input layer may hold values ​​representing the raw pixel values ​​of the image as capacities (e.g., width, height, and color channels (e.g., RGB), e.g., 32x32x3).

[0039] One or more layers may include a convolutional layer. A convolutional layer can compute the output of neurons connected to local regions in the input layer, where each neuron computes the dot product between its weights and the small region to which it is connected in the input capacitance. The result of the convolutional layer may be another capacitance having one of the dimensions based on the number of filters applied (e.g., width, height, and number of filters; for example, 32x32x12 if 12 is the number of filters).

[0040] One or more of the layers may include a rectified linear unit (ReLU) layer. The ReLU layer may have an element-wise activation function that thresholds at zero, for example, max(0,x). The resulting capacity of the ReLU layer may be the same as the capacity of the input to the ReLU layer.

[0041] One or more of the layers may include a pooling layer. The pooling layer may perform a downsampling operation along the spatial dimensions (e.g., height and width) which may result in a smaller capacitance than the input to the pooling layer (e.g., 16x16x12 from a 32x32x12 input capacitance).

[0042] One or more of the layers may contain one or more fully connected layers. Each neuron in a fully connected layer may be connected to the respective neuron in the previous capacity. A fully connected layer may compute a class score, and the resulting capacity may be 1x1x the number of classes. In some examples, a CNN may contain fully connected layers such that the output of one or more of the layers of the CNN may be provided as input to a fully connected layer of the CNN. In some examples, one or more convolutional streams may be implemented by a machine learning model 104, and some or all of the convolutional streams may contain their respective fully connected layers.

[0043] In some non-limiting embodiments, the machine learning model 104 may include a series of convolutional and max-pool layers to facilitate image feature extraction, followed by multiscale extended convolutional and upsampling layers to facilitate global context feature extraction.

[0044] Input layers, convolutional layers, pooling layers, ReLU layers, and fully connected layers are discussed herein in relation to machine learning model 104, but this is not intended to be an limitation. For example, additional or alternative layers may be used in machine learning model 104, e.g., normalization layers, SoftMax layers, and / or other layer types.

[0045] In embodiments where the machine learning model 104 includes a CNN, different orders and / or numbers of layers in the CNN may be used depending on the embodiment. In other words, the order and number of layers in the machine learning model 104 are not limited to any single architecture.

[0046] In addition, some of the layers may include parameters (e.g., weights and / or biases), such as convolutional layers and fully connected layers, while others, such as ReLU layers and pooling layers, may not. In some examples, the parameters may be learned by the machine learning model 104 during training. Furthermore, some of the layers may include additional hyperparameters (e.g., learning rate, stride, epoch, etc.), such as convolutional layers, fully connected layers, and pooling layers, while others, such as ReLU layers, may not. The parameters and hyperparameters are not limiting and may vary depending on the embodiment.

[0047] Referring here to Figure 4, each block of Method 400 as described herein includes a computation process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions may be performed by a processor that executes instructions stored in memory. Method 400 may be performed as computer-usable instructions stored on a computer storage medium. Method 400 may be provided, to name a few, as a standalone application, a service, or a hosted service (in combination with another hosted service, or as a standalone service), or as a plug-in to another product. In addition, Method 400 is described as an example with respect to process 100 in Figure 1. However, these methods may be performed additionally or alternatively by any one system, or any combination of systems, including but not limited to those described herein.

[0048] Figure 4 is a flowchart illustrating a method 400 for training a neural network to detect intersection areas, according to some embodiments of the present disclosure. Method 400 includes receiving image data representing an image indicating an intersection in block B402. For example, sensor data 102 may be received, where sensor data 102 includes image data representing an image indicating an intersection.

[0049] Method 400 includes generating annotations in block B404 that represent boundary shapes corresponding to intersection areas and corresponding semantic class labels corresponding to the semantic classes of each area. For example, annotation 110 may be generated and / or received, where the annotation represents boundary shapes corresponding to intersection areas 110A and corresponding classifications 110B of each intersection area 110A.

[0050] Method 400 involves calculating a signed distance function for each semantic class type among the semantic class types corresponding to an intersection in block B406, the signed distance function including a first signed value for a first pixel in the image inside the area, a second signed value for a second pixel in the image outside the area, and a third signed value for a third pixel in the image along the boundary of the area. For example, the GT signed distance function 116 may be calculated for each intersection area type of classification 110B corresponding to an intersection. The GT signed distance function 116 may include positive distance values ​​for pixels inside intersection area 110A, negative distance values ​​for pixels outside intersection area 110A, and zero values ​​for pixels along the boundary of intersection area 110A.

[0051] Method 400 includes training a deep neural network (DNN) using a signed distance function as ground truth data in block B408. For example, the GT signed distance function 116 may be used as ground truth data to train a machine learning model 104.

[0052] Referring here to Figure 5, which is a data flowchart illustrating an exemplary process 500 for detecting an intersection area using outputs from vehicle sensors in real time or near real time, according to some embodiments of the present disclosure. Sensor data 102 may be similar to those described herein with respect to at least Figure 1. During deployment, the sensor data 102 may be applied to a machine learning model 104 trained according to process 100 to compute a signed distance function 108 (and corresponding classification) using the training sensor data 102. As such, the machine learning model 104 may output—or regress—one or more of the signed distance functions 108, as will be described in more detail with respect to the signed distance function 108 in at least Figure 1.

[0053] The signed distance function 108 may be applied to the post-processor 502 to generate an instance partition mask 510. In some examples, the post-processor 502 may use one or more post-processing algorithms (e.g., temporal smoothing, upsampling, thresholding, clustering, mapping, etc.) to compute the instance partition mask 510. For example, the post-processor 502 may use the signed distance function 108 to generate a representation of the intersection area, which may then be used to navigate the intersection—for example, by a vehicle control component 516. The instance partition mask 510 may include the intersection area, its classification, and / or other information corresponding to the intersection (e.g., polygons corresponding to the intersection area, which may be used to generate a visualization of the intersection for debugging).

[0054] The post-processor may perform smoothing 504, upsampling 506, and / or mapping 508 to generate an instance partition mask 510. In some embodiments, the signed distance function 108 may undergo pixel-by-pixel smoothing 504 to smooth out sharp peaks and fluctuations in the pixel values. In some examples, smoothing 504 may include Gaussian smoothing of each signed distance function 108 output by the machine learning model 104. Sharp peaks and fluctuations in the signed distance function 108 may be smoothed by applying a smoothing algorithm to each signed distance function 108 to generate a corresponding smoothed signed distance function.

[0055] In some embodiments, one or more of the signed distance function 108 and / or smoothed signed distance functions may undergo upsampling 506. For example, the signed distance function 108 may be generated at a first spatial resolution (e.g., a downsampled spatial resolution) different from a second spatial resolution corresponding to the input spatial resolution of the sensor data 102. As such, upsampling 506 may decode the data from the first spatial resolution of the signed distance function 108 and convert or translate the data to the second spatial resolution of the sensor data 102. In some examples, bilinear upsampling may be applied to each signed distance function 108 and / or each smoothed signed distance function to recover the original resolution of the sensor data 102 and to generate the corresponding upsampled signed distance function. As a result, the processing speed of the machine learning model 104 may be increased while simultaneously conserving computational resources for other tasks of the system—for example, by processing reduced spatial resolution data. In addition, by increasing processing speed, the system runtime can be reduced, thereby enabling real-time or near-real-time deployment of 500 processes within the system (e.g., vehicle 900).

[0056] The signed distance function 108, the smoothed signed distance function, and / or the upsampled signed distance function can then be mapped to polygon areas representing intersection areas based on the negative and positive values ​​associated with each pixel in the corresponding signed distance function—for example, using a mapping 508 such as a mapping algorithm. Mapping 508 may include applying the signed distance function 108 to a thresholding and / or clustering algorithm. In some embodiments, a watershed clustering algorithm can be used as mapping 508 to generate instance partition maps 510 corresponding to each detected intersection area for each signed distance function 108. In such embodiments, for each intersection, pixels with areas, areas that are too high, and areas that are too low can be filtered out using predetermined distance thresholds—thereby reducing the computational resources required. The remaining pixels can be used to determine the boundaries of the associated intersection areas and / or to determine each pixel associated with a given region of an intersection having an associated classification type. Each pixel within the boundary of the polygon representing the intersection area can then be encoded with a pixel value indicating the corresponding division of the intersection area (e.g., a class label or value). The output of the post-processor 502 may also be an instance partition mask 510—or an instance thereof—for each classification type corresponding to the location (e.g., 2D coordinates) of different classification type regions of the intersection in image space.

[0057] The instance partition mask 510 can then be applied to the coordinate converter 512 to convert the 2D positions in image space to 3D world space coordinates of each detected intersection area within the instance partition mask 510. The 3D world space coordinates can be used by the vehicle 900 for route planning, control decisions, and / or other processes for safely and efficiently navigating through intersections in the world space environment. Any known method for projecting 2D positions into 3D space can be used. In some examples, the flat model assumption can be used to project positions into 3D space. In some other examples, LiDAR data for depth and / or distance estimation can be used for projection, and / or the future movement of the vehicle 900 as it crosses a road surface can be used, for example, to determine the slope or contour of the road surface and project the 2D points more accurately into 3D space (for example, to generate more accurate ground truth for training a network). In some embodiments, the machine learning model 104 can be trained on 3D data to directly predict the location of intersection areas in 3D world space.

[0058] As such, 3D world space coordinates determined from the instance partition map 510 may be used by the control components 514 of the vehicle 900 to perform one or more actions. As a non-limiting example, the perception layer of the autonomous driving software stack may update information about the environment based on intersection information, the world model manager may update the world model to reflect the position, distance, attributes, and / or other information about the intersection and other vehicles, and / or the control layer may use information (e.g., based on attributes such as waiting conditions, competition areas, vehicles with priority, the size of the intersection, the distance to the intersection, etc.) to determine controls when approaching, navigating, and / or leaving an intersection.

[0059] Referring here to Figure 6, Figure 6 shows exemplary signed distance functions (e.g., signed distance function 108 in Figure 5) that are converted to corresponding instance partition masks (e.g., instance partition mask 510 in Figure 5) using a post-processor (e.g., post-processor 502) according to some embodiments of the present disclosure. The signed distance functions 610A, 620A, 630A, and 640A include visualizations of the predicted intersection areas for individual intersection area classes. Each of the signed distance functions 610A, 620A, 630A, and 640A may be smoothed and upsampled (e.g., using smoothing 504, upsampling 506, etc.) to produce smoothed and upsampled signed distance functions 610B, 620B, 630B, and 640B, respectively. For example, pixels of the signed distance function 610A may undergo Gaussian smoothing to first generate a smoothed signed distance function by smoothing out sharp peaks and / or fluctuations in the signed distance function 610A. The smoothed signed distance function may then undergo upsampling 506 to recover the initial input image resolution of the sensor data (e.g., sensor data 102) and generate the corresponding smoothed and upsampled signed distance function 610B. Furthermore, each smoothed and upsampled signed distance function 610B, 620B, 630B, and 640B may be mapped using mapping 508 to generate the corresponding instance piecewise maps 610C, 620C, 630C, and 640C, respectively. For example, the smoothed and upsampled signed distance function 610B may undergo a watershed clustering algorithm to generate an instance piecewise mask 612C that can be overlaid on the initial image as seen in the resulting instance piecewise map 610C. In this way, intersection areas detected in an image may be represented by a set of polygons for each intersection area classification.

[0060] Referring to Figures 7A–7B, Figures 7A–7B show exemplary intersection areas predicted in 3D space based on a signed distance function computed by a neural network, according to some embodiments of the present disclosure. Figure 7A shows a visualization of predictions 700A for intersection areas 720, 712, 714, 716, 718, 720, and 722, which may be used to determine the right of way to cross an intersection for other vehicles in the intersection. Different intersection area classifications may be represented by solid lines, dashed lines, different colors, etc., to represent different classifications. This is not intended to be limiting, and any visual definition of a classification may be used to indicate differences in the classification of features (e.g., intersection areas) in an image. For example, intersection areas 712, 714, and 716 may be understood to belong to a single intersection area classification. Similarly, intersection areas 720 and 722 may belong to different intersection area classifications.

[0061] Figure 7B shows another prediction 700B of intersection areas 740, 742, 744, 746, 748, 750, and 720, which can be used to determine the right of way to cross the intersection for other vehicles by a control component (e.g., control component 514 of vehicle 900). Different intersection area classifications may be represented by solid lines, dashed lines, different colors, etc., to represent different classifications. This is not intended to be limiting, and any visual designation of a classification may be used to indicate differences in the classification of features (e.g., intersection areas) in an image for visualization, debugging, or other purposes. For example, intersection areas 742 and 744 may be understood to belong to a first intersection area classification. Similarly, intersection areas 750 and 752 may belong to a second intersection area classification, and so on. In this way, intersection areas may be easily detectable as belonging to their corresponding classifications.

[0062] Referring here to Figure 8, each block of Method 800 as described herein includes a computation process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions may be performed by a processor that executes instructions stored in memory. Method 800 may also be performed as computer-usable instructions stored on a computer storage medium. Method 800 may be provided, to name a few, as a standalone application, a service, or a hosted service (in combination with another hosted service, or as a standalone service), or as a plug-in to another product. In addition, Method 800 is described as an example with respect to Process 500 in Figure 5. However, these methods may be performed additionally or alternatively by any one system, or any combination of systems, including but not limited to those described herein.

[0063] Figure 8 is a flowchart illustrating a method 800 for detecting intersection areas according to some embodiments of the present disclosure. Method 800 includes, in block B802, applying image data representing intersections within the field of view of a vehicle's image sensor to a neural network. For example, sensor data 102 may be applied to a machine learning model 104. Sensor data 102 may represent intersections within the field of view of a vehicle's image sensor 900.

[0064] Method 800 includes, in block B804, using a neural network to compute first data representing a first signed distance function corresponding to a first intersection area class and a second signed distance function corresponding to a second intersection area class, at least in part, based on image data. For example, a machine learning model 104 may output a signed distance function 108 for each of a plurality of intersection area classes based on sensor data 102.

[0065] Method 800 includes applying a clustering algorithm to a first signed distance function and a second signed distance function in block B806. For example, a mapping 508, which includes clustering, may be performed on the signed distance function 108.

[0066] Method 800 includes, in block B808, computing a first instance partition map corresponding to a first intersection area class and a second instance partition map corresponding to a second intersection area class, at least in part on a clustering algorithm. For example, the instance partition map 510 may be computed corresponding to each intersection area class as detected in the signed distance function 108 based on the mapping 508.

[0067] Method 800 includes determining the world space locations corresponding to instances of a first intersection area class and instances of a second intersection area class in block B810, based at least partially on a first instance partition map and a second instance partition map. For example, the 3D world space locations corresponding to instances of an intersection area class can be determined by the coordinate converter 512 based on the instance partition mask 510.

[0068] Method 800 includes transmitting a second data representing a world spatial location to a vehicle planning component in block B810. For example, the 3D world spatial location of an intersection area may be transmitted to a control component 514 of the vehicle 900 for further processing.

[0069] Exemplary autonomous vehicle Figure 9A shows an exemplary autonomous vehicle 900 according to some embodiments of the present disclosure. The autonomous vehicle 900 (or referred to herein as "vehicle 900") may include, but is not limited to, passenger vehicles such as cars, trucks, buses, first responder vehicles, shuttles, electric or motorized bicycles, motorcycles, fire engines, police vehicles, ambulances, boats, construction vehicles, submarines, drones, and / or other types of vehicles (e.g., unmanned and / or carrying one or more passengers). Autonomous vehicles are generally described in terms of automation levels as defined by the National Highway Traffic Safety Administration (NHTSA), departments within 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 Vehicle" (standard number J3016-201806, published June 15, 2018; standard number J3016-201609, published September 30, 2016; and previous and future versions of this standard). Vehicle 900 may have the capability to perform functions at one or more of the autonomous driving levels from Level 3 to Level 5. For example, depending on the embodiment, Vehicle 900 may have the capability of conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5).

[0070] Vehicle 900 may include components such as the vehicle's chassis, body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other components. Vehicle 900 may include a propulsion system 950, such as an internal combustion engine, a hybrid power unit, a fully electric engine, and / or another propulsion system type. The propulsion system 950 may be connected to the vehicle's drivetrain, which may include a transmission, to enable propulsion for the vehicle 900. The propulsion system 950 may be controlled in response to receiving signals from a throttle / accelerator 952.

[0071] A steering system 954, which may include a steering wheel, may be used to steer the vehicle 900 (for example, along a desired course or route) when the propulsion system 950 is operating (for example, when the vehicle is moving). The steering system 954 may receive signals from the steering actuator 956. The steering wheel may also be an option for fully automated (level 5) functionality.

[0072] The brake sensor system 946 may be used to operate the vehicle brakes in response to receiving signals from the brake actuator 948 and / or the brake sensor.

[0073] The controller 936, which may include one or more system-on-a-chip (SoC) 904 (Figure 9C) and / or GPUs, can provide signals (e.g., expressions of commands) to one or more components and / or systems of the vehicle 900. For example, the controller can send signals to actuate the vehicle brakes via one or more brake actuators 948, actuate the steering system 954 via one or more steering actuators 956, and actuate the propulsion system 950 via one or more throttle / accelerators 952. The controller 936 may include one or more onboard (e.g., integrated) computing devices (e.g., supercomputers) that process sensor signals and output operational commands (e.g., signals representing commands) to enable autonomous driving and / or assist the driver in driving the vehicle 900. The controller 936 may include a first controller 936 for autonomous driving functions, a second controller 936 for functional safety functions, a third controller 936 for artificial intelligence functions (e.g., computer vision), a fourth controller 936 for infotainment functions, a fifth controller 936 for redundancy in emergency situations, and / or other controllers. In some examples, a single controller 936 may handle two or more of the aforementioned functions, and two or more controllers 936 may handle a single function, and / or any combination thereof.

[0074] The controller 936 can provide signals for controlling one or more components and / or systems of the vehicle 900 in response to sensor data (e.g., sensor inputs) received from one or more sensors. Sensor data may be received from, for example and without limitation, global navigation satellite system sensors 958 (e.g., global positioning system sensors), RADAR sensors 960, ultrasonic sensors 962, LIDAR sensors 964, inertial measurement unit (IMU) sensors 966 (e.g., accelerometers, gyroscopes, magnetic compasses, magnetometers, etc.), microphones 996, stereo cameras 968, wide-view cameras 970 (e.g., fisheye cameras), infrared cameras 972, surround cameras 974 (e.g., 360-degree cameras), long-range and / or medium-range cameras 998, speed sensors 944 (e.g., for measuring the speed of a vehicle 900), vibration sensors 942, steering sensors 940, brake sensors (e.g., as part of a brake sensor system 946), and / or other sensor types.

[0075] One or more of the controllers 936 may receive inputs (represented, for example, by input data) from the instrument cluster 932 of the vehicle 900 and provide outputs (represented, for example, by output data, display data, etc.) via a human-machine interface (HMI) display 934, an audible annunciator, a loudspeaker, and / or other components of the vehicle 900. The outputs may include information such as vehicle velocity, speed, time, map data (e.g., HD map 922 in Figure 9C), location data (e.g., the location of the vehicle 900, such as on a map), direction, the location of other vehicles (e.g., occupied grid), and information about objects and the status of objects as perceived by the controller 936. For example, the HMI display 934 may display information regarding the presence of one or more objects (e.g., road signs, warning signs, changes in traffic signals, etc.) and / or driving operations that the vehicle has performed, is performing, or will perform (e.g., changing lanes now, exiting exit 34B within 3.22 km (2 miles), etc.).

[0076] The vehicle 900 further includes a network interface 924 that can communicate over one or more networks using one or more wireless antennas 926 and / or a modem. For example, the network interface 924 may have the capability to communicate over LTE, WCDMA®, UMTS, GSM, CDMA2000, etc. The wireless antennas 926 can also enable communication between objects in the environment (e.g., vehicles, mobile devices, etc.) using local area networks such as Bluetooth®, Bluetooth LE, Z-Wave, ZigBee, and / or low-power wide-area networks (LPWAN) such as LoRaWAN, SigFox.

[0077] Figure 9B shows examples of camera positions and fields of view of the exemplary autonomous vehicle 900 of Figure 9A according to several embodiments of the present disclosure. The cameras and their respective fields of view are exemplary embodiments and are not intended to limit the scope. For example, additional and / or alternative cameras may be included, and / or cameras may be placed in different positions on the vehicle 900.

[0078] The camera type may include, but is not limited to, a digital camera that can be used with components and / or systems of the vehicle 900. The camera may operate at Automotive Safety Integrity Level (ASIL) B and / or other ASILs. Depending on the embodiment, the camera type may have the capability of any image capture rate, such as 60 frames per second (fps), 920 fps, 240 fps, etc. The camera may have the capability to use a roll shutter, a global shutter, another type of shutter, or a combination thereof. In some examples, the color filter array may include an RCCC (red clear clear clear) color filter array, an RCCB (red clear clear blue) color filter array, an RBGC (red blue green clear) 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, clear pixel cameras, such as cameras having RCCC, RCCB, and / or RBGC color filter arrays, may be used in efforts to increase light sensitivity.

[0079] In some applications, one or more cameras may be used to perform advanced driver assistance system (ADAS) functions (e.g., as part of a redundant or fail-safe design). For example, a multi-function mono-camera may be installed to provide functions including lane departure warning, traffic sign assist, and intelligent headlamp control. One or more cameras (e.g., all cameras) may simultaneously record and provide image data (e.g., video).

[0080] One or more of the cameras may be mounted in custom-designed (3D-printed) mounting parts to eliminate stray light and reflections from inside the vehicle (e.g., reflections from the dashboard reflected in the windshield mirror) that may interfere with the camera's image data capture capability. Referring to side mirror mounting parts, the side mirror parts may be custom 3D-printed so that the camera mounting plate conforms to the shape of the side mirror. In some examples, the camera may be integrated within the side mirror. For side-view cameras, the camera may also be integrated within four struts located at each corner of the cabin.

[0081] A camera having a field of view that includes a portion of the environment in front of the vehicle 900 (e.g., a forward-facing camera) may be used for surround view to help identify the forward path and obstacles and, with the help of one or more controllers 936 and / or control SoCs, to help provide information essential for generating an occupied grid and / or determining a preferred vehicle path. The forward-facing camera may also be used to perform many of the same ADAS functions as LIDAR, including emergency braking, pedestrian detection, and collision avoidance. The forward-facing camera may also be used for ADAS functions and systems, including other functions such as Lane Departure Warning ("LDW"), Autonomous Cruise Control ("ACC"), and / or traffic sign recognition.

[0082] Various cameras may be used in forward-facing configurations, including, for example, a monocular camera platform that includes a CMOS (complementary metal oxide semiconductor) color imaging device. Another example may be a wide-view camera 970, which can be used to capture objects entering the view from the surroundings (e.g., pedestrians, crossing traffic, or bicycles). Although only one wide-view camera is shown in Figure 9B, any number of wide-view cameras 970 may be present in the vehicle 900. In addition, long-range cameras 998 (e.g., a long-view stereo camera pair) may be used for depth-based object detection, particularly for objects for which the neural network has not yet been trained. Long-range cameras 998 may also be used for object detection and classification, as well as basic object tracking.

[0083] One or more stereo cameras 968 may also be included in the forward-facing configuration. The stereo camera 968 may include an integrated control unit with an expandable processing unit that may provide a programmable logic (FPGA) and a multi-core microprocessor with an integrated CAN or Ethernet® interface on a single chip. Such a unit may be used to generate a 3D map of the vehicle's environment, including distance estimates of all points in the image. An alternative stereo camera 968 may include a compact stereo vision sensor that includes two camera lenses (one on the left and one on the right) and an image processing chip that can measure the distance from the vehicle to an object and activate autonomous emergency braking and lane departure warning functions using the generated information (e.g., metadata). Other types of stereo cameras 968 may be used in addition to or instead of those described herein.

[0084] A camera having a field of view including a portion of the environment to the sides of the vehicle 900 (e.g., a side-view camera) may be used for surround view, providing information used to create and update the occupancy grid and generate side impact collision warnings. For example, surround cameras 974 (e.g., four surround cameras 974 as shown in Figure 9B) may be positioned on the vehicle 900. The surround cameras 974 may include wide-view cameras 970, fisheye cameras, 360-degree cameras, and / or similar. For example, four fisheye cameras may be positioned in front of, behind, and to the sides of the vehicle. In an alternative configuration, the vehicle may use three surround cameras 974 (e.g., left, right, and rear) and utilize one or more other cameras (e.g., forward-facing cameras) as a fourth surround view camera.

[0085] A camera having a field of view that includes a portion of the environment behind the vehicle 900 (e.g., a rear-view camera) may be used for parking assistance, surround view, rear collision warning, and creation and updating of the occupancy grid. A wide variety of cameras may be used, including, but not limited to, cameras suitable as forward-facing cameras (e.g., long-range and / or medium-range cameras 998, stereo camera 968), infrared camera 972, etc., as described herein.

[0086] Figure 9C is a block diagram of an exemplary system architecture of the exemplary autonomous vehicle 900 of Figure 9A, according to some embodiments of the present disclosure. It should be understood that this and other arrangements described herein are merely illustrative. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, groupings of functions, etc.) may be used in addition to or instead of those shown, and some elements may be omitted together. Furthermore, many of the elements described herein are functional entities that can be implemented as individual or distributed components or in combination with other components, and in any appropriate combination and location. Various functions described herein as being performed by entities may be performed by hardware, firmware, and / or software. For example, various functions may be performed by a processor that executes instructions stored in memory.

[0087] Each component, feature, and system of vehicle 900 in Figure 9C is illustrated as being connected via bus 902. Bus 902 may include a Controller Area Network (CAN) data interface (or referred to as the "CAN bus"). CAN may also be a network within vehicle 900 used to help control various features and functions of vehicle 900, such as the operation of brakes, acceleration, steering, windshield wipers, etc. The CAN bus may be configured to have dozens or hundreds of nodes, each having its own unique identifier (e.g., CAN ID). The CAN bus may be read to find steering angle, ground speed, engine revolutions per minute (RPM), button position, and / or other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0088] Bus 902 is described herein as a CAN bus, but this is not intended to limit it. For example, FlexRay and / or Ethernet® may be used in addition to, or as an alternative to, a CAN bus. In addition, a single line is used to represent bus 902, but this is not intended to limit it. There may be any number of buses 902, which may include, for example, one or more CAN buses, one or more FlexRay buses, one or more Ethernet® buses, and / or one or more other types of buses using different protocols. In some examples, two or more buses 902 may be used to perform different functions and / or for redundancy. For example, a first bus 902 may be used for collision avoidance functions, and a second bus 902 may be used for operation control. In any example, each bus 902 may communicate with any of the components of vehicle 900, and two or more buses 902 may communicate with the same component. In some examples, each SoC904, each controller936, and / or each computer within the vehicle may have access to the same input data (e.g., input from sensors in the vehicle900) and may be connected to a common bus such as a CAN bus.

[0089] The vehicle 900 may include one or more controllers 936, such as those described herein with respect to Figure 9A. The controllers 936 may be used for a variety of functions. The controllers 936 may be connected to any of the various other components and systems of the vehicle 900 and may be used for the control of the vehicle 900, the artificial intelligence of the vehicle 900, infotainment for the vehicle 900, and / or the like.

[0090] Vehicle 900 may include a system-on-a-chip (SoC) 904. The SoC 904 may include a CPU 906, a GPU 908, a processor 910, a cache 912, an accelerator 914, a data store 916, and / or other components and features not shown. The SoC 904 may be used to control vehicle 900 in various platforms and systems. For example, the SoC 904 may be coupled in a system (e.g., a system of vehicle 900) that has an HD map 922 that can obtain map refreshes and / or updates via a network interface 924 from one or more servers (e.g., server 978 in Figure 9D).

[0091] The CPU906 may include a CPU cluster or CPU complex (also referred to as "CCPLEX"). The CPU906 may include multiple cores and / or L2 caches. For example, in some embodiments, the CPU906 may include eight cores in a coherent multiprocessor configuration. In some embodiments, the CPU906 may include four dual-core clusters, each cluster having its own dedicated L2 cache (e.g., 2MBL2 cache). The CPU906 (e.g., CCPLEX) may be configured to support concurrent cluster operation, allowing any combination of the CPU906 clusters to be active at any given time.

[0092] The CPU906 can implement power management capabilities that include one or more of the following features: individual hardware blocks may be automatically clock-gated when idle to conserve dynamic power; each core clock may be gated when a core is not actively executing instructions by executing WFI / WFE instructions; each core may be independently power-gated; each core cluster may be independently clock-gated when all cores are clock-gated or power-gated; and / or each core cluster may be independently power-gated when all cores are power-gated. The CPU906 can further implement enhanced algorithms for managing power states, where acceptable power states and expected wake-up times are specified, and the hardware / microcode determines the best power state to input to the cores, clusters, and CCPLEX. The processing core may support a simplified power state input sequence in software where the work is offloaded to the microcode.

[0093] The GPU908 may include an integrated GPU (or, as referred to herein, "iGPU"). The GPU908 may be programmable and efficient for parallel workloads. In some embodiments, the GPU908 may be able to use an enhanced tensor instruction set. The GPU908 may include one or more streaming microprocessors, each of which may include an L1 cache (e.g., an L1 cache with a storage capacity of at least 96KB), and two or more of the streaming microprocessors may share a cache (e.g., an L2 cache with a storage capacity of 512KB). In some embodiments, the GPU908 may include at least eight streaming microprocessors. The GPU908 may be able to use a Computation Application Programming Interface (API). In addition, the GPU908 may be able to use one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0094] The GPU908 can be power-optimized for optimal performance in automotive and embedded use cases. For example, the GPU908 can be manufactured on a FinFET (Fin field-effect transistor). However, this is not intended to be a limitation, and the GPU908 can be manufactured using other semiconductor manufacturing processes. Each streaming microprocessor can incorporate several mixed-precision processing cores divided into multiple blocks. Not limited to, for example, 64 PF32 cores and 32 PF64 cores may be divided into four processing blocks. In such an example, each processing block may be allocated 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, 2 mixed-precision NVIDIA tensor cores for deep learning matrix operations, an L0 instruction cache, a warp scheduler, a dispatch unit, and / or a 64KB register file. In addition, the streaming microprocessor may include independent parallel integer and floating-point data paths to provide efficient execution of workloads with a mixture of computation and addressing operations. A streaming microprocessor may include independent thread scheduling capabilities to enable finer-grained synchronization and coordination between concurrent threads. A streaming microprocessor may also include a combined L1 data cache and shared memory unit to simplify programming while improving performance.

[0095] In some examples, the GPU908 may include high-bandwidth memory (HBM) and / or a 16GB HBM2 memory subsystem to provide a peak memory bandwidth of 900 GB / s. In some examples, in addition to or instead of HBM memory, synchronous graphics random-access memory (SGRAM), such as graphics double data rate type five synchronous random-access memory (GDDR5), may be used.

[0096] The GPU908 can incorporate unified memory technology, including access counters, to enable more precise movement of memory pages to the processor that most frequently accesses them, thereby improving the efficiency of shared memory ranges between processors. In some examples, address translation service (ATS) support may be used to allow the GPU908 to directly access the CPU906 page table. In such examples, when the GPU908 memory management unit (MMU) experiences a miss, an address translation request may be sent to the CPU906. In response, the CPU906 can look up its page table for virtual-to-real-address mapping and send the translation back to the GPU908. As such, unified memory technology can enable a single, unified virtual address space for both the CPU906 and GPU908 memory, thereby simplifying GPU908 programming and porting of applications to the GPU908.

[0097] In addition, the GPU908 may include an access counter that can record how often the GPU908 accesses the memory of other processors. The access counter can help ensure that memory pages are moved to the physical memory of the processor that accesses that page most frequently.

[0098] The SoC904 may include any number of caches 912, including those described herein. For example, the cache 912 may include an L3 cache available to both the CPU 906 and the GPU 908 (e.g., connected to both the CPU 906 and the GPU 908). The cache 912 may include a write-back cache that can record line states, for example, by using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). The L3 cache may include 4 MB or more, depending on the embodiment, although a smaller cache size may be used.

[0099] The SoC904 may include an arithmetic logic unit (ALU) that can be used to perform processing for any of the various tasks or operations of the vehicle 900 (for example, a processing DNN). In addition, the SoC904 may include a floating-point unit (FPU) (or other mass coprocessor or numerical coprocessor type) for performing mathematical operations within the system. For example, the SoC104 may include one or more FPUs integrated as execution units within the CPU906 and / or GPU908.

[0100] The SoC904 may include one or more accelerators 914 (e.g., a hardware accelerator, a software accelerator, or a combination thereof). For example, the SoC904 may include a hardware acceleration cluster that may include an optimized hardware accelerator and / or a large on-chip memory. The large on-chip memory (e.g., 4MB of SRAM) may enable the hardware acceleration cluster to accelerate neural networks and other computations. The hardware acceleration cluster may be used to complement the GPU908 and to offload some of the GPU908's tasks (e.g., to free up more cycles of the GPU908 to perform other tasks). As an example, accelerator 914 may be used for target workloads that are sufficiently stable to be suitable for acceleration (e.g., perception, convolutional neural networks (CNNs), etc.). In this specification, the term "CNN" may include all types of CNNs, including region-based or regional convolutional neural networks (RCNNs) and fast RCNNs (for example, as used for object detection).

[0101] The accelerator 914 (e.g., a hardware acceleration cluster) may include a deep learning accelerator (DLA). The DLA may include one or more tensor processing units (TPUs) that can be configured to provide an additional 10 trillion operations per second for deep learning applications and inference. The TPU may also be an accelerator configured and optimized to perform image processing functions (e.g., CNN, RCNN, etc.). The DLA may further be optimized for a specific set of neural network types and floating-point operations, as well as for inference. The design of the DLA can provide more performance per millisecond than a general-purpose GPU and significantly exceed the performance of a CPU. The TPU can perform several functions, including, for example, single-instance convolutional functions supporting INT8, INT16, and FP16 data types for both features and weights, as well as post-processing functions.

[0102] DLA can quickly and efficiently run neural networks, particularly CNNs, on processed or unprocessed data for any of a variety of functions, including but not limited to: CNNs for object recognition and detection using data from camera sensors; CNNs for distance estimation using data from camera sensors; CNNs for emergency vehicle detection, identification, and detection using data from microphones; CNNs for facial recognition and vehicle owner identification using data from camera sensors; and / or CNNs for security and / or safety-related events.

[0103] DLA can perform any function of GPU908, and by using inference accelerators, for example, a designer can target either DLA or GPU908 for any function. For example, a designer can focus on CNN and floating-point arithmetic processing on DLA, and leave other functions to GPU908 and / or other accelerators 914.

[0104] The accelerator 914 (for example, a hardware accelerator cluster) may include a programmable vision accelerator (PVA), which may be referred to herein as a computer vision accelerator. A 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. A PVA can provide a balance between performance and flexibility. For example, each PVA may, but is not limited to, any number of reduced instruction set computer (RISC) cores, direct memory access (DMA), and / or any number of vector processors.

[0105] A RISC core can interact with an image sensor (for example, the image sensor of one of the cameras described herein), an image signal processor, and / or similar devices. Each RISC core may contain any amount of memory. Depending on the embodiment, a RISC core may use one of several protocols. In some examples, a RISC core can run a real-time operating system (RTOS). A RISC core may be implemented using one or more integrated circuit devices, application-specific integrated circuits (ASICs), and / or memory devices. For example, a RISC core may include an instruction cache and / or tightly coupled RAM.

[0106] DMA can enable PVA components to access system memory independent of the CPU906. DMA can support any number of features used to bring optimizations to the PVA, including but not limited to supporting multidimensional addressing and / or circular addressing. In some examples, DMA can support up to six or more dimensions of addressing, which may include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0107] A vector processor may also be a programmable processor that can be designed to efficiently and flexibly execute the programming of computer vision algorithms and provide signal processing capabilities. In some examples, a PVA may include a PVA core and two vector processing subsystem partitions. The PVA core may include a processor subsystem, a DMA engine (e.g., two DMA engines), and / or other peripherals. The vector processing subsystem can act as the primary processing engine of the PVA and may include a vector processing unit (VPU), an instruction cache, and / or vector memory (e.g., VMEM). The VPU core may include a digital signal processor, such as a single-instruction, multiple-data (SIMD), or very-long instruction word (VLIW) digital signal processor. A combination of SIMD and VLIW can increase throughput and speed.

[0108] Each vector processor may include an instruction cache and be linked to dedicated memory. As a result, in some examples, each vector processor may be configured to run independently of other vector processors. In other examples, the vector processors included in a particular PVA may be configured to use data parallelism. For example, in some embodiments, multiple vector processors included in a single PVA can run the same computer vision algorithm, but on different regions of an image. In other examples, the vector processors included in a particular PVA can run different computer vision algorithms simultaneously on the same image, or even run different algorithms sequentially on the image or parts of an image. In particular, any number of PVAs may be included in a hardware acceleration cluster, and any number of vector processors may be included in each PVA. In addition, a PVA may include additional error correction code (ECC) memory to enhance overall system safety.

[0109] The accelerator 914 (for example, a hardware accelerator cluster) may include a computer vision network on-chip and SRAM to provide high-bandwidth, low-latency SRAM for the accelerator 914. In some examples, the on-chip memory may include at least 4 MB of SRAM consisting of eight field-configurable memory blocks, which may be accessible by both the PVA and DLA, for example, and not limited to. 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 can access the memory via a backbone that provides the PVA and DLA with high-speed access to the memory. The backbone may include a computer vision network on-chip that interconnects the PVA and DLA to the memory (for example, using an APB).

[0110] A computer vision network on-chip may include an interface that determines whether both the PVA and DLA are activatable and enable signals before any control signals / addresses / data are transmitted. Such an interface can provide separate phases and separate channels for transmitting control signals / addresses / data, as well as burst-type communication for continuous data transfer. This type of interface may conform to ISO 26262 or IEC 61508 standards, but other standards and protocols may be used.

[0111] In some embodiments, the SoC904 may include a real-time ray tracing hardware accelerator, as described in U.S. Patent Application No. 16 / 101,232, filed August 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the location and size of objects (e.g., in a world model) to generate real-time visualization simulations for RADAR signal interpretation, acoustic propagation synthesis and / or analysis, SONAR system simulation, general wave propagation simulation, comparison to LIDAR data for localization and / or other functions, and / or other uses. In some embodiments, one or more tree traversal units (TTUs) may be used to perform one or more ray tracing-related operations.

[0112] The accelerator 914 (e.g., a hardware accelerator cluster) has diverse applications for autonomous driving. The PVA may also be a programmable vision accelerator that can be used in critical processing stages in ADAS and autonomous vehicles. The PVA's capabilities are suitable for areas of algorithms requiring predictable processing at low power and low latency. In other words, the PVA performs well in semi-high density or high density typical computations, even on small data sets, where predictable execution time is required along with low latency and low power. Therefore, because the PVA is efficient in object detection and integer computation, in relation to a platform for autonomous vehicles, the PVA is designed to run classic computer vision algorithms.

[0113] For example, according to one embodiment of this technology, PVA is used to perform computer stereo vision. While semi-global matching-based algorithms may be used in some examples, this is not intended to be a limitation. Numerous applications for Level 3-5 autonomous driving require motion estimation / stereo matching on the fly (e.g., SFM (structure from motion), pedestrian recognition, lane detection, etc.). PVA can perform computer stereo vision functions with input from two monocular cameras.

[0114] In some applications, PVA can be used to perform high-density optical flow by processing raw RADAR data (e.g., using 4D Fast Fourier Transform) to provide processed RADAR data. In other applications, PVA is used for flight depth processing, for example, by processing raw flight data to provide processed flight data.

[0115] DLA can be used to run any type of network to enhance control and driving safety, for example, a neural network that outputs a confidence value for each object detection. Such confidence values ​​can be interpreted as probabilities or as providing the relative "weight" of each detection compared to other detections. This confidence value allows the system to make further decisions about which detections should be considered true positives rather than false positives. For example, the system can set a confidence threshold and consider only detections that exceed the threshold as true positives. In an automatic emergency braking (AEB) system, a false positive detection would cause the vehicle to automatically apply the emergency brakes, which is obviously undesirable. Therefore, only the most confident detection should be considered as a trigger for the AEB. DLA can run a neural network that regresses on the confidence values. The neural network can accept at least a subset of parameters as its input, such as bounding box dimensions, ground plane estimation acquired (e.g., from another subsystem), vehicle orientation, distance, inertial measurement unit (IMU) sensor output correlated with 3D position estimation of an object acquired from the neural network and / or other sensors (e.g., LIDAR sensor 964 or RADAR sensor 960), and others.

[0116] The SoC904 may include a data store 916 (for example, memory). The data store 916 may also be the on-chip memory of the SoC904 and can store neural networks that will run on the GPU and / or DLA. In some examples, the data store 916 may have a capacity large enough to store multiple instances of the neural network for redundancy and safety. The data store 912 may include an L2 or L3 cache 912. References to the data store 916 may include references to memory associated with the PVA, DLA, and / or other accelerators 914, as described herein.

[0117] The SoC904 may include one or more processors 910 (e.g., integrated processors). The processors 910 may include a boot and power management processor, which may be a dedicated processor and subsystem for handling boot power and management capabilities and associated security enforcement. The boot and power management processor may also be part of the SoC904 boot sequence and can provide runtime power management services. The boot power and management processor may provide clock and voltage programming, assistance with system low-power state transitions, management of SoC904 thermal and temperature sensors, and / or management of SoC904 power states. Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to temperature, and the SoC904 may use the ring oscillators to detect the temperatures of the CPU 906, GPU 908, and / or accelerator 914. If the temperature is determined to have exceeded a threshold, the boot and power management processor may enter a temperature fault routine, placing the SoC904 into a lower power state and / or putting the vehicle 900 into chauffeur safe shutdown mode (for example, safely shutting down the vehicle 900).

[0118] The processor 910 may further include a set of integrated processors that can perform the functions of an audio processing engine. The audio processing engine may also be an audio subsystem that enables full hardware support for multi-channel audio through multiple interfaces and a wide and flexible range of audio I / O interfaces. In some examples, the audio processing engine is a dedicated processor core having a digital signal processor with dedicated RAM.

[0119] The processor 910 may further include an always-on processor engine that can provide the necessary hardware features to support low-power sensor management and wake use cases. The always-on processor engine may include a processor core, tightly coupled RAM, support peripherals (e.g., timer and interrupt controllers), various I / O controller peripherals, and routing logic.

[0120] The processor 910 may further include a safety cluster engine, which includes a dedicated processor subsystem for handling safety management in 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 function as a single core with comparison logic to detect any differences between their operations.

[0121] The processor 910 may further include a real-time camera engine, which may include a dedicated processor subsystem for handling real-time camera management.

[0122] The processor 910 may further include a high dynamic range signal processor, which may include an image signal processor, a hardware engine that is part of the camera processing pipeline.

[0123] The processor 910 may include a video image synthesizer, which may also be a processing block (for example, implemented on a microprocessor) that implements post-video processing functions required by the video playback application to produce the final image for the player window. The video image synthesizer can perform lens distortion correction on the wide-view camera 970, the surround camera 974, and / or the in-cabin surveillance camera sensors. The in-cabin surveillance camera sensors are preferably monitored by a neural network running on another instance of the advanced SoC, configured to identify and appropriately respond to in-cabin events. The in-cabin system can activate cellular services and perform lip-reading to make phone calls, transcribe emails, change the vehicle's destination, activate or change the vehicle's infotainment system and settings, or provide voice-activated web surfing. Certain functions are available to the driver only when operating in autonomous mode and are otherwise disabled.

[0124] A video image synthesizer may include enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, if motion occurs in the video, noise reduction reduces the weight of information provided by adjacent frames and appropriately weights the spatial information. If the image or part of the image does not contain motion, the temporal noise reduction performed by the video image synthesizer can use information from previous images to reduce noise in the current image.

[0125] The video image synthesizer can also be configured to perform stereo rectification on the input stereo lens frame. Furthermore, the video image synthesizer can be used for user interface compositing when the operating system desktop is in use, so that the GPU908 is not required to continuously render new surfaces. Even when the GPU908 is powered on and actively performing 3D rendering, the video image synthesizer can be used to offload the GPU908 to improve performance and responsiveness.

[0126] The SoC904 may further include a Mobile Industry Processor Interface (MIPI) camera serial interface, a high-speed interface, and / or a video input block that can be used for camera and associated pixel input functions to receive video and input from a camera. The SoC904 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 committed to a specific role.

[0127] The SoC904 may further include a wide range of peripheral interfaces to enable communication with peripheral devices, audio codecs, power management, and / or other devices. The SoC904 may be used to process data from cameras (connected, for example, via Gigabit Multimedia Serial Link and Ethernet®), sensors (e.g., LiDAR sensor 964, RADAR sensor 960, etc., which may be connected via Ethernet®), data from bus 902 (e.g., vehicle speed, steering wheel position, etc.), and data from GNSS sensor 958 (connected, for example, via Ethernet® or CAN bus). The SoC904 may further include a dedicated high-performance mass storage controller, which may include its own DMA engine and may be used to free up CPU 906 from routine data management tasks.

[0128] The SoC904 may also be an inter-terminal platform with a flexible architecture that extends to automation levels 3-5, thereby providing a comprehensive functional safety architecture that leverages and efficiently uses computer vision and ADAS techniques for diversity and redundancy, and, together with deep learning tools, provides a platform for a flexible, reliable driving software stack. The SoC904 can be faster, more reliable, more energy-efficient, and more space-efficient than conventional systems. For example, when the accelerator 914 is coupled with the CPU 906, the GPU 908, and the data store 916 can provide a fast and efficient platform for autonomous vehicles at levels 3-5.

[0129] Therefore, this technology brings capabilities and functionality that cannot be achieved by conventional systems. For example, computer vision algorithms can be executed on a CPU, 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 cannot meet the performance requirements of many computer vision applications, such as those related to execution time and power consumption. Specifically, many CPUs cannot execute real-time complex object detection algorithms, which are required for in-vehicle ADAS applications and actual Level 3-5 autonomous vehicles.

[0130] In contrast to conventional systems, by providing a CPU complex, a GPU complex, and a hardware acceleration cluster, the technologies described herein enable multiple neural networks to run simultaneously and / or sequentially, and the results to be combined to enable Level 3–5 autonomous driving capabilities. For example, a DLA or a CNN running on a dGPU (e.g., GPU920) may include text and word recognition, enabling a supercomputer to read and understand traffic signs, including signs for which the neural network has not been specifically trained. The DLA may further include a neural network capable of identifying, interpreting, and providing a semantic understanding of signs and passing that semantic understanding to a route planning module running on the CPU complex.

[0131] As another example, multiple neural networks may run simultaneously, as required for Level 3, 4, or 5 driving. For instance, a warning sign consisting of a flashing light and the text "Caution: Flashing light indicates frozen conditions" may be interpreted independently or collectively by several neural networks. The sign itself may be identified as a traffic sign by a first deployed neural network (e.g., a trained neural network), and the text "Flashing light indicates frozen conditions" may be interpreted by a second deployed neural network that informs the vehicle's route planning software (preferably running on a CPU complex) that frozen conditions are present when flashing light is detected. The flashing light may be identified by informing the vehicle's route planning software of the presence (or absence) of the flashing light, and by operating a third deployed neural network through multiple frames. All three neural networks can run simultaneously within the DLA and / or on the GPU908, for example.

[0132] In some applications, a CNN for facial recognition and vehicle owner identification can use data from camera sensors to identify the presence of the legitimate driver and / or owner of vehicle 900. An always-on sensor processing engine may be used to unlock the vehicle and turn on the lights when the owner approaches the driver's side door, and in security mode, to stop the vehicle when the owner leaves the vehicle. In this way, SoC904 provides security against theft and / or vehicle hijacking.

[0133] In another example, a CNN for emergency vehicle detection and identification can detect and identify emergency vehicle sirens using data from microphone 996. In contrast to conventional systems that use a general classifier to detect sirens and manually extract features, SoC904 uses a CNN for classifying environmental and urban sounds, as well as for classifying visual data. In a preferred embodiment, a CNN running on DLA is trained to identify the relative terminal velocity of emergency vehicles (for example, by using the Doppler effect). The CNN may also be trained to identify emergency vehicles specific to the local area in which the vehicle is operating, as identified by GNSS sensor 958. Thus, for example, when operating in Europe, the CNN would attempt to detect European sirens, and when in the United States, the CNN would attempt to identify only North American sirens. After an emergency vehicle is detected, a control program may be used, with the assistance of ultrasonic sensor 962, to perform emergency vehicle safety routines such as slowing down the vehicle, stopping it at the side of the road, parking the vehicle, and / or idling the vehicle until the emergency vehicle has passed.

[0134] The vehicle may include a CPU 918 (e.g., a separate CPU, or dCPU) which can be connected to the SoC 904 via a high-speed interconnect (e.g., PCIe). The CPU 918 may include, for example, an x86 processor. The CPU 918 may be used to perform any of a variety of functions, including, for example, mediating the consequences of a potential mismatch between ADAS sensors and the SoC 904, and / or monitoring the status and condition of the controller 936 and / or the infotainment SoC 930.

[0135] Vehicle 900 may include a GPU 920 (e.g., a separate GPU, or dGPU) which can be connected to SoC 904 via a high-speed interconnect (e.g., NVIDIA NVLINK). The GPU 920 can provide additional artificial intelligence capabilities, such as by running redundant and / or different neural networks, and may be used to train and / or update neural networks based on input from sensors in Vehicle 900 (e.g., sensor data).

[0136] Vehicle 900 may further include a network interface 924 which may include one or more wireless antennas 926 (e.g., one or more wireless antennas for different communication protocols, such as a cellular antenna and a Bluetooth antenna). The network interface 924 may be used to enable wireless connectivity to a cloud via the Internet (e.g., a server 978 and / or other network devices), to other vehicles, and / or to computing devices (e.g., passenger client devices). To communicate with other vehicles, a direct link may be established between two vehicles, and / or an indirect link may be established (e.g., via a network and via the Internet). The direct link may be provided using a vehicle-to-vehicle communication link. The vehicle-to-vehicle communication link can provide vehicle 900 information about vehicles in close proximity to vehicle 900 (e.g., vehicles in front of, beside, and / or behind vehicle 900). This function may also be part of the vehicle 900's joint adaptive cruise control function.

[0137] The network interface 924 may include an SoC that provides modulation and demodulation functions and enables the controller 936 to communicate over a wireless network. The network interface 924 may include a radio frequency front end for up-conversion from baseband to radio frequency and down-conversion from radio frequency to baseband. Frequency conversion can be performed through well-known processes and / or using a superheterodyne process. In some examples, the radio frequency front end functionality may be provided by a separate chip. The network interface may include wireless functionality for communication over LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-Wave, ZigBee, LoRaWAN, and / or other wireless protocols.

[0138] The vehicle 900 may further include a data store 928 which may include storage outside the chip (for example, outside the SoC 904). The data store 928 may include one or more storage elements, including RAM, SRAM, DRAM, VRAM, flash, hard disk, and / or other components and / or devices capable of storing at least one bit of data.

[0139] The vehicle 900 may further include GNSS sensors 958. The GNSS sensors 958 (e.g., GPS, assisted GPS sensors, differential GPS (DGPS) sensors, etc.) assist in mapping, perception, occupy grid generation, and / or route planning functions. Any number of GNSS sensors 958 may be used, including, but not limited to, GPS using a USB connector with Ethernet® to a serial (RS-232) bridge.

[0140] Vehicle 900 may further include a RADAR sensor 960. The RADAR sensor 960 may be used by vehicle 900 for long-range vehicle detection, even in darkness and / or severe weather conditions. The RADAR functional safety level may be ASIL B. In some examples, the RADAR sensor 960 may use CAN and / or bus 902 for control and to access object tracking data (for example, to transmit data generated by the RADAR sensor 960) using Ethernet® access for accessing raw data. A wide variety of RADAR sensor types may be used. For example, and without limitation, the RADAR sensor 960 may be suitable for front, rear, and side RADAR use. In some examples, a pulsed Doppler RADAR sensor may be used.

[0141] The RADAR sensor 960 may include different configurations, such as long-range with a narrow field of view, short-range with a wide field of view, and short-range side coverage. In some examples, the long-range RADAR may be used for adaptive cruise control functions. The long-range RADAR system can provide a wide field of view achieved by two or more independent scans, such as within a range of 250m. The RADAR sensor 960 can help distinguish between static and moving objects and may be used by ADAS systems for emergency brake assist and forward collision warning. The long-range RADAR sensor may include monostatic multimodal RADARs having multiple (e.g., six or more) fixed RADAR antennas and high-speed CAN and FlexRay interfaces. In one example with six antennas, the four central antennas may create a focused beam pattern designed to record the area around the vehicle 900 at high speed with minimal interference from traffic in adjacent lanes. The other two antennas can widen the field of view, enabling rapid detection of vehicles entering or leaving the lane of the vehicle 900.

[0142] As an example, a medium-range RADAR system may include a range of up to 960m (front) or 80m (rear) and a field of view of up to 42 degrees (front) or 950 degrees (rear). A short-range RADAR system may include, but is not limited to, RADAR sensors designed to be mounted on both ends of the rear bumper. When mounted on both ends of the rear bumper, such a RADAR sensor system can create two beams that constantly monitor the blind spots behind and beside the vehicle.

[0143] Short-range radar systems can be used in ADAS systems for blind spot detection and / or lane change assistance.

[0144] The vehicle 900 may further include ultrasonic sensors 962. Positioned on the front, rear, and / or sides of the vehicle 900, the ultrasonic sensors 962 may be used for parking assistance and / or for creating and updating the occupancy grid. A wide variety of ultrasonic sensors 962 may be used, and different ultrasonic sensors 962 may be used for detection of different ranges (e.g., 2.5m, 4m). The ultrasonic sensors 962 may operate at a functional safety level of ASIL B.

[0145] The vehicle 900 may include a LiDAR sensor 964. The LiDAR sensor 964 may be used for object and pedestrian detection, emergency braking, collision avoidance, and / or other functions. The LiDAR sensor 964 may also have a functional safety level of ASIL B. In some examples, the vehicle 900 may include multiple LiDAR sensors 964 (e.g., two, four, six, etc.) that can use Ethernet® (for example, to provide data to a Gigabit Ethernet® switch).

[0146] In some examples, the LIDAR sensor 964 may have the ability to provide a list of objects and their distances within a 360-degree field of view. A commercially available LIDAR sensor 964 may have an advertised range of approximately 900m, for example, with an accuracy of 2cm to 3cm and support for 900Mbps Ethernet® connectivity. In some examples, one or more non-protruding LIDAR sensors 964 may be used. In such examples, the LIDAR sensor 964 may be implemented as a small device that can be incorporated into the front, rear, side, and / or corners of a vehicle 900. In such examples, the LIDAR sensor 964 may have a range of 200m even for low-reflection objects and can provide a field of view up to 920 degrees horizontal and 35 degrees vertical. A front-mounted LIDAR sensor 964 may be configured for a horizontal field of view between 45 and 135 degrees.

[0147] In some applications, LiDAR technologies such as 3D flash LiDAR may also be used. 3D flash LiDAR uses a laser flash as a source to illuminate the area around the vehicle up to approximately 200m. The flash LiDAR unit includes receptors that record the laser pulse travel time and reflected light on each pixel, sequentially corresponding to the range from the vehicle to the object. Flash LiDAR can enable the generation of high-precision and distortion-free images of the surroundings with every laser flash. In some applications, four flash LiDAR sensors may be deployed, one on each side of the vehicle. Available 3D flash LiDAR systems include solid-state 3D steering array LiDAR cameras (e.g., non-scanning LiDAR devices) that have no moving parts other than a blower. Flash LiDAR devices can use 5 nanosecond Class I (eye-safe) laser pulses per frame and can capture reflected laser light in the form of a 3D range point cloud and co-documented intensity data. By using flash LiDAR, and because flash LiDAR is a solid-state device with no moving parts, the LiDAR sensor 964 can be made less susceptible to motion blur, vibration, and / or shock.

[0148] The vehicle may further include an IMU sensor 966. In some examples, the IMU sensor 966 may be positioned in the center of the rear axle of the vehicle 900. The IMU sensor 966 may include, but is not limited to, an accelerometer, magnetometer, gyroscope, magnetic compass, and / or other sensor types. In some examples, such as in a 6-axis application, the IMU sensor 966 may include an accelerometer and a gyroscope, while in a 9-axis application, the IMU sensor 966 may include an accelerometer, a gyroscope, and a magnetometer.

[0149] In some embodiments, the IMU sensor 966 may be implemented as a miniature, high-performance GPS-aided inertial navigation system (GPS / INS) that combines a micro-electro-mechanical system (MEMS) inertial sensor, a high-sensitivity GPS receiver, and an advanced Kalman filtering algorithm to provide estimates of position, velocity, and attitude. As such, in some examples, the IMU sensor 966 may enable the vehicle 900 to estimate its direction of travel without requiring input from a magnetic sensor by directly observing and correlating velocity changes from the GPS to the IMU sensor 966. In some embodiments, the IMU sensor 966 and the GNSS sensor 958 may be combined in a single integrated unit.

[0150] The vehicle may include a microphone 996 placed inside and / or around the vehicle 900. The microphone 996 may, among other things, be used for emergency vehicle detection and identification.

[0151] The vehicle may further include any number of camera types, including a stereo camera 968, a wide-view camera 970, an infrared camera 972, a surround camera 974, a long-range and / or medium-range camera 998, and / or other camera types. The cameras may be used to capture image data around the entire exterior surface of the vehicle 900. The type of camera used will depend on the embodiment and requirements of the vehicle 900, and any combination of camera types may be used to achieve the required coverage around the vehicle 900. In addition, the number of cameras may vary depending on the embodiment. For example, the vehicle may include six cameras, seven cameras, ten cameras, twelve cameras, and / or another number of cameras. The cameras may, as an example, support Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet®. Each camera is described in more detail herein in relation to Figures 9A and 9B.

[0152] The vehicle 900 may further include a vibration sensor 942. The vibration sensor 942 can measure vibrations of vehicle components, such as axles. For example, a change in vibration may indicate a change in the road surface. In another example, when two or more vibration sensors 942 are used, the difference in vibration may be used to determine friction or slippage of the road surface (for example, when the difference in vibration is between a power-driven axle and a free-rotating axle).

[0153] Vehicle 900 may include ADAS system 938. In some examples, ADAS system 938 may include SoC. ADAS system 938 may include autonomous / adaptive / automatic cruise control (ACC), cooperative adaptive cruise control (CACC), forward crash warning (FCW), automatic emergency braking (AEB), lane departure warning (LDW), lane keep assist (LKA), blind spot warning (BSW), rear cross-traffic warning (RCTW), collision warning system (CWS), lane centering (LC), and / or other features and functions.

[0154] The ACC system may use a radar sensor 960, a lithium-ion sensor 964, and / or a camera. The ACC system may include longitudinal ACC and / or transverse ACC. Longitudinal ACC monitors and controls the distance to the vehicle immediately in front of vehicle 900 and automatically adjusts the vehicle speed to maintain a safe distance from the vehicle ahead. Transverse ACC performs distance maintenance and advises vehicle 900 to change lanes when necessary. Transverse ACC is related to other ADAS applications such as LCA and CWS.

[0155] CACC uses information from other vehicles that can be received from other vehicles via a wireless link through a network interface 924 and / or a wireless antenna 926, or indirectly via a network connection (e.g., via the Internet). Direct links may be provided by vehicle-to-vehicle (V2V) communication links, while indirect links may be infrastructure-to-vehicle (I2V) communication links. Generally, the V2V communication concept provides information about the vehicle immediately ahead (e.g., a vehicle in the same lane as vehicle 900, immediately in front of vehicle 900), while the I2V communication concept provides information about traffic further ahead. A CACC system may include either or both I2V and V2V information sources. Given information about vehicles ahead of vehicle 900, CACC can be more reliable, and CACC has the potential to make traffic flow smoother and reduce road congestion.

[0156] The FCW system is designed to warn the driver of hazards so that the driver can take corrective action. The FCW system uses a forward-facing camera and / or radar sensor 960, coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to driver feedback such as a display, speaker, and / or vibration components. The FCW system can provide warnings in the form of audible, visual, vibration, and / or quick brake pulses.

[0157] An AEB system can detect an imminent forward collision with another vehicle or object and automatically apply the brakes if the driver does not take corrective action within a specified time or distance parameter. The AEB system may use a forward-facing camera and / or radar sensor 960 coupled to a dedicated processor, DSP, FPGA, and / or ASIC. When the AEB system detects a hazard, it typically first warns the driver to take corrective action to avoid the collision. If the driver does not take corrective action, the AEB system may automatically apply the brakes as part of an effort to prevent, or at least mitigate, the impact of the anticipated collision. The AEB system may include techniques such as dynamic brake support and / or impending collision braking.

[0158] The LDW system warns the driver when the vehicle crosses a lane marking by providing visual, audible, and / or tactile warnings, such as vibration of the steering wheel or seat. The LDW system does not activate when the driver indicates an intentional lane departure by activating the turn signal. The LDW system may use a forward-facing camera connected to a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to driver feedback, such as a display, speaker, and / or vibration components.

[0159] The LKA system is a modified version of the LDW system. The LKA system provides steering input or braking to correct the vehicle 900 if it begins to drift out of its lane.

[0160] The BSW system detects and warns the driver of a vehicle in the vehicle's blind spot. The BSW system can provide visual, audible, and / or tactile warnings to indicate that merging or changing lanes is unsafe. The system can provide additional warnings when the driver uses the turn signal. The BSW system can use a rear-facing camera and / or radar sensor 960 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, electrically coupled to driver feedback, such as a display, speaker, and / or vibration component.

[0161] The RCTW system can provide visual, audible, and / or haptic notifications when an object is detected outside the range of the rear camera while the vehicle 900 is reversing. Some RCTW systems include AEB to ensure that the vehicle brakes are applied to avoid a collision. The RCTW system may use one or more rear-facing RADAR sensors 960 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which are electrically coupled to driver feedback, such as a display, speaker, and / or vibration component.

[0162] Conventional ADAS systems warn the driver and allow the driver to determine whether a safe condition truly exists and act accordingly. However, conventional ADAS systems have sometimes tended to produce misjudgments that, while not usually catastrophic, can be troubling and distracting to the driver. In the autonomous vehicle 900, however, if the results are contradictory, the vehicle 900 itself must decide whether to heed the results from the primary computer or the secondary computer (e.g., the first controller 936 or the second controller 936). For example, in some embodiments, the ADAS system 938 may also be a backup and / or secondary computer for providing perceptual information to a backup computer rationality module. The backup computer rationality monitor can run a variety of redundant software on hardware components to detect failures in perceptual and dynamic driving tasks. The output from the ADAS system 938 may be provided to the supervisory MCU. If the outputs from the primary and secondary computers are contradictory, the supervisory MCU must decide how to reconcile the contradiction to ensure safe operation.

[0163] In some implementations, a primary computer may be configured to provide a supervising MCU with a reliability score indicating the reliability of the primary computer in a selected outcome. If the reliability score exceeds a threshold, the supervising MCU may follow the primary computer's instructions, regardless of whether the secondary computer gives conflicting or inconsistent results. If the reliability score does not meet the threshold, and the primary and secondary computers produce different results (e.g., conflicting results), the supervising MCU may mediate between the computers to determine an appropriate outcome.

[0164] The supervisory MCU may be configured to run a neural network trained and configured to determine, based on the outputs from the primary and secondary computers, when a secondary computer is providing a false alarm. Thus, the neural network in the supervisory MCU can learn when the output of the secondary computer is reliable and when it is not. For example, when the secondary computer is a radar-based forward crossing (FCW) system, the neural network in the supervisory MCU can learn when the FCW is identifying metal objects that are not actually dangerous, such as sewer grates or manhole covers that trigger an alarm. Similarly, when the secondary computer is a camera-based lane departure warning (LDW) system, the neural network in the supervisory MCU can learn to ignore the LDW when a cyclist or pedestrian is present and lane departure is actually the safest operation. In embodiments involving a neural network running on the supervisory MCU, the supervisory MCU may include at least one of a DLA or GPU suitable for running a neural network with associated memory. In a preferred embodiment, the supervisory MCU may comprise and / or be included as a component of the SoC904.

[0165] In other examples, the ADAS system 938 may include a secondary computer that performs ADAS functions using conventional rules of computer vision. As such, the secondary computer may use classical computer vision rules (if-then), and the presence of a neural network within the supervisory MCU can improve reliability, safety, and performance. For example, diverse implementations and intentional non-identities make the entire system more fault-tolerant, particularly to failures caused by software (or software-hardware interface) functions. For instance, if a software bug or error exists in the software running on the primary computer, and non-identical software code running on the secondary computer produces the same overall result, the supervisory MCU may have greater confidence that the overall result is correct and that the bug in the software or hardware on the primary computer did not cause a critical error.

[0166] In some examples, the output of the ADAS system 938 may be supplied to the perception block and / or the dynamic driving task block of the primary computer. For example, if the ADAS system 938 indicates a forward collision warning due to an object immediately ahead, the perception block can use this information when identifying the object. In other examples, the secondary computer may have its own neural network, which is trained as described herein and therefore reduces the risk of misjudgment.

[0167] Vehicle 900 may further include an infotainment SoC 930 (for example, an in-vehicle infotainment system (IVI)). Although illustrated and described as an SoC, the infotainment system does not have to be an SoC and may include two or more separate components. The infotainment SoC 930 may include a combination of hardware and software that can be used to provide vehicle 900 with audio (e.g., music, personal digital assistant, navigation commands, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), telephone (e.g., hands-free calling), network connectivity (e.g., LTE, Wi-Fi, etc.), and / or information services (e.g., navigation system, rear parking assist, radio data system, fuel level, total mileage, brake fuel level, oil level, door open / close, air filter information, and other vehicle-related information). For example, the infotainment SoC 930 may also include wireless, disc player, navigation system, video player, USB and Bluetooth connectivity, car computer, in-car entertainment, Wi-Fi, steering wheel audio control unit, hands-free voice control, heads-up display (HUD), HMI display 934, telematics device, control panel (for example, for controlling and / or interacting with various components, features, and / or systems), and / or other components. The infotainment SoC 930 may be further used to provide information (for example, visual and / or audible) to the vehicle user, such as information from the ADAS system 938, autonomous driving information such as planned vehicle operation, trajectory, surrounding environment information (for example, intersection information, vehicle information, road information, etc.), and / or other information.

[0168] The infotainment SoC 930 may include GPU functionality. The infotainment SoC 930 can communicate with other devices, systems, and / or components of the vehicle 900 via bus 902 (e.g., CAN bus, Ethernet®, etc.). In some examples, the infotainment SoC 930 may be coupled to a supervisory MCU so that the infotainment system's GPU can perform certain self-drive functions in the event of a primary controller 936 (e.g., the vehicle 900's primary and / or backup computer) failure. In such examples, the infotainment SoC 930 can put the vehicle 900 into a chauffeur-safe stop mode as described herein.

[0169] Vehicle 900 may further include an instrument cluster 932 (e.g., a digital dash, an electronic instrument cluster, a digital instrument panel, etc.). The instrument cluster 932 may include a controller and / or a supercomputer (e.g., a separate controller or supercomputer). The instrument cluster 932 may include a set of instruments such as a speedometer, fuel level indicator, oil pressure indicator, tachometer, odometer, turn signals, gear shift position indicator, seat belt warning light, parking brake warning light, engine fault light, airbag (SRS) system information, lighting control device, safety system control device, and navigation information. In some examples, information may be displayed and / or shared between the infotainment SoC 930 and the instrument cluster 932. In other words, the instrument cluster 932 may be included as part of the infotainment SoC 930, and vice versa.

[0170] Figure 9D is a system diagram of communication between the cloud-based server of Figure 9A and an exemplary autonomous vehicle 900, according to some embodiments of the present disclosure. System 976 may include a server 978, a network 990, and a vehicle including the vehicle 900. Server 978 may include a plurality of GPUs 984(A) to 984(H) (collectively referred to herein as GPU 984), PCIe switches 982(A) to 982(H) (collectively referred to herein as PCIe switch 982), and / or CPUs 980(A) to 980(B) (collectively referred to herein as CPU 980). The GPUs 984, CPUs 980, and PCIe switches may be interconnected by high-speed interconnects, such as, for example, NVLink interfaces 988 and / or PCIe connections 986 developed by NVIDIA. In some examples, the GPU984 is connected via NVLink and / or NVSwitch SoCs, and the GPU984 and PCIe switch 982 are connected via PCIe interconnects. Eight GPU984s, two CPU980s, and two PCIe switches are illustrated, but this is not intended to be an limitation. Depending on the embodiment, each server 978 may contain any number of GPU984s, CPU980s, and / or PCIe switches. For example, server 978 may contain eight, sixteen, thirty-two, and / or more GPU984s, respectively.

[0171] Server 978 can receive image data from vehicles via network 990, representing images showing unexpected or altered road conditions, such as recently started road construction. Server 978 can transmit map information 994, including information about traffic and road conditions, to vehicles via network 990, including information about the neural network 992, updated neural network 992, and / or map information 994. Updates to map information 994 may include updates to HD map 922, such as information about construction sites, potholes, detours, floods, and / or other obstacles. In some examples, the neural network 992, updated neural network 992, and / or map information 994 may have arisen from new training and / or experience represented in data received from any number of vehicles in the environment, and / or based on training performed in a data center (for example, using server 978 and / or other servers).

[0172] Server 978 may be used to train a machine learning model (e.g., a neural network) based on training data. The training data may be generated by a vehicle and / or in a simulation (e.g., using a game engine). In some instances, the training data is tagged (e.g., if the neural network benefits from supervised learning) and / or otherwise pre-processed, while in other instances, the training data is not tagged and / or pre-processed (e.g., if the neural network does not require supervised learning). Training may be performed according to any one or more classes of machine learning techniques, including but not limited to the following: supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, associative learning, transfer learning, feature learning (including key component and cluster analysis), multilinear subspace learning, manifold learning, representation learning (including pre-dictionary learning), rule-based machine learning, anomaly detection, and variations or combinations thereof. After the machine learning model has been traced, it may be used by the vehicle (for example, transmitted to the vehicle via network 990), and / or the machine learning model may be used by server 978 to remotely monitor the vehicle.

[0173] In some examples, Server 978 can receive data from vehicles and apply it to a state-of-the-art real-time neural network for real-time intelligent inference. Server 978 may include deep learning supercomputers and / or dedicated AI computers powered by GPU 984, such as the DGX and DGX Station Machines developed by NVIDIA. However, in some examples, Server 978 may include deep learning infrastructure that uses only CPU-powered data centers.

[0174] The deep learning infrastructure of server 978 can have the capability for high-speed real-time inference, which can be used to evaluate and verify the condition of the processor, software, and / or associated hardware within vehicle 900. For example, the deep learning infrastructure can receive periodic updates from vehicle 900, such as images of a sequence and / or objects located within images of that sequence (e.g., via computer vision and / or other machine learning object classification techniques). The deep learning infrastructure can run its own neural network to identify objects and compare them with objects identified by vehicle 900. If the results do not match and the infrastructure concludes that the AI ​​within vehicle 900 is not functioning correctly, server 978 can send a signal to vehicle 900 instructing the vehicle's fail-safe computer to infer control, notify passengers, and complete a safe parking operation.

[0175] For inference, server 978 may include GPU 984 and one or more programmable inference accelerators (e.g., NVIDIA TensorRT). The combination of a GPU-powered server and inference accelerator can enable real-time responsiveness. In other examples, such as when high performance is not required, a server powered by a CPU, FPGA, and other processors may be used for inference.

[0176] Exemplary computing devices Figure 10 is a block diagram of an example of a computing device 1000 suitable for use in implementing some embodiments of the present disclosure. The computing device 1000 may include an interconnection system 1002 that indirectly or directly connects the following devices: memory 1004, one or more central processing units (CPUs) 1006, one or more graphics processing units (GPUs) 1008, a communication interface 1010, input / output (I / O) ports 1012, input / output components 1014, a power supply unit 1016, one or more presentation components 1018 (e.g., a display), and one or more logic units 1020.

[0177] The various blocks in Figure 10 are shown connected by lines via the interconnection system 1002, but this is not intended to be restrictive and is simply for clarity. For example, in some embodiments, a presentation component 1018, such as a display device, could be considered an I / O component 1014 (for example, if the display is a touchscreen). In another example, the CPU 1006 and / or GPU 1008 could include memory (for example, memory 1004 could represent a storage device in addition to the memory of the GPU 1008, CPU 1006, and / or other components). In other words, the computing devices in Figure 10 are merely illustrative. Categories such as “workstation,” “server,” “laptop,” “desktop,” “tablet,” “client device,” “mobile device,” “handheld device,” “game console,” “electronic control unit (ECU),” “virtual reality system,” and / or other device or system types are all intended to fall within the scope of the computing devices in Figure 10 and are therefore not distinguished.

[0178] The interconnection system 1002 may represent one or more links or buses, for example, an address bus, a data bus, a control bus, or a combination thereof. The interconnection system 1002 may include one or more bus or link types, for example, an industry standard architecture (ISA) bus, an extended industry standard architecture (EISA) bus, a VESA (video electronics standards association) bus, a peripheral component interconnect (PCI) bus, a peripheral component interconnect express (PCIe) bus, and / or other types of buses or links. In some embodiments, direct connections exist between components. For example, the CPU 1006 may be directly connected to the memory 1004. Furthermore, the CPU 1006 may be directly connected to the GPU 1008. Where direct or point-to-point connections exist between components, the interconnection system 1002 may include PCIe links for implementing the connections. In these examples, the PCI bus does not need to be included in computing device 1000.

[0179] The memory 1004 may include any of various computer-readable media. The computer-readable media may be any available media that can be accessed by the computing device 1000. The computer-readable media may include both volatile and non-volatile media, and removable and non-removable media. For example, but not limited to, the computer-readable media may include computer storage media and communication media.

[0180] Computer storage media may include both volatile and non-volatile media and / or removable and non-removable media implemented in any method or technique for storing information such as computer-readable instructions, data structures, program modules, and / or other data types. For example, memory 1004 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 memory technologies, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other media that can be used to store desired information and can be accessed by computing device 1000. In this specification, computer storage media does not include signals themselves.

[0181] Computer storage media include any information distribution medium that can implement computer-readable instructions, data structures, program modules, and / or other data types in modulated data signals such as carrier waves or other transfer mechanisms. The term “modulated data signal” may refer to a signal that has been modified in a manner that has one or more of its characteristic sets or encodes information within the signal. For example, but not limited to, computer storage media may include wired media such as wired networks or direct wired connections, and wireless media such as acoustic, RF, infrared, and other wireless media. Any combination of the foregoing should also be included in the scope of computer-readable media.

[0182] The CPU 1006 may be configured to execute at least some computer-readable instructions to control one or more components of the computing device 1000 to execute one or more of the methods and / or processes described herein. The CPU 1006 may include one or more cores (e.g., one, two, four, eight, twenty-eight, seventy-two, etc.) each capable of processing a large number of software threads simultaneously. The CPU 1006 may include any type of processor, and depending on the type of computing device 1000 in which it is implemented, it may include different types of processors (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 1000, the processor may be an Advanced RISC Machines (ARM) processor implemented using Reduced Instruction Set Computing (RISC), or an x86 processor implemented using Complex Instruction Set Computing (CISC). The computing device 1000 may include one or more CPUs 1006 in one or more microprocessors or auxiliary coprocessors, such as a computing coprocessor.

[0183] In addition to or instead of the CPU 1006, the GPU 1008 may be configured to execute at least some computer-readable instructions to control one or more components of the computing device 1000 to execute one or more of the methods and / or processes described herein. One or more of the GPUs 1008 may be an integrated GPU (for example, one or more of the CPUs 1006), and / or one or more of the GPUs 1008 may be discrete GPUs. In embodiments, one or more of the GPUs 1008 may be coprocessors of one or more of the CPUs 1006. The GPUs 1008 may be used by the computing device 1000 to render graphics (for example, 3D graphics) or to perform general-purpose computing. For example, the GPUs 1008 may be used for general-purpose computing on a GPU (GPGPU). It may be used for a GPU. The GPU 1008 may include hundreds or thousands of cores capable of processing hundreds or thousands of software threads simultaneously. The GPU 1008 can generate pixel data for an output image in response to rendering commands (for example, rendering commands from CPU 1006 received via the host interface). The GPU 1008 may include graphics memory, such as display memory, for storing pixel data or any other suitable data, such as GPGPU data. The display memory may be included as part of memory 1004. GPU1008 may include two or more GPUs operating in parallel (for example, via a link). The link can connect directly to the GPUs (for example, using NVLINK) or via a switch (for example, using NVSwitch). When coupled together, each GPU1008 can generate pixel data or GPGPU data for different parts of the output or different outputs (for example, the first GPU for the first image and the second GPU for the second image). Each GPU may have its own memory or may share memory with other GPUs.

[0184] In addition to or instead of the CPU 1006 and / or GPU 1008, the logic unit 1020 may be configured to execute at least some computer-readable instructions to control one or more of the computing devices 1000 to execute one or more of the methods and / or processes described herein. In embodiments, the CPU 1006, GPU 1008, and / or the logic unit 1020 can execute any combination of methods, processes, and / or parts thereof discretely or congruently. One or more of the logic units 1020 may be part of and / or integrated with one or more of the CPU 1006 and / or GPU 1008, and / or one or more of the logic units 1020 may be discrete components of the CPU 1006 and / or GPU 1008 or otherwise external to them. In embodiments, one or more of the logic units 1020 may be coprocessors of one or more of the CPU 1006 and / or one or more of the GPU 1008.

[0185] Examples of the logic unit 1020 include one or more processing cores and / or components thereof, such as Tensor Cores (TCs), Tensor Processing Units (TPUs), Pixel Visual Cores (PVCs), Vision Processing Units (VPUs), Graphics Processing Clusters (GPCs), Texture Processing Clusters (TPCs), Streaming Multiprocessors (SMs), Tree Traversal Units (TTUs), Artificial Intelligence Accelerators (AIAs), Deep Learning Accelerators (DLAs), Logical Units (ALUs), Application-Specific Integrated Circuits (ASICs), Floating-Point Units (FPUs), Input / Output (I / O) elements, Peripheral Component Interconnect (PCI) or Peripheral Component Interconnect Express (PCIe) elements, and / or similar.

[0186] The communication interface 1010 may include one or more receivers, transmitters, and / or transceivers that enable the computing device 1000 to communicate with other computing devices via an electronic communication network, including wired and / or wireless communication. The communication interface 1010 may include components and functions to enable communication over any of several different networks, such as wireless networks (e.g., Wi-Fi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), wired networks (e.g., communicating via Ethernet® or InfiniBand), low-power wide-area networks (e.g., LoRaWAN, SigFox, etc.), and / or the Internet.

[0187] The I / O port 1012 can enable the computing device 1000 to be logically connected to other devices, including I / O components 1014, presentation components 1018, and / or other components, some of which can be built into (e.g., integrated into) the computing device 1000. Exemplary I / O components 1014 include microphones, mice, keyboards, joysticks, gamepads, game controllers, satellite dishes, scanners, printers, wireless devices, etc. The I / O components 1014 can provide a natural user interface (NUI) that processes air gestures, voice, or other physiological inputs generated by the user. In some cases, the input may be transmitted to appropriate network elements for further processing. The NUI may implement any combination of voice recognition, stylus recognition, face recognition, biometric recognition, on-screen and beside-screen gesture recognition, air gestures, head and target tracking, and touch recognition related to the display of the computing device 1000 (as described in more detail later). The computing device 1000 may include depth cameras, such as stereoscope camera systems, infrared camera systems, RGB camera systems, touchscreen technology, and combinations thereof, for gesture detection and recognition. Additionally, the computing device 1000 may include accelerometers or gyroscopes that enable motion detection (for example, as part of an inertia measurement unit (IMU)). In some examples, the output of the accelerometer or gyroscope may be used by the computing device 1000 to render immersive augmented reality or virtual reality.

[0188] The power supply device 1016 may include a hardwired power supply device, a battery power supply device, or a combination thereof. The power supply device 1016 can supply power to the computing device 1000 to enable the components of the computing device 1000 to operate.

[0189] The presentation component 1018 may include a display (e.g., a monitor, touch screen, television screen, head-up display device (HUD), other display types, or a combination thereof), a speaker, and / or other presentation components. The presentation component 1018 can receive data from other components (e.g., GPU 1008, CPU 1006, etc.) and output data (e.g., as images, videos, sounds, etc.).

[0190] Exemplary network environment A network environment suitable for use in implementing the embodiments of this disclosure may include one or more client devices, servers, network-attached storage (NAS), other backend devices, and / or other device types. Each client device, server, and / or other device type (for example, each device) may be implemented as one or more instances of the computing device 1000 in Figure 10, for example, each device may include similar components, features, and / or functionalities of the computing device 1000.

[0191] Components of a network environment may communicate with one another via the network, either wired, wirelessly, or both. A network may include multiple networks, or a network of networks. For example, a network may include one or more wide area networks (WANs), one or more local area networks (LANs), one or more public networks, such as the Internet and / or the Public Switched Telephone Network (PSTN), and / or one or more private networks. If a network includes a wireless telecommunications network, its components, such as base stations, towers, or access points (and other components), may provide wireless connectivity.

[0192] Compatible network environments may include one or more peer-to-peer network environments (in which case servers may not be included in the network environment) and one or more client-server network environments (in which case one or more servers may be included in the network environment). In a peer-to-peer network environment, the functionality described herein with respect to the server can be implemented on any number of client devices.

[0193] In at least one embodiment, the network environment may include one or more cloud-based network environments, distributed computing environments, or a combination thereof. The cloud-based network environment may include a framework layer, a job scheduler, a resource manager, and a distributed file system implemented on one or more of the servers, which may include one or more core network servers and / or edge servers. The framework layer may include a framework to support the software in the software layer and / or one or more applications in the application layer. The software or applications may each include web-based service software or applications. In the embodiment, one or more of the client devices may use the web-based service software or applications (for example, by accessing the service software and / or applications via one or more application programming interfaces (APIs)). The framework layer may be, but is not limited to, a type of free and open-source software web application framework that may use a distributed file system for, for example, large-scale data processing (e.g., “big data”).

[0194] A cloud-based network environment may provide cloud computing and / or cloud storage that implements any combination of the computing and / or data storage functions (or one or more of them) described herein. Any of these various functions may be distributed across multiple locations from a central or core server (e.g., one or more data centers that may be distributed across states, territories, countries, or the world). If the connection to the user (e.g., a client device) is relatively close to the edge server, the core server may delegate at least a portion of its functionality to the edge server. The cloud-based network environment may be private (e.g., limited to a single organization), public (e.g., available to multiple organizations), and / or a combination thereof (e.g., a hybrid cloud environment).

[0195] A client device may include at least some of the components, features, and functionalities of the exemplary computing device 1000 described herein with respect to Figure 10. As an example, and not limited to, a client device may be implemented as a personal computer (PC), laptop computer, mobile device, smartphone, tablet computer, smartwatch, wearable computer, personal digital assistant (PDA), MP3 player, virtual reality headset, global positioning system (GPS) or device, video player, video camera, surveillance device or system, vehicle, boat, airship, virtual machine, drone, robot, handheld communication device, hospital device, gaming device or system, entertainment system, vehicle computer system, embedded system controller, remote control, instrument, consumer electronic device, workstation, edge device, any combination of these depicted devices, or any other suitable device.

[0196] This disclosure may be described in general terms with computer code or machine-usable instructions, including computer-executable instructions such as program modules, which are executed by computers or other machines, such as personal digital assistants or other handheld devices. Generally, a program module, including routines, programs, objects, components, and data structures, refers to code that performs a specific task or implements a specific abstract data type. This disclosure may be implemented in a variety of configurations, including handheld devices, consumer electronics, general-purpose computers, and more specialized computing devices. This disclosure may also be implemented in a distributed computing environment where tasks are performed by remote processing devices linked over a communication network.

[0197] In this specification, any “and / or” statement relating to two or more elements should be interpreted as meaning 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 element A, at least one element B, or at least one element A and at least one element B. Furthermore, “at least one of element A and element B” may include at least one element A, at least one element B, or at least one element A and at least one element B.

[0198] The subject matter of this disclosure is described in a manner that is specific in order to satisfy statutory requirements. However, the description itself is not intended to limit the scope of this disclosure. Rather, the inventors intend that the claimed subject matter may be carried out in other ways, including different steps or combinations of steps similar to those described herein, in conjunction with other current or future technologies. Furthermore, the terms “step” and / or “block” may be used herein to imply different elements of the way in which they are used, but these terms should not be construed as implying any particular order among the various steps disclosed herein unless the order of the individual steps is expressly stated and, when so, expressed.

Claims

1. A method comprising the steps of: applying image data representing an intersection to a neural network; using the neural network to compute first data representing a first signed distance function corresponding to a first intersection area class and a second signed distance function corresponding to a second intersection area class, based at least partially on the image data; applying a clustering algorithm to the first signed distance function and the second signed distance function; computing a first instance partition map corresponding to the first intersection area class and a second instance partition map corresponding to a second intersection area class, based at least partially on the clustering algorithm; determining the world space location corresponding to an instance of the first intersection area class and an instance of the second intersection area class, based at least partially on the first instance partition map and the second instance partition map; and transmitting second data representing the world space location to a vehicle planning component, wherein the first signed distance function and the second signed distance function are used to compute values ​​indicating the pixel distance of a pixel from the boundary of the first intersection area class and the second intersection area class.

2. The method according to claim 1, further comprising the step of applying a smoothing operation to the first signed distance function and the second signed distance function to generate a smoothed first signed distance function and a smoothed second signed distance function, wherein the step of applying the clustering algorithm is performed on the smoothed first signed distance function and the smoothed second signed distance function.

3. The method according to claim 2, wherein the smoothing operation is applied to the value of each pixel of an image representing the image data, and the value of each pixel corresponds to the first signed distance function and the second signed distance function.

4. The method according to claim 2, wherein the smoothing operation is a Gaussian smoothing operation.

5. The method according to claim 1, wherein the image data is applied to the neural network at a first spatial resolution, and the first signed distance function and the second signed distance function are output at a second spatial resolution smaller than the first spatial resolution, and the method further includes the step of applying an upsampling operation to the first signed distance function and the second signed distance function to generate an upsampled first signed distance function and an upsampled second signed distance function at the first spatial resolution.

6. The method according to claim 1, wherein the clustering algorithm is a watershed clustering algorithm.

7. The method according to claim 1, wherein the signs of the first signed distance function and the second signed distance function are based on a zero threshold, and the clustering algorithm is applied after the zero threshold step.

8. The method according to claim 1, wherein the first instance partition map and the second instance partition map are represented in image space, and the step of determining the world space position includes the step of converting the image space position to the world space position using one or more of intrinsic sensor parameters or external sensor parameters.

9. A system comprising one or more image sensors on a vehicle, one or more processors, and one or more memory devices that store instructions causing the one or more processors to perform an operation when executed using the one or more processors, wherein the operation involves applying image data representing an intersection to a deep neural network (DNN), and using the DNN to compute first data representing signed distance functions based at least in part on the image data, wherein each of the signed distance functions corresponds to an intersection area class trained to predict the DNN, and clustering A system comprising: applying an algorithm to the signed distance function; calculating an instance partition map in image space, at least in part on the clustering algorithm, such that each instance partition map corresponds to the intersection area class of the corresponding signed distance function; determining the world space location corresponding to the instance partition map; and transmitting second data representing the world space location to a vehicle planning component, wherein the first and second signed distance functions calculate values ​​indicating the pixel distance of a pixel from the boundary of the first and second intersection area classes.

10. The system according to claim 9, wherein the operation further includes applying a smoothing operation to each of the signed distance functions of the signed distance function to generate a corresponding smoothed first signed distance function, and the application of the clustering algorithm is performed on the smoothed first signed distance function.

11. The system according to claim 10, wherein the smoothing operation is applied to the value of each pixel of an image representing the image data, and the value of each pixel corresponds to each signed distance function among the signed distance functions.

12. The system according to claim 9, wherein the operation further includes applying a zero threshold to each sign of the signed distance function, and the application of the clustering algorithm occurs after the application of the zero threshold.

13. The system according to claim 9, wherein the image data is applied to the DNN at a first spatial resolution, the signed distance function is output at a second spatial resolution smaller than the first spatial resolution, and the operation further includes applying an upsampling operation to each of the signed distance functions to generate a corresponding upsampled signed distance function at the first spatial resolution.

14. A method comprising the steps of: receiving image data representing an image showing an intersection; generating annotations representing boundary shapes corresponding to the area of ​​the intersection and corresponding semantic class labels corresponding to the semantic classes of each of the areas; calculating a signed distance function for each of the semantic class types corresponding to the intersection, wherein the signed distance function includes a first signed value for a first pixel of the image inside the area, a second signed value for a second pixel of the image outside the area, and a third signed value for a third pixel of the image along the boundary of the area; and training a deep neural network (DNN) using the signed distance function as ground truth data, wherein the signed distance function calculates a value indicating the pixel distance of a pixel from the boundary of the intersection area.

15. The method according to claim 14, wherein the first signed value, the second signed value, and the third signed value correspond to the pixel distance to the nearest boundary among the boundaries of the area.

16. The method according to claim 14, wherein the step of training the DNN includes the step of constraining the output of the DNN using one or more loss functions.

17. The method according to claim 16, wherein the one or more loss functions include a separate loss function for each semantic class type.

18. The method according to claim 14, wherein the boundary shape includes a polygon that matches the shape of the area.

19. The method according to claim 14, wherein the semantic class type includes one or more of internal ground, external ground, partially visible, competitor exiting, competitor entering, no competitor lane, crosswalk area, and fully visible.

20. The method according to claim 14, wherein the step of generating the annotations includes the step of individually labeling each instance of each semantic class type of the image.

Images