Stationary object detection

Abstract

The present disclosure relates to stationary object detection, in particular to a method of detecting stationary objects implemented by a control system of an autonomous vehicle, comprising: receiving sensor signal data including stationary detections and non-stationary detections from a surrounding environment of the vehicle; determining at least one stationary detection combination that satisfies one or more lateral position selection criteria based on a lateral position of each stationary detection from a direction that the vehicle faces; determining at least one stationary detection combination that satisfies one or more combination regularity selection criteria based on a regularity of a position difference between pairs of stationary detections in the combination that are sequentially positioned in the direction that the vehicle faces; determining one or more stationary detection combinations that satisfy the lateral selection criteria and the combination regularity selection criteria for combinations of stationary detections corresponding to at least one stationary object; removing stationary detections corresponding to the at least one stationary object from the sensor signal data output to one or more data processing components of the control system.

Description

Technical Field

[0001] The disclosed technology relates to methods for detecting stationary objects, such as methods for detecting stationary objects with specific types of physical properties in object data that includes sensed stationary objects and sensed non-stationary objects, and related aspects.

[0002] The disclosed techniques are applicable to vehicles, including semi-autonomous vehicles that assist the driver and fully autonomous vehicles. The disclosed techniques can be implemented in the control systems of such vehicles (also referred to herein as ego vehicles). The disclosed techniques can be implemented in the control systems of ego vehicles to detect and remove one or more types of stationary object detections from a stream of sensed object detections generated by one or more radar sensors of the ego vehicle. In some embodiments, the disclosed techniques are particularly useful for the sensing detection of specific types of stationary objects, such as road or lane-side structures (e.g., guardrails or similar obstacle-like structures that may be arranged alongside the path traversed by the vehicle). Such structures can extend over long distances and can generate a large number of sensor detections during a sensing cycle.

[0003] The disclosed technologies seek to provide solutions to the problem of how to reliably reduce the amount of data that the perception systems of autonomous vehicles need to process in order to provide reliable information about the sensed environment around the autonomous vehicle. The disclosed technologies seek to provide a way to quickly determine whether a combination of sensed stationary detections sharing one or more attributes is jointly a result of sensing part (or all) of a specific type of stationary object (such as part (or all) of a road or lane-side barrier or guardrail).

[0004] The disclosed technology allows for the reliable and time-saving removal of sensor detections such as guardrails / roadblocks during the preprocessing stage of batch processing. This reduces the possibility of misclassifying other objects and generating similar sensor detections, leading to errors. Such similar sensor detections include, for example, radar detection of temporarily stationary objects such as vehicle platoons, which can extend along roads / lanes in a similar manner (especially at side intersections on relatively straight road sections). The disclosed technology reduces the amount of data including sensor detections during the preprocessing stage, which would otherwise require the processing components of the perception and / or control systems to analyze the sensor detections in greater detail to distinguish between temporarily stationary objects and truly stationary objects in order to better understand the autonomous vehicle's environment. Background Technology

[0005] Autonomous or driver-assisted vehicles are equipped with sophisticated sensor systems to allow their control systems to understand the vehicle's environment, such as its position and lane location, enabling it to detect and navigate around potential obstacles. Autonomous vehicles traveling (or stationary) along highway lanes face the challenge of reliably distinguishing between moving and stationary sensed objects within the sensor stream generated by their systems. Moving objects, which may represent other vehicles, pedestrians, or animals, can be monitored and tracked, and predicted paths can be generated to avoid collisions. Stationary objects, on the other hand, can represent buildings, lane dividers, side streets, bridges, disabled vehicles, side walls, and guardrails, and are typically not tracked in the same way as moving objects.

[0006] One problem that perception systems must address is that a platoon of stationary vehicles can generate sensor detections that are easily confused with those of elongated structures such as roadblocks or guardrails. As roadblocks and guardrails move along the road, they can generate a large number of sensor detections via remote sensors, such as radar sensor systems, in the same way as a platoon of vehicles.

[0007] It would be desirable to be able to distinguish, as quickly as possible, large, stationary objects (e.g., slender structures forming guardrails or roadside barriers) from temporarily stationary objects (e.g., a line of stationary vehicles). However, the sheer number of such detections that can be sensed within a sensing cycle makes it problematic to achieve this safely and reliably.

[0008] Known techniques for removing roadblock and guardrail detections to reduce the amount of sensed object detections requiring further processing and classification by the autonomous vehicle's perception and / or control systems may be unsatisfactory in various ways. For example, some techniques can process data fast enough but cannot reliably distinguish such stationary objects from pseudo-stationary objects such as temporarily stationary platoons of vehicles. Some techniques may take too long to reliably identify which sensed detections can be reliably removed. Some techniques may not be efficient enough, for example, the number of sensed detections reliably identified (i.e., those that can be removed with high confidence because they are truly stationary objects) is too low as a proportion of the total number of sensed detections generated per sensing cycle.

[0009] The disclosed technology seeks to remove and / or mitigate the problems of known stationary object detection techniques by providing a computer-implemented method that offers a reliable and rapid technique for detecting certain types of stationary objects. This technique is particularly suitable for large stationary objects with properties that result in a statistically significant number of sensor detections generated in any given sensing cycle to share certain features in the sensor data, as this allows for the safe and rapid removal of such sensor detections during the batch preprocessing phase of an autonomous vehicle perception system. This stationary object generation features a combination of sensor detections with certain characteristics that allow them to be rapidly distinguished from sensor detections of pseudo-stationary (i.e., temporarily stationary) objects (such as, for example, queued vehicles) during the preprocessing phase. Summary of the Invention

[0010] The summary section presents some features of the present disclosure in a concise form. Preferred aspects and embodiments of the present disclosure are defined by the appended independent and dependent claims, as well as one or more clauses of the description preceding the claims.

[0011] This disclosure includes examples of various embodiments of methods and related aspects relating to the detection, identification, and / or classification of sensor detection in a sensor detection stream, which includes the sensor detection of one or more stationary objects. Specifically, this disclosure relates to methods and related aspects that can be used to distinguish truly stationary objects from seemingly stationary but potentially moving objects detected by the autonomous vehicle's sensor array in the environment surrounding the vehicle.

[0012] This disclosed example method can be used to identify and / or classify sensor detections of one or more stationary objects, where the objects have physical properties that cause such combinations of sensor detections to share certain features or attributes, such as location attributes. While such shared attributes are detectable in individual sensor detections, it is only possible to have a higher confidence that they originate from the same object when processed as a combination. Advantageously, such a large combination of shared features or attributes allows each sensory stationary detection associated with a particular combination to be more reliably identified as a sensor detection of a specific type of stationary object. This allows sensory stationary detections to be processed as a batch of stationary detections or combinations of stationary detections, and distinguishes all stationary detections in a combination of stationary detections from other sensor detections of stationary or pseudo-stationary objects at the combination level.

[0013] Examples of stationary objects of this type with such physical properties include elongated object types such as roadblocks or guardrails. This type of structure is often found positioned alongside roads / lanes or can be provided in other locations along the paths traversed by autonomous vehicles. Such stationary objects typically possess regular physical properties, or exhibit a degree of regularity over a long distance alongside a road or lane. For example, support posts can often be provided at regular intervals to support roadblocks or guardrails along the lanes or paths traversed by autonomous vehicles. The gaps between sections of roadblocks and / or guardrails can also be regularly spaced and have consistent dimensions. This regularity in structural characteristics can result in a degree of regularity in the multiple sensing locations detected by long-range sensing systems such as radar systems for autonomous vehicles when detecting such large objects.

[0014] The range of large stationary objects, such as roadblocks or tracks, generates a large number of sensor detections in any sensing cycle. When these sensor detections are analyzed as a combination, the regular physical properties of such large stationary objects produce corresponding regularities. For example, if not obstructed by other sensed objects, roadblocks or guardrails can generate radar sensor detections in one or more of the autonomous vehicle's rear, front, and side areas. Some preferred embodiments of this disclosure take into account sensor detections from the entire surrounding environment of the autonomous vehicle. This allows for the joint analysis of a large number of sensor detections compared to using only sensor detections from the front of the vehicle, which increases the reliability of detecting truly stationary objects using the techniques of this disclosure.

[0015] Some disclosed exemplary embodiments of the preferred computer-readable storage medium are stored in a program configured to be executed by one or more processors of the vehicle control system, the program including instructions for performing a method according to any one of the embodiments disclosed in the specification, the claims, or the drawings.

[0016] Some disclosed example embodiments of preferred vehicles include a perception system comprising at least one sensor for monitoring the vehicle's surroundings. The perception system may be part of a control system or apparatus according to any of the embodiments disclosed herein. Vehicle data may be provided to the perception system and / or control system from other vehicle components (e.g., from autonomous vehicle motion sensing devices). For example, an inertial measurement system or unit (in... Figure 1A The IMU (shown as IMU 20) can measure the inertial movement of the vehicle, and the positioning system (e.g., such as...) Figure 1A The positioning system 22) can be used to monitor the vehicle's geographical location and heading.

[0017] The computer-readable storage medium and vehicle aspects of this disclosure have similar advantages and preferred features as the method aspects of this disclosure.

[0018] The term "non-transitory" as used herein is intended to describe computer-readable storage media (or "memory") that do not transmit electromagnetic signals, but is not intended to otherwise limit the types of physical computer-readable storage devices covered by the terms computer-readable media or memory. For example, the terms "non-transitory computer-readable media" or "tangible memory" are intended to cover types of storage devices that do not necessarily permanently store information, including, for example, random access memory (RAM). Program instructions and data stored in a non-transitory form on a tangible computer-accessible storage medium can be further transmitted via a transmission medium or a signal such as an electrical signal, electromagnetic signal, or digital signal, which can be transmitted via a communication medium such as a network and / or a wireless link. Therefore, the term "non-transitory" as used herein is a limitation on the medium itself (i.e., tangible, not signal-based), rather than a limitation on the persistence of data storage (e.g., RAM and ROM).

[0019] It should be emphasized that when the term "comprising / including" is used in this specification, the term "comprising / including" is used to specify the presence of the stated feature, integral, step, or component. It does not exclude the presence or addition of one or more other features, integrals, steps, components, or combinations thereof.

[0020] These and other features and advantages of the present invention will be further illustrated below with reference to the embodiments described in more detail below. Attached Figure Description

[0021] Further objects, features, and advantages of the present disclosure will now be described with reference to the accompanying drawings, which illustrate embodiments by way of example only, and in which:

[0022] Figure 1A This is a schematic side view of an example of an autonomous vehicle according to some embodiments disclosed herein;

[0023] Figure 1B It includes Figure 1A A schematic diagram of the functional components of an example control system for an autonomous vehicle shown;

[0024] Figure 2A It is based on Figure 1A and Figure 1B A schematic top view of an example environment for autonomous vehicles in the outer lane;

[0025] Figure 2B It is based on Figure 1A and Figure 1B A schematic top view of another example environment for autonomous vehicles in the inner lane;

[0026] Figure 3The diagram illustrates an example of environmental sensing detection by an autonomous vehicle.

[0027] Figure 4 The schematic map illustrates example histograms of lateral position groups of sense detection according to some embodiments of the methods for detecting stationary objects disclosed herein;

[0028] Figure 5 The schematic map illustrates an example of sensing detection in the environment of an autonomous vehicle, associated with the confidence level of sensing detection of a specific type of object;

[0029] Figure 6 The figures illustrate some embodiments of how to use the methods for detecting stationary objects disclosed herein, and examples of sensing detections that identify sensing detections as a specific type of object using confidence levels;

[0030] Figure 7 The figures illustrate example embodiments of a method for detecting stationary objects according to some embodiments of the techniques disclosed herein;

[0031] Figure 8 The figure illustrates another example embodiment of a method for detecting stationary objects according to some embodiments of the techniques disclosed herein;

[0032] Figure 9 The diagram illustrates additional steps performed in a method for detecting stationary objects according to some embodiments of the techniques disclosed herein;

[0033] Figure 10A The illustration shows an example of autonomous vehicle sensing detection prior to performing a method according to some embodiments of the techniques disclosed herein; and

[0034] Figure 10B The diagram illustrates how to remove a stationary object after some embodiments of the techniques disclosed herein have been executed. Figure 10A Some of the sensing detections shown are illustrated. Detailed Implementation

[0035] Those skilled in the art will understand that the steps, services, and functions described herein can be implemented using separate hardware circuitry, software running in conjunction with a programmed microprocessor or general-purpose computer, one or more application-specific integrated circuits (ASICs), and / or one or more digital signal processors (DSPs). It will also be understood that, when this disclosure is described according to a method, it can also be implemented in one or more processors and in one or more memories coupled to one or more processors, where one or more memories store one or more programs that, when executed by one or more processors, perform the steps, services, and functions disclosed herein.

[0036] In the following description of exemplary embodiments, the same reference numerals denote the same or similar parts. In this context, a vehicle is understood to be a road vehicle or an off-road vehicle. Examples of vehicles include automobiles, buses, trucks, or construction vehicles. The term "autonomous vehicle" refers to a vehicle equipped with suitable components providing at least one sensor system and a control system, the control system including a perception control system for implementing a method for detecting stationary objects according to embodiments disclosed herein.

[0037] Figure 1A This is a schematic side view of an example embodiment of an autonomous vehicle 10, which includes a vehicle control system 12 that performs methods according to any of the embodiments disclosed herein. Figure 1A The control system 12 shown includes a sensing system 14, one or more sensors 16, a camera system 18, a vehicle inertial measurement system 20 (shown as IMU 20 in Figure 1), a positioning system 22, a vehicle control system 30, and a data communication module 34 (in Figure 1A (shown as TX / RX in the image).

[0038] In some embodiments, Figure 1AThe control system 12 of the autonomous vehicle 10 shown is configured to implement a method for detecting stationary objects based on positional attributes of sensor detection combinations of one or more specific types of stationary objects. Some examples of this method include receiving sensor signal data, including stationary and non-stationary detections, from the autonomous vehicle's surrounding environment, such as in a sensor detection stream. The method determines at least one combination of stationary detections that satisfies one or more lateral position selection criteria based on the lateral position of each stationary detection from the direction the autonomous vehicle is facing. The method also determines at least one combination of stationary detections that also satisfies one or more combination regularity selection criteria and shares a similar lateral position relative to the longitudinal axis of the autonomous vehicle. An example of combination regularity selection criteria is a regularity selection criterion based on the positional difference between pairs of stationary detections sequentially positioned in a direction along the longitudinal axis of the autonomous vehicle (e.g., along the longitudinal axis in the direction the autonomous vehicle is facing and / or along the same axis but in a direction behind the autonomous vehicle). The method further includes determining one or more combinations of stationary detections that satisfy both the lateral selection criteria and the combination regularity selection criteria. Stationary detections in such combinations can then be identified in batches or classified as corresponding to elongated stationary objects of the same type. In some examples, the method is used to detect one or more combinations of stationary detections corresponding to at least one predetermined type of stationary object in each sensing cycle (where a sensing cycle refers to each cycle in which sensor signals are detected and processed). Then, each combination of sensing detections can be batch-labeled and retained or filtered (or otherwise removed) from the sensor signal data stream generated in each sensing cycle before the cycle stream of sensing detections is forwarded to one or more other data processing components of the sensing system 14 of the control system 12 for further analysis.

[0039] Some embodiments of the control system 12 include a speed determination device used by the sensing system 14 in some embodiments of the disclosed technology. In some embodiments, the speed determination device for monitoring the speed of the vehicle 10 includes a positioning system 22 and an inertial measurement unit 20 for generating data used by the sensing system; however, in some embodiments, only the positioning system 22 is used as the speed determination device.

[0040] A computer-readable storage medium storing one or more programs can be used to execute the methods disclosed herein, wherein one or more programs are configured to be executed by one or more processors of a vehicle control system on such an autonomous vehicle. One or more programs, including instructions for performing this method, can be used to configure the processor as software, or provided in the form of configuring processing steps of the method in a circuit.

[0041] Now we will describe it in more detail. Figure 1AThis is an example embodiment of the perception system 14 of the control system 12 shown. The perception system 14 receives raw sensor data from one or more sensors or a sensor array 16. This sensor data includes sensing detections from detection areas 32a, 32b, 32c, and 32d, which form a sensing environment 32 surrounding the autonomous vehicle 10. The sensing detections include both stationary object sensing detections and non-stationary object sensing detections. These are processed by the perception system 14 to implement methods for detecting stationary objects, including detecting one or more types of stationary objects. Figure 1A As shown in the embodiment, the sensing system includes a main processing unit 28 and a preprocessing unit 24. The preprocessing unit 24 includes... Figure 1A The schematically shown is a processing unit 26 for detecting stationary objects, which performs a method for detecting stationary objects according to embodiments of the invention disclosed herein.

[0042] Sensing system 14, such as preprocessing unit 24, first distinguishes between the sensing detection of moving objects and the sensing detection of objects that are stationary when they are sensed. In some embodiments of sensing system 14, this may be performed by preprocessing stationary object detection unit 26. For example, in some embodiments, the radial velocity for stationary detection is calculated by preprocessing unit 24 using suitable techniques (e.g., techniques known to those skilled in the art). For example, in some embodiments, a stationary check is performed within stationary object detection unit 26 by checking that the radial velocity reaches a threshold (e.g., 1 m / s). If the amount of noise generated by the sensors and the signal quality of the autonomous vehicle's motion (among other factors) support this (e.g., 0.5 m / s or less), the threshold can be lowered to a lower value.

[0043] In some embodiments, the sensing system 14 is further configured to fuse sensing detections from the sensor 16 with image data generated by the camera system 18. In some embodiments, the fusion of sensor detections, such as radar sensing detections, with camera images is performed by the processing unit 28. The fused radar sensing detections may include a filtered stream of sensing detections that have already been processed by the preprocessing unit 24.

[0044] In some embodiments, the preprocessing stage 24 is configured to use resources different from those used by the main processing unit 28. However, in some embodiments, including those where the preprocessing stage 24 uses information such as information related to the yaw rate and / or curvature of the autonomous vehicle fed back from the main processing unit 28, one or more processor and / or memory resources may be shared with the main processing unit 28. In some embodiments, the main processing unit 28 and / or the preprocessing unit 24 are configured to determine road curvature and / or autonomous vehicle yaw to control whether the preprocessing unit 26 performs a method for stationary object detection according to the disclosed embodiments. If the road curvature and / or yaw of the autonomous vehicle does not meet certain conditions to allow for the evaluation of the position of the sensor-detected combination in the autonomous vehicle coordinate system, in some embodiments, execution of the method is suspended (although in other embodiments, the method may be modified to take into account road curvature / yaw conditions). In this way, assuming that the road is substantially straight and the autonomous vehicle is substantially aligned with the road, for example, such that the longitudinal axis of the autonomous vehicle and the road are substantially parallel within one or more detection areas of the autonomous vehicle, the object recognition algorithm for implementing the method of detecting objects of a predetermined type will be executed by the stationary object detection component / module / unit 26 of the preprocessing stage 24.

[0045] In some example embodiments of the control system 12 disclosed herein, the preprocessing component 24 may be implemented as a separate component independent of the sensing system 14, rather than as... Figure 1A In the illustrated embodiment, it is part of the sensing system 14.

[0046] exist Figure 1A In the illustrated embodiment, some types of sensor signals (e.g., imaging data such as from camera / camera array 18) are sent directly to the main processing unit 28 of the sensing system 14. Sensor detections acquired from signals from one or more other types of sensors 16 are first preprocessed by the preprocessing unit 24 by performing a method for detecting whether the sensed object is stationary or moving. Various techniques are known in the art for determining whether a sensor, such as a radar sensor, that has emitted a sensing signal has sensed a moving or stationary object.

[0047] Examples of sensors 16 include lidar / radar sensors / sensor arrays that can provide long-range detection (e.g., up to 80 meters or even longer, up to 200 meters or more). Such long-range sensors 16 can generate a high-volume stream of sensed detections due to their range. By sending the sensed detections to the preprocessing unit 24, the stationary object detection unit 26 can determine which sensed detections are of a specific type of stationary object in a manner that allows them to be distinguished from sensed detections of other stationary objects.

[0048] In some embodiments of the sensing system 14, the preprocessing stage stationary object detection module 26, for example, uses one or more techniques disclosed herein to label sensed detections as sensed detections of a specific type of stationary object. In some embodiments, the labeled sensed detections as sensed detections of that specific type of stationary object are provided with an indication of a confidence score. In some embodiments, the label is provided as metadata appended to the data representing the sensed detections.

[0049] In some embodiments, once a sensing detection is determined to be a sensing detection of a specific type of object, it is removed from the sensing detection stream from the sensor processed by the preprocessing unit, such that the amount of sensor data received by the main processing unit 28 is less than the amount of sensor data it would receive if the sensor data were not preprocessed first. In this way, by removing the sensing detection of certain types of stationary objects before the sensor stream is forwarded to the main processing unit 28, the main processing unit 28 generates fewer erroneous objects and the computational load on the main processing unit 28 can be reduced.

[0050] In some embodiments, the stationary object detection component 26 of the preprocessing component 24 provides a stationary object classification confidence score, which classifies the sensed detection as a specific type of stationary object using a confidence indication. The classification score is binary, meaning that the sensed detection is either classified as a specific type of stationary object or not. In other example embodiments, the classification may include one or more intermediate levels of confidence / confidence score values.

[0051] In some embodiments, all sensed detections are labeled using a confidence level indicator, for example, they may be sensed detections of a specific type of object, they may be assigned high, medium, or low confidence levels, and all are forwarded to the main processing unit 28. In some embodiments, sensed detections are labeled only using a medium confidence level and are forwarded to the main processing unit 28 as candidate detections of that specific type of stationary object, and high-confidence sensed detections are removed / discarded.

[0052] In some embodiments, a particular type of stationary object includes an object of a predetermined type having one or more attributes, the one or more attributes being represented by one or more detectable features detected by one or more stationary sensors.

[0053] In some embodiments, the method is applied to multiple different types of stationary objects, wherein multiple different types of stationary objects share one or more attributes represented by one or more detectable features detected by one or more stationary sensors.

[0054] In some embodiments, the method is applied to a plurality of different specific types of stationary objects, wherein each different type of stationary object is associated with one or more attributes representing one or more detectable features detected by one or more stationary sensors.

[0055] In this way, the disclosed method, described as being related to identifying a particular type of stationary object, should also be considered suitable for distinguishing the sensing detection of multiple different types of stationary objects from the sensing detection of other stationary objects.

[0056] In some embodiments, if a sensing detection is detected or classified as a stationary object belonging to one or more predetermined types, such that the detection or classification of the sensing detection can be considered to correspond to a truly stationary object, then such sensing detection is filtered or otherwise removed from the sensing detection stream, which is passed to the main processing unit 28 for further analysis.

[0057] In some embodiments, the main processing unit 28 performs further analysis to distinguish between stationary and non-stationary objects and / or track non-stationary objects sensed to be in the environment 32 of the autonomous vehicle. In some embodiments, the processing unit 28 performs further processing of the received sensor data, including sensing detection, such as fusing the sensor detection data with camera image data and / or mappings and / or data provided by the inertial system 20 or the positioning system 22, to determine more information about the environment of the autonomous vehicle 10 and / or further detect stationary and / or non-stationary objects. The sensed non-stationary objects can then be tracked, and their movement can be predicted by the perception system 14 or another component of the control system 12.

[0058] Using preprocessing component 24 to remove sensor detections identified as corresponding to a specific type of stationary object from the data stream advantageously enables other components, such as processing component 28 of perception system 14, to generate information about the sensing environment 32 surrounding autonomous vehicle 10 using a smaller amount of data. This method also potentially allows analysis to use fewer computational resources and / or provide results faster, as data processing involves handling a smaller amount of data. This allows for... Figure 1A The vehicle control system 30 shown uses the output from the perception system 14, including information about the sensed environment 32 of the autonomous vehicle 10, to control the movement of the autonomous vehicle 10 in a more responsive manner, which can make the autonomous vehicle safer in certain rapidly changing environments 32.

[0059] Some embodiments of the perception system 14 using the method for detecting stationary objects receive raw sensor data from one or more cameras 18 and one or more types of long-range sensors 16 (e.g., lidar, radar, or ultrasonic sensors), where the stationary objects include one or more types of predetermined stationary objects as described. Data from the sensors 16 (e.g., radar or lidar sensors 16) is received only after the data has been preprocessed. Then, some embodiments of the perception system 14 fuse the sensing detections from the sensors 16 with image data from the cameras 18 to convert the raw signal data, including the sensing detections and images, into scene understanding, which is then output to other parts of the vehicle control system 12 (such as…). Figure 1A The vehicle control component 30 shown.

[0060] Some embodiments of the perception system 14 disclosed below also utilize positioning data and / or IMU data, and may also consider map data, for generating scene understanding of the environment of the autonomous vehicle 10. For example, in some embodiments, such as Figure 1A As shown, the positioning data is provided by positioning system 22, which is configured to monitor the geographic location and heading of vehicle 10, for example, using a Global Navigation Satellite System (GNSS) such as Geolocation System (GPS). However, in other embodiments of positioning system 22, other positioning technologies may be used instead or additionally. For example, in some embodiments, if higher accuracy is desired, the positioning data may alternatively be implemented using Real-Time Kinematics (RTK) GPS.

[0061] For example, in some embodiments, IMU 20 is configured to measure the inertial movement of an autonomous vehicle 10 with six degrees of freedom using three accelerometers and three gyroscopes; however, in some other embodiments of the disclosed technology, other IMU devices may be configured. Therefore, various combinations of positioning and IMU technologies are possible.

[0062] In some examples of some embodiments, IMU 20 and / or positioning unit 22 are configured to provide data to the main data processing unit 28 of the perception system 14 and other components of the control system 12, such as vehicle controller 30, or are configured to provide data to the main data processing unit 28 of the perception system 14 but not to other components of the control system 12, such as vehicle controller 30. However, in some embodiments, additionally or alternatively, some or all of the information generated by IMU 20 and / or positioning unit 22 is provided directly or indirectly to preprocessing unit 24 (e.g., the main processing unit 28 may provide this information to preprocessing unit 24).

[0063] Information data from IMU 20 and / or positioning unit 22 can be used to generate autonomous vehicle motion signals for use by the main processing unit 28 of the perception system. For example, the motion signals can provide information about the yaw rate of vehicle 10. In some embodiments, where the preprocessing unit has access to this data, the vehicle's yaw (i.e., its angular orientation relative to the longitudinal axis of the road) is used in a method for classifying stationary objects, according to some embodiments of the technology disclosed later below.

[0064] In some embodiments of the control system 12, vehicle data includes data related to the geographical location of the vehicle 10, the heading (or orientation) of the vehicle 10, and the speed of the vehicle 10. In some embodiments, vehicle data may be obtained, for example, directly from sources such as... Figure 1A The GPS unit of the positioning component 22 of the vehicle 10 shown is used. In some embodiments, this data may be used by the processing component 28 and / or the preprocessing component 24.

[0065] Figure 1A The environment 32 surrounding the autonomous vehicle 10 includes a sensing area within which sensors 16, 18 of the vehicle 10 can detect stationary and non-stationary objects. In some embodiments, the autonomous vehicle does not need to move when performing the method for detecting stationary objects.

[0066] Environment 32 in Figure 1A The area is represented by multiple sensor detection regions 32a, 32b, 32c, and 32d. Figure 1A In the example embodiments shown, multiple detection areas 32a, 32b, 32c, and 32d collectively provide near 360° or 360° coverage (i.e., 360 degrees in the field of view), enabling the method to utilize a greater number of detections than a sensor facing only forward can provide. When evaluating how consistently stationary detections can be aligned at different lateral distances from the autonomous vehicle 10, detections located behind the autonomous vehicle can be evaluated relative to the vehicle path. For example, at least one camera 18 used by the perception system 14 looks forward at area 32a and is configured to detect objects such as at least lane markings on a road, such as those in at least the forward detection area 32a shown in FIG. 1. Other directions can be covered by one or more cameras; for example, one or more cameras can face rearward (relative to the direction the vehicle 10 is facing) and thus provide coverage for the rear detection area 32, while other cameras can have fields of view covering the side detection areas 32b and 32d. Alternatively, in some embodiments, one or more omnidirectional cameras (or a camera array) can be used.

[0067] One or more other sensors 16 also provide a similar level of 360° or near-360° coverage. For example, sensor 16 may include radar, lidar, or ultrasonic sensors configured to provide sensor coverage in the front detection area 32a, the rear detection area 32c, and the side detection areas 32b and 32d, so that the 360-degree environment around the vehicle is fully sensed.

[0068] In some embodiments, the processor 40 forming the data processing unit 28 is configured to receive map data including lane geometry of the vehicle's surrounding environment in the form of candidate lane groups. The map data may be in the form of an HD map including information about a road with multiple parallel lanes, each of which has a centerline represented as a polyline. Polylines are typically two-dimensional, defined by the latitude and longitude of the start and end points of each line segment. Furthermore, the left and right lane markings representing lane boundaries are polylines. They are also associated with their marking type (e.g., dashed or solid lines). The map data may also include road demarcation symbols such as guardrails, speed limits, lane orientations, etc.

[0069] Radar sensor systems can have a significantly longer range than camera sensors (typically, camera range is 25 meters, but can reach 100 meters). For example, a typical radar sensor range might be 80 meters, but longer-range radar systems can support ranges from 80 meters to 200 meters or even greater. In some embodiments, the protected areas shown in Figure 1 may not extend laterally as they do in front and / or behind, but in other embodiments they may. In some examples, the actual sensing range may be limited by nearby objects obstructing the sensors.

[0070] like Figure 1A As shown, the autonomous vehicle 10 also has a suitable receiver / transmitter 34 that can connect to an external network 2 via a wireless data connection or link. In some examples, communication with the autonomous vehicle along the wireless link is bidirectional (i.e., data can also be uploaded from the autonomous vehicle using the same type of wireless channel) for example, to download map data from a distant source, but in some examples, different types of wireless connections are used. The same type of wireless connection channel (i.e., wireless link) or some other wireless link can be used to communicate with other vehicles 40 near vehicle 10 (see Figure 2 and...). Figure 3(Those in the text) or communicating with local infrastructure components. Examples of wireless connectivity include connections provided using cellular communication technologies, which can be used for long-distance communication such as with external networks, and, if the cellular communication technology used has low latency, can also be used for vehicle-to-vehicle, vehicle-to-vehicle (V2V), and / or vehicle-to-infrastructure (V2X) communication. Examples of cellular wireless technologies are GSM, GPRS, EDGE, LTE, 5G, 5G NR, etc., and also include future cellular solutions. However, in some solutions, short-to-medium range communication technologies are used, such as wireless local area networks (LANs), for example, solutions based on IEEE 802.11. ETSI is investigating cellular standards for vehicle communication, and, for example, 5G is considered a suitable solution due to its low latency and efficient handling of high bandwidth and communication channels.

[0071] In some embodiments, the preprocessing component 24 determines coordinates representing the sensing locations of stationary and non-stationary objects from the surrounding environment of the vehicle 10 in a suitable reference frame. The sensing locations of the objects can be given using coordinates based on the vehicle's own reference frame (e.g., provided in the reference frame of an autonomous vehicle) or coordinates based on a standardized "global" coordinate system.

[0072] A preferred example is to use a Cartesian coordinate system relative to the autonomous vehicle to determine the position measurements of the sensed detections. This allows the lateral position of the sensed detections to be given by, for example, their distance from the autonomous vehicle along the lateral axis, and the longitudinal position of the sensed detections to be given by their distance in the orthogonal direction (i.e., along the longitudinal axis of the autonomous vehicle).

[0073] The use of a reference frame based on the autonomous vehicle advantageously simplifies the geometry because it makes it easier to determine when senses share the same lateral or longitudinal position relative to the autonomous vehicle. The longitudinal distance position of the senses can be used to determine regularity criteria by calculating the distance differences between senses at selected lateral positions. For example, if senses are grouped based on their lateral positions, in one embodiment, a combination or groups of senses with the highest number are then sorted according to their longitudinal positions. This allows for the determination of differences between longitudinal positions. If the determined longitudinal differences are nearly identical for a particular lateral combination (i.e., if the senses at a specific lateral distance from the autonomous vehicle are longitudinally uniformly spaced), this indicates regularity in the longitudinal positions of the senses at that lateral distance from the autonomous vehicle.

[0074] Embodiments of the invention associate certain combinations of sensing detections that share the same lateral distance from the autonomous vehicle (within a tolerance level) and whose sensing detections of large (e.g., longer than the longitudinal sensor range ahead and behind the autonomous vehicle) elongated objects at that lateral distance exhibit regularity in their longitudinal direction (within another tolerance level). Examples of possible objects large enough to extend all or a significant distance within the sensor detection range of the autonomous vehicle include, for example, guardrails.

[0075] In some embodiments, the difference regularity criteria for lateral grouping of sensor detections include at least the following criteria: the longitudinal distance difference between the sensor detections in each selected lateral group and the autonomous vehicle is less than a predetermined threshold. For example, all differences can be binned, and peaks can be checked against known characteristics of roadblocks in the road segment where the autonomous vehicle is located to qualify the detection that the peak in the difference regularity corresponds to a roadblock and not to another object.

[0076] Some embodiments of the preprocessing unit 24 of the sensing system 14 are configured to perform a method of static object detection by detecting one or more types of static sensing objects based on the attributes of the characteristics of a particular type of object, the attributes of the characteristics of the particular type of object causing features in the sensing detection combination of that type of object.

[0077] Some embodiments of methods for detecting stationary objects analyze a combination of sensor detections, such as radar detections from the environment 32 of the autonomous vehicle 10. This allows for the analysis of a larger sample set of candidate stationary object detections at any given time, rather than utilizing only the forward-facing detection area. By examining the sensor detections of the autonomous vehicle's environment, certain types of objects with sufficiently large structures can be identified that are detectable within a range of locations within the environment. Specifically, at multiple different locations within the 360-degree environment 32, elongated structures (such as roadblocks or guardrails) along a road or lane will be sensed multiple times. This allows for the use of more sensor detections, and if the autonomous vehicle is currently or is traveling along a relatively straight path parallel to the sensed structure, objects such as... will be seen in the sensor detections at specific lateral positions relative to the autonomous vehicle. Figure 3 and Figure 5 The parallel trajectory shown.

[0078] At any given point in time, a large number of sensing detections resulting from the sensing of large structures in the environment of the autonomous vehicle can be used to quickly identify such detections and remove them from the group of sensing detections of the environment, provided that it is possible to quickly identify what attributes from such large structures would lead to the sensing of the environment. In other words, embodiments of the stationary object detection method disclosed herein analyze sensing stationary object detections from the environment surrounding the vehicle to determine shared characteristics of sensing combinations or groups of stationary objects. In the case where the structure is a guardrail or road barrier traveling parallel to the path of the autonomous vehicle 10, a shared characteristic of the sensing group of stationary objects of that guardrail or road barrier is a shared lateral distance from the path of the autonomous vehicle.

[0079] In some embodiments, vehicle data, providing geographic location (latitude, longitude), vehicle heading (yaw angle), and vehicle speed, is also received by the preprocessing unit 24 for determining vehicle yaw and / or road curvature. Vehicle data can be received from a positioning system 22 (e.g., a Global Navigation Satellite System, GNSS) and / or an inertial measurement unit 20 (IMU) (e.g., as shown in Figure 1) of vehicle 10, and in some embodiments may also include map data.

[0080] Some of the received sensor data includes sensor detections indicating the position of at least one stationary object detected within one or more detection areas 32a, 32b, 32c, 32d around the environment of vehicle 10. The stationary object may include a road reference, such as a lane marking, traffic sign, road edge, road barrier, or any other suitable landmark. In some embodiments, the position of one or more stationary objects is determined with reference to the vehicle's local coordinate system, but may alternatively or additionally be determined with reference to a global coordinate system, depending on the application and implementation choice.

[0081] In some example embodiments, before the main processing unit 28, which provides scene understanding of, for example, the environment of an autonomous vehicle, performs any additional analysis, the preprocessing unit 24 reduces the amount of raw sensor data acquired from sensors 16, 18 that is converted into sensor detections. In some examples, the data stream including raw sensor data is reduced by removing stationary sensing detections that include sensing detections from one or more different types of sensors such as, for example, radar sensors, lidar sensors, ultrasonic sensors, etc., using methods for detecting stationary objects of a specific type of predetermined object claimed herein.

[0082] The attached diagram Figure 1B Various components of a platform implementing a sensing system 14 according to some example embodiments are shown. For example... Figure 1BAs shown, the perception system 14 is implemented using one or more processors 40 (e.g., a general-purpose processing unit and / or one or more other processing units), a memory component 42, a sensor interface component 44, and a data communication interface component 46. The processor 40 may be implemented in hardware and / or software, and in some embodiments may be referred to as control circuitry 40 or control line 40. The processor 40 is configured to execute instructions stored in memory to perform one or more methods according to the method embodiments disclosed later below. In some embodiments, one or more of the components may be implemented at least as part of a distributed computer system, rather than entirely on the autonomous vehicle. While the currently preferred implementation of the obstacle detection algorithm of the examples of the methods disclosed herein includes executing the algorithm entirely on the autonomous vehicle, some or all of the performed steps may be implemented in a more distributed manner if communication speeds and connection reliability improve sufficiently in the future. Other applications currently available for use by the autonomous vehicle perception system that can be implemented on the autonomous vehicle or in a distributed manner include, for example, map processing.

[0083] The memory 42 used by the processor 40 includes one or more (non-transitory) computer-readable storage media for storing computer-executable instructions, which, when executed by one or more computer processors of the processing unit 28, are executed.

[0084] In some embodiments of the sensing system 14, the memory 42 used by the processor 40 includes one or more (non-transitory) computer-readable storage media, which are also used by the processor 40 implementing the preprocessing unit 26, for example, to store computer-executable instructions that, when executed by one or more computer processors of the preprocessing unit 26, cause the data preprocessing unit 26 to perform techniques such as those of example embodiments of the method for detecting stationary objects described below. However, in some embodiments, the preprocessing system 26 uses different memory and / or processors than those used by the main processing system 28, or in addition to those used by the main processing system 28.

[0085] Examples of suitable forms of memory 42 include high-speed random access memory such as DRAM, SRAM, DDR RAM, or other random access solid-state storage devices. Other examples of memory 42 include non-volatile memory such as one or more disk storage devices, optical disk storage devices, flash memory devices, or other non-volatile solid-state storage devices.

[0086] In some embodiments, the processor 40 of the perception system 14 is configured to receive vehicle data, such as data from the IMU 20 and / or the positioning unit 22, using the data communication interface 46, and to receive sensor data from the camera system 18 via the sensor interface 44, and also to receive signals from the sensor system 16 via the signal preprocessing unit 24.

[0087] Turn now Figure 2A and Figure 2B , Figure 2A and Figure 2B The schematic map illustrates an example environment in which methods for detecting objects of a predetermined type can be implemented according to some disclosed embodiments. As mentioned above, when positioning along such a straight or near-straight road segment 50, the autonomous vehicle 10 will collect a continuous stream of sensor data (such as, for example, including sensing detection and one or more attributes of the sensing detection) from its sensor systems 16, 18.

[0088] Figure 2A and Figure 2B The images show a generally straight or nearly straight road segment 50 within the sensing environment 32 of an autonomous vehicle 10 having, for example, the control system described above. Figure 2A and Figure 2B The common diagram illustrates non-static objects (e.g., such as...) Figure 2B The queue of vehicles 70a, 70b, 70c, 70d, etc. shown is incorrectly classified as stationary objects (e.g., such as...). Figure 2A The example scenario shown is a roadblock or guardrail (62). As mentioned earlier, while it is important that the removal of certain detected categories can be achieved by... Figure 1A The preprocessing component 24 shown executes quickly, but it is dangerous to misclassify the sensing detection of stationary objects when the objects are only temporarily stationary, for example, the objects may not be able to be tracked correctly afterwards.

[0089] Figure 2A The schematic map illustrates an example road segment 50, along which an autonomous vehicle 10, such as that shown in Figure 1, travels in the outer lane 60a in the first direction. Figure 2A As shown, road segment 50 includes other traffic vehicles 52 and 56. Figure 2AAs shown, vehicle 52 travels along lane 60c in a direction opposite to that facing the autonomous vehicle. Lane 60c travels substantially parallel to lane 60a, and traffic travels along lane 60c in a direction substantially opposite to that of the autonomous vehicle traveling along lane 60a. Lane 60c is typically separated from inner lane 60b by a traffic lane divider, which in some examples may take the form of an elongated object 54, along which traffic 56 travels along inner lane 60b in a direction substantially the same as that of autonomous vehicle 10 (which is shown positioned / traveling along lane 60a). Figure 2A In the example road segment shown, lane 60b is also separated from lane 60a by a suitable lane divider, such as a road barrier or guardrail 58.

[0090] exist Figure 2A In the example road segment shown, along the outer edge of the outer lane 60a, another elongated structure 68 serving as a lane edge marker for the outer lane 60a is illustrated. In some embodiments, the elongated structure 68 includes a road barrier or guardrail and may have a similar design to... Figure 2A The guardrails / barriers shown in Figures 58 and 54 have similar structures.

[0091] In some example embodiments, local, national, or international regulations may define the structural features of any edge guardrail 68 or lane divider 54, 58. Such regulations may vary depending on the type of road.

[0092] exist Figure 2A In the middle, the elongated structure 68 includes tracks 62 supported by regularly spaced pillars or supports 64a, 64b, 64c, 64d, and 64e. Several repeating structural elements 64a, 64b, 64c, 64d, and 64e are in... Figure 2A The center is shown positioned substantially parallel to the lane segment 60a along which the autonomous vehicle travels. Several regularly spaced features (such as struts or supports 64a, 64b, 64c, 64d) supporting continuous or near-continuous elements 62 are located within... Figure 2A Within the detection area 32 of the autonomous vehicle. In this example, such as Figure 2A As shown, the supports 64b, 64c, and 64d of object 68 are each spaced from the adjacent supports 64a…64e by a regular distance d. reg .

[0093] like Figure 2AAs shown in the example, at any given time point, the near 360-degree detection area 32 around the autonomous vehicle 10 correspondingly includes sensing detection of stationary features 64a, 64b to the rear of the vehicle, 64c to the side of the vehicle, and 64d and 64e to the front of the vehicle, as well as sensing detection from the non-stationary vehicle 56 to the side of the vehicle. Other sensing detection may include stationary features generated by lane dividers 54, 58.

[0094] Figure 2B The diagram illustrates another example of a road segment 50 including lanes 60a, 60b, 60c, where the autonomous vehicle 10 is now in the inner lane 60b of traffic traveling in the first direction, and a stationary traffic queue including stationary vehicles 70a, ..., 70e is shown occupying the outer lane 60a. Figure 2B The schematic map illustrates how an autonomous vehicle misclassifies the sensor detection of temporarily stationary vehicles 70a, ..., 70b as the sensor detection of permanently stationary elongated structures such as roadblocks or guardrails, because stationary vehicles 70a, ..., 70e can exhibit different characteristics over time. Figure 2A The near-continuous sensing detection trajectory of the elongated structure 62 schematically shown in the figure is similar to that of a near-continuous sensing detection trajectory.

[0095] The detected vehicle queues 70a, ..., 70e will also generate a large number of sensed stationary detections, also referred to herein as pseudo-stationary detections. For example, in each processing cycle, a large number of both queue features and obstacle type structures can be sensed at similar lateral locations. Therefore, it is advantageous if the sensing signals from sensor 16 can be preprocessed sufficiently reliably and quickly by preprocessing unit 24 to distinguish between truly stationary sensed object detections and pseudo-stationary sensed object detections. In some embodiments, truly stationary sensed object detections can then be flagged or discarded from the sensor stream to reduce the computational load on data processing unit 28 (e.g., for tracking moving objects). Discarding truly stationary objects does not necessarily mean that they are deleted or no longer stored in memory 42. In some embodiments, one or more other sensor processing applications use sensed stationary object detections to further analyze the environment of autonomous vehicle 10.

[0096] In some example embodiments, sensor 16 is a radar sensor, and the raw data includes multiple radar signal detections. Figure 1A The signal preprocessing unit 24 shown is configured to reduce the number of radar sensing detections received by the data processing unit 28 by performing one of the disclosed method embodiments, provided that any qualifying conditions of the method to be performed are met. Examples of such qualifying conditions include first determining that the autonomous vehicle is substantially parallel to, for example, Figure 2A and Figure 2BThe straight section shown moves in a straight direction. This type of road and driving provides suitable conditions for the position of the sensor detection given in a reference frame, which can be determined relatively quickly relative to the road geometry and / or the autonomous vehicle's driving position on the road.

[0097] Under these conditions, the amount of data in any sensor detection stream used by the sensing system to track moving objects and passed to the main processing unit 28 is reliably reduced because the method removes roadblock or guardrail detection, but does not remove vehicle queue sensing detection, since the common structure of the queue indicated by the queue sensor detection is not as regular in the reference frame based on the longitudinal and transverse axes of the autonomous vehicle as the common structure of guardrail detection in the same reference frame.

[0098] For example, in Figure 2B In this context, each vehicle can have a different vehicle length or the same vehicle length (in other words, such as...). Figure 2B As shown in the figure, D_vehicle_length_A ≠ D_vehicle_length_B ≠ D_vehicle_length_C ≠ D_vehicle_length_D, but theoretically it is possible for D_vehicle_length_A = D_vehicle_length_B = D_vehicle_length_C = D_vehicle_length_D. Similarly, the distance between adjacent vehicles in the vehicle queue (shown in Figure 2 as d2 ≠ d3 ≠ d4 ≠ d5), but theoretically d2 = d3 = d4 = d5 is certainly possible.

[0099] Because it is highly unlikely that D_vehicle_length = constant for all vehicles or vehicles in a queue to be regularly separated, making d2 = d3 = d4 = d5a constant, the disclosed method will be able to detect sensing (such as...) in almost all cases. Figure 2B The radar sensing detection of the convoy of vehicles 70a, 70b, 70c, 70d, and 70e shown in the diagram, along with objects such as guardrails or roadside barriers (illustratively shown as...) Figure 2A The sensing detection distinguishes the true static structure of the regular structure 62 shown in the figure.

[0100] In some embodiments, the sensed detections have metadata appended by the preprocessing unit 24 based on their possible classifications to reflect a confidence score or indicator for a particular type of sensed detection. This allows the processing unit 28 to consider the confidence level of certain types of sensed detections when processing the received data.

[0101] In some embodiments, the processor 40 (e.g., as...) Figure 1BThe steps (shown) are at least partially configured in a circuit to implement one or more or all of the steps of a method for determining stationary sensing detection according to the disclosed technology. In some embodiments, one or more or all of the steps forming the method are provided by executable instructions, which are included in a non-transitory computer-readable storage medium or other computer program product configured to be executed by one or more processors 40.

[0102] Figure 3 An example is shown, such as a diagram generated using raw data from radar sensor array 16, plotted in Cartesian coordinates relative to autonomous vehicle 10 (autonomous vehicle 10 is also shown as a reference point). However, it will be apparent to anyone in the art or of ordinary skill that other coordinate systems can also be used, and that the reference point can be other than autonomous vehicle 10. Figure 3 In the drawing, the drawing position is indicated by the longitudinal distance from the vehicle 10 along one axis and the lateral distance from the vehicle 10 along another axis. Figure 3 The diagram clearly shows a large number of sensor detections aligned in a direction that is substantially parallel to the longitudinal direction of the path traveled by the autonomous vehicle 10, such as when the autonomous vehicle is traveling along a substantially straight section of road that is substantially parallel to various roadblocks (this is illustrated in the example diagram by referring to a large number of detection counts at the same lateral position shared by the autonomous vehicle).

[0103] Figure 3 The sensor detections detected by the autonomous vehicle at the snapshot are shown, including reflections sensed in the rear detection area 32c and the front detection area 32a, as well as on the sides of the vehicle (e.g., within the side detection areas 32b, 32c). Figure 3 This illustrates a combination of stationary sensing detection methods that can determine the same lateral position shared by an autonomous vehicle 10 along certain straight road segments. For example, such as... Figure 3 As shown, at least four sensor detection clusters exist, and within each cluster or combination, the sensor detections share the same "y" value relative to the autonomous vehicle 10. Assuming the road geometry is substantially straight within the detection area 32, this provides a strong indication that some detections may be stationary obstacle signals. The preprocessing unit 24 can discard stationary obstacle signals to reduce the amount of sensor detections sent from the preprocessing unit 26 to the main processing unit 28 of the perception system 14.

[0104] Roadblocks can cause problems for vehicle tracking systems because obstacle sensing detection may produce erroneous trajectories, obstacle detections may be correlated with the true trajectories (including, for example, the trajectories of other vehicles that are only temporarily stationary), and a large number of obstacle detections may place an unnecessary load on the tracking system's processor.

[0105] Therefore, embodiments of the disclosed technology seek to identify sensing detections, such as radar detections, and remove roadblock detections in a preprocessing step if the sensing detections can be identified as roadblock detections with a sufficiently high confidence level.

[0106] Some embodiments of the disclosed technology are particularly suitable for autonomous vehicles traveling along straight or relatively straight road segments. For example, a relatively straight or substantially straight road segment is one along which any lateral curvature of the road does not unduly increase the computational load of determining the lateral position of a stationary detection relative to the longitudinal axis of the autonomous vehicle and / or relative to the longitudinal axis of the path the autonomous vehicle is traveling. Although the term "road segment" is used herein, the disclosed methods can be applied to any off-road scenario where a vehicle can travel along a similarly substantially straight path, such as in a parking lot.

[0107] As the autonomous vehicle 10 travels in a straight line, roadblocks, guardrails, or other similar elongated structures 62 that are substantially parallel to the path traveled by the autonomous vehicle within its detection area will have a similar lateral position relative to the autonomous vehicle within the detection area 32. If the sensing detections are grouped based on their lateral distance from the autonomous vehicle, these parallel detection trajectories should be detectable, as the highest peak should indicate structures with a consistent lateral distance from the autonomous vehicle, such as those represented by roadblock detections.

[0108] To determine which lateral groupings of the sensing detections might include the sensing detections of objects extending longitudinally through the sensing environment surrounding the autonomous vehicle, the sensing detection combinations are ordered relative to the lateral extent covered by the autonomous vehicle 10. Figure 4 The image shows an example of lateral sorting grouping for sensor detection. In the illustrated example, in Figure 4 The histogram clearly shows three peaks. Figure 4 In the figure shown, the left peak 72 has a low count (<10) of sensed detections. The central peak 74 in the cluster of three peaks has the highest count (>50), and the right peak 746 has a count between the lowest peak 72 and the highest peak 74 (shown here as approximately 30 counts). The numbers on the horizontal axis in the figure only indicate the distance to the autonomous vehicle, as they represent the number of histograms that have been grouped to generate the histogram. In this figure, lateral position 0 (i.e., the position of the autonomous vehicle) is located at the center of the figure and is represented by histogram number 25.

[0109] In some embodiments, sensor detections may be removed solely based on their lateral location. For example, in some embodiments, lateral boundaries are removed so that sensor detections that are too far or too close to the autonomous vehicle are removed. Typical edge locations are chosen based on the premise that detection processing resources are discarded above the far edge from the autonomous vehicle because fewer histograms need to be processed. Removing locations from inner edges reduces the risk of misclassifying stationary detections that are directly located in the autonomous vehicle's lane (i.e., not adjacent). In some embodiments, an example near-end boundary of 2.5 meters can be used so that sensor detections within this range are discarded. In some embodiments, an example far-end boundary of approximately 40 meters can be used so that sensor detections beyond approximately 40 meters of the far edge are also discarded / ignored. For example, in a preferred embodiment, the near-end boundary is 1 meter and the far-end boundary is 25 meters. Figure 4 In the example shown, several peaks can be identified within the lateral cutoff edge. If the histogram of the peak with the highest sensed detection (i.e., the combination) only includes detections from obstacle 62, the computational load on the main processing unit 28 can be reduced by removing sensed detections 74 from that group from the sensed detections streamed to the main processing unit 28 of the sensing system. Alternatively, they can remain in the stream but be forwarded along with additional metadata, such as tags or labels that indicate whether it is a "roadblock type" or a "candidate for roadblock type".

[0110] In some examples, the perception system 14 first receives sensing detections from one or more sensors 16, and then categorizes the sensing detections into those identified as non-stationary (i.e., moving) objects and those as stationary objects (also referred to herein as stationary object detection or simply as stationary detection 70). Next, in some embodiments, vehicle data and / or mapping data are used to first ensure that road geometry and vehicle yaw meet certain conditions before performing the method for stationary object detection according to embodiments disclosed herein. For example, before performing other steps in the method, it may be checked whether the straightness of road segment 50 (and / or lane 60) is sufficient and whether the autonomous vehicle 10's driving does not generate excessive yaw. In other words, before performing the method for detecting objects of a predetermined type, checks are performed to ensure that one or more conditions, such as a straight road, straight driving (or substantially straight road, substantially straight driving), are met. If the conditions are suitable, the preprocessing unit 26 performs a method for classifying the sensing detections according to one of the embodiments disclosed herein.

[0111] If the conditions are met, stationary detections 70 with similar lateral positions relative to the longitudinal axis of the autonomous vehicle are grouped into combinations. These lateral groups of stationary detections 70 are ordered according to the range of combined lateral positions covered by their positions relative to the autonomous vehicle 10. Each combination can be associated with a range of lateral positions, which can be a predetermined range or a dynamically determined range, for example, based on the overall range of the autonomous vehicle's sensing capabilities.

[0112] Next, determine which combination (e.g.) Figure 4 As shown, combination 74 includes the maximum number of stationary sensing detections. This may be determined in some examples as an absolute number, or in other examples as a relative peak value on adjacent combinations (e.g., by determining the crest factor of each combination). Once the criteria for selecting lateral combinations are met, the stationary sensing detections in that combination (e.g., in combination 74) are considered a combination to collectively represent at least a portion of an elongated structure traveling substantially along a substantially straight path 62 of the autonomous vehicle 10, which travels substantially along a road segment 50 within the sensing environment 32 of the autonomous vehicle. Such an elongated structure 62 may be, for example, a road barrier or guardrail traveling parallel to the road. Some embodiments require the crest factor of the combination to be above a threshold such that there is greater confidence that the combination of stationary detections sensed within the lateral range of that combination is associated with a structure such as a road barrier or guardrail.

[0113] While the above embodiments can allow for rapid detection of roadblock tracks based on the relative number of stationary detections at a specific lateral distance from the autonomous vehicle compared to the total number of detections at other lateral distances, this may not be a sufficiently reliable indicator in all traffic conditions.

[0114] If in relation to, such as Figure 2B In schematic depictions of dense traffic with temporary stillness in one or more adjacent lanes of an autonomous driving lane, some embodiments of the above methods are less effective in reliably reducing the number of obstacle detections. Very dense traffic areas can lead to problems such as congestion of autonomous vehicles and other vehicles queuing perpendicularly when in a standstill. This results in false stationary sensing detections, leading to histogram errors, which is problematic because it can cause the sensing detection of temporarily stationary vehicles to be misclassified as sensing obstacle detections.

[0115] In some embodiments, one way to address this issue is to modify the lateral extent of each combination relative to the autonomous vehicle 10. However, while a wider width allows for the use of more sensor detections to determine peaks, it also increases the risk of capturing non-obstacle detections. On the other hand, a smaller lateral extent for each sensor detection combination provides more accurate identification of sensor detections from obstacles, but if the obstacle is not perfectly parallel to the straight path the autonomous vehicle is traveling on or the orientation of the autonomous vehicle, there is a higher risk that sensor detections will be assigned to different combinations (which would flatten the peak combinations).

[0116] In some embodiments, the lateral range of the position relative to the longitudinal axis of the autonomous vehicle is not predetermined, but is determined based on the total lateral range of the sensing detection and the number of combinations associated with the sub-ranges of the lateral range. These combinations correspond to histograms in a histogram, and the typical lateral range of each combination is about 1.5 meters, with 50 combinations used, although this may vary with frame rate (as sensing detection forms a data stream over time).

[0117] In some embodiments, longitudinal constraints may be imposed to avoid the risk of misclassifying roadside sensor detections as roadblocks at a greater distance. An example of a longitudinal constraint is a distance of 50 meters from the autonomous vehicle. For example, in a preferred embodiment, the preferred number of histograms is 48, the width is 1 meter, and the longitudinal constraint is typically 100 meters in each direction.

[0118] In some embodiments, for example by Figure 4 The histogram depicted in the diagram represents a combination of detections that must satisfy one or more lateral position selection criteria so that the sensed detections in the combination are classified as belonging to an elongated structure such as a roadblock or guardrail. In some embodiments, such lateral position selection criteria include, for example, a first roadblock detection criterion and a second roadblock detection criterion.

[0119] Therefore, in some embodiments, the first lateral selection condition is related to the count of sensed detections that form a lateral combination exceeding a peak selection threshold, the lateral combination having a peak value exceeding the average count of sensed detections or at least exceeding the peak value of sensed detections forming one or more combinations of adjacent lateral ranges (i.e., exceeding). Figure 4 (The example histogram shows adjacent histograms). In some embodiments, the peak over averages are also referred to as the peak factor. Figure 4 In the example, the peak factor either exceeds the average peak threshold of 1.5.

[0120] In some embodiments, the second lateral selection detection condition is combined with a combination that also exceeds the count selection threshold (i.e., in...). Figure 4The number of sensed detections (within any given histogram of the example histogram) is related to the number of sensed detections. For example, in some embodiments, a count selection threshold of 10 can be used.

[0121] Some embodiments of the present invention impose prerequisites before performing the method. In some embodiments, the prerequisites include:

[0122] Sensing detection is the minimum stationary detection confidence condition for sensing detection of stationary objects (also referred to as stationary sensing detection in this paper); and / or

[0123] A maximum yaw rate threshold condition is established such that if the yaw rate exceeds the threshold, embodiments of the method for determining stationary objects as disclosed herein are not implemented to avoid problems when road sections are curved.

[0124] Some embodiments adjust the lateral position of the sensing detection to account for vehicle yaw, which allows for a narrower lateral range for each combination.

[0125] Some implementations distinguish between the left and right lateral sides of an autonomous vehicle when determining the lateral combination because there can be significant differences in the crest factor.

[0126] Some embodiments may not consider sensors located behind the autonomous vehicle in the count of combined sensors covering a range of lateral positions; however, by taking these sensors into account, greater confidence can be gained that the object is a very large longitudinal object such as a roadblock or guardrail.

[0127] Some embodiments implement the method for classifying stationary sensor detections in the preprocessor 26 instead of the main processor; however, this is not required in all embodiments. However, by performing the processing at the preprocessing stage, this method avoids impacting the resources required for the road estimation processing performed by the main processing unit 28.

[0128] The embodiments of the method for classifying sensed detections disclosed herein implement an algorithm that reliably and computationally efficiently addresses one or more technical problems that may arise when classifying the detection of specific types of stationary objects, such as roadblocks and guardrails (which can constitute the majority of detections that need to be processed by the perception system). For example, a large number of sensed detections can be generated by guardrails traveling parallel to the road, which can lead to erroneous trajectories on the guardrails, incorrect association of guardrail detections with existing vehicle trajectories, and overload of the vehicle control system CPU.

[0129] By enabling radar detection that identifies guardrails, tracking and sensor fusion algorithms can be executed more efficiently, and other algorithms such as road estimation algorithms can also benefit.

[0130] In some embodiments, the method for classifying sensed detections is performed only if the following prerequisites are met:

[0131] The road curvature is below a threshold (which can be determined using map data in some embodiments); and

[0132] When an autonomous vehicle is traveling along a road, it does not make any obvious turns, for example, if the autonomous vehicle is overtaking, which may happen (this can be determined by autonomous motion data, and during this period, the execution of the algorithm of the embodiment of the method for classifying stationary objects is prohibited).

[0133] In some embodiments, the autonomous vehicle is stationary when the algorithm of an embodiment implementing a method for classifying stationary objects is executed.

[0134] Some embodiments of a method for classifying stationary sensor detections will now be described, wherein the above-described method embodiments are performed, and furthermore, a second grouping of the stationary sensor detections is performed based on the longitudinal position attribute of the sensor detections.

[0135] In addition to the first lateral position roadblock condition and the second lateral position roadblock condition, which include lateral conditions and additional conditions as described above, embodiments of the invention will now be described, wherein the selection of a combination is additionally based on the regularity of the position of the stationary detection in the combination in the longitudinal direction from the autonomous vehicle. This regularity stems from the structural regularity of the stationary object itself, for example, from the support structure of a guardrail, which includes pillars spaced regularly along the guardrail.

[0136] In the following description, some embodiments of the method are disclosed, which relate to classifying stationary sensing detection as roadblock or guardrail detection. However, it will be apparent to anyone in the art or of ordinary skill that the method can be used to classify other types of elongated stationary objects that extend in a direction parallel to the longitudinal direction faced by the autonomous vehicle and exhibit structural characteristics similar to those of roadblocks or guardrails.

[0137] For the sake of brevity, the descriptions of the various embodiments described above will not be repeated herein, as the method steps may also be used in the embodiments, in which, in addition to grouping the stationary detections first based on their lateral position relative to the autonomous vehicle and selecting one or more combinations that satisfy the first lateral selection criterion and the second lateral selection criterion, the selected groups are further analyzed before classifying these combinations as combinations of stationary detections from large, slender structures such as guardrails.

[0138] In some examples, similar to the embodiments described above, where a histogram is used to represent the counts of stationary sensing detections at various lateral distances of 10 from the autonomous vehicle in the histogram (or combinations thereof), peak values, average peak values, etc., can be evaluated to determine whether they meet one or more lateral position selection criteria. However, if the lateral selection criteria are now met, the combination is only considered as a candidate combination of stationary detections.

[0139] Since other stationary objects can also give strong peaks in the first histogram, for example, in some cases, as mentioned above, vehicle queues can do so, in some examples, a second set of longitudinal combinations is generated, one for each guardrail candidate combination, based on the difference in longitudinal positions of the stationary detections that are sequentially separated for each candidate combination. In some embodiments, timestamp information is not required as the sensing system progresses in iterations, where new detections from approximately the same time are processed in each processing loop iteration. Detections belonging to a strong histogram in the lateral histogram are then sorted according to the longitudinal position and the difference determined between the longitudinal positions. For example, if the longitudinal (x-position) of each stationary detection is x0, x1, x2, x3, x4, then if x1 - x0 = dreg (e.g., as... Figure 2A As shown in the diagram, and x2-x1 = dreg and x3-x3 = dreg, and x4-x3 = dreg, then the regularity of each stationary detection is perfect. In practice, some or all of the differences in the longitudinal positions of sequential stationary detection pairs may differ by a small amount of Δd, but if Δd is small enough, i.e., if the intervals between consecutive stationary detections are sufficiently regular, this provides additional confidence that these stationary detections are detections of large, elongated structures such as roadblocks or guardrails. In other words, the algorithm determines the difference d0 = x1-x0, the difference d2 = x2-x1, etc., where x0, x1, x2 are the increasing longitudinal distances from the autonomous vehicle, and groups the differences so that a histogram of all differences d found for a particular lateral combination can be created. The difference d with the highest number of combination members (the peak in the histogram) can also be compared with information indicating the expected “d” of the roadblock or guardrail structural features that can be detected based on the autonomous vehicle's map location and / or road type.

[0140] In some example embodiments, the method also creates a second grouping in the form of a second histogram, which is created for the difference in longitudinal position. As mentioned above, this allows for greater confidence that structures such as guardrails with very regular constructions of metal components at fixed intervals will be distinguished from vehicle queues, even if the lateral positions are fairly consistent, vehicle queues generally do not exhibit the same degree of regularity in the longitudinal direction.

[0141] Therefore, the method according to these exemplary embodiments enables the more reliable determination of a combination of sensed detections representing a real guardrail (or any other elongated structure with similar structural regularity detectable by radar sensors) by analyzing longitudinal peak values. If these longitudinal regularity counts of the sensed detections form peaks in the histogram that satisfy one or more selection criteria, for example, if the counts of regular sensed detections belonging to lateral groupings or histograms form sufficiently high histogram peaks, they indicate sensed detections of objects with uniform or regular intervals having a fairly consistent lateral position from the autonomous vehicle, and thus can be more confidently classified as guardrails compared to methods that only consider lateral groupings of sensed detections.

[0142] Some example embodiments described herein classify stationary detections based on confidence scores. In some embodiments of methods for classifying stationary objects as, for example, roadside or lane-side guardrails or barriers, the coordinate system uses the autonomous vehicle origin located at the center of the autonomous vehicle's rear axle. For example, the Y-axis label value may refer to the longitudinal position relative to the autonomous vehicle origin, and the X-axis label value may refer to the lateral position relative to the autonomous vehicle origin. However, different reference frames and / or coordinate systems may be used instead of those used in some of the disclosed embodiments.

[0143] For example, such as Figure 5 As shown in the example, a detection map can be provided, wherein the longitudinal position is relative to the position of the autonomous vehicle along the longitudinal axis of the autonomous vehicle 10, and the lateral position includes the lateral position relative to the longitudinal axis of the autonomous vehicle 10. For example... Figure 5 As shown in the example diagram, at a certain point in time, multiple stationary detections 70 are detected within the detection area 32 of the autonomous vehicle. According to the techniques disclosed herein, stationary detections 78 that are discarded because they are too far or too close to the vehicle to detect guardrails or obstacles are not necessarily processed by the algorithm; they are not necessarily deleted from memory 42, as they may be processed by other parts of the perception system 14. However, stationary detections 80 are very likely to be detections of obstacle or guardrail type structures, while stationary detections 82 are identified as stationary detections of obstacle or guardrail type structures with a lower confidence level.

[0144] The two levels of static detection are classified as follows: Figure 5 As shown in the example diagram, Figure 5 The example image is generated based on a dual histogram, where sensed detections of stationary objects are first grouped laterally, and then classified only if each group contains sufficiently regular vertical distances between vertically adjacent sensed detections. Figure 5 In the example shown, at a given lateral distance (such as...) Figure 6(Illustratively shown) Two different thresholds for the peak values ​​in the longitudinal grouping of the distance between detections provide two confidence levels of score, combined to correspond to stationary objects. The process of applying the confidence scores will be described in more detail below.

[0145] Figure 6 The figure illustrates an example of how the method disclosed in some embodiments identifies a sensed detection based on a confidence score that the sensed detection is a specific type of stationary object. Figure 6 In the diagram, the vertical axis represents the peak value of each longitudinal distance group between detections within a histogram covering a specific lateral range from the detection location of the autonomous vehicle, and the horizontal axis represents the number of histograms covering the specific lateral range. Each histogram covering the specific lateral range can include detection samples of both stationary objects (such as roadblocks) and pseudo-stationary objects (such as vehicle queues). Figure 6 The figure illustrates high and low thresholds (this is optional). Objects belonging to the histogram with peaks above the higher threshold are considered stationary objects. In some embodiments, objects between the two thresholds are optionally labeled as candidate stationary objects. This allows more information about objects for which there is insufficient confidence to classify them as stationary objects (they would typically be removed in subsequent analyses such as object tracking), but retaining the information that they might be stationary objects can be useful for subsequent analyses, as these analyses can increase the confidence score that they are stationary objects.

[0146] See now Figure 7 This illustrates an example embodiment of a method for classifying stationary sensing detections to identify whether they are stationary objects of a predetermined type. Selection criteria are applied sequentially, and in different embodiments, lateral-then-regular or regular-then-lateral grouping may be performed to determine whether a sensing detection is, for example, an elongated object extending in a direction substantially parallel to the longitudinal direction faced by the autonomous vehicle. Examples of elongated objects include roadside or lane-side barriers or guardrails. In a preferred embodiment, sensing detections can be generated from sensors facing the rear, sides, and front of the autonomous vehicle to ensure a maximum number of sensing detections per sensing cycle.

[0147] exist Figure 7 In this method, the method includes receiving (step 700) sensor signals, such as signals generated by the radar sensor system of an autonomous vehicle. For example, as referenced above... Figure 4As described, the sensed detections can be grouped according to their lateral position relative to the autonomous vehicle 10. For example, as described above, combinations of sensed detections are then sorted according to the range of lateral position of each combination relative to the autonomous vehicle. Then, static detection combinations that satisfy one or more lateral combination selection criteria are determined (step 702) as candidate object combinations (e.g., guardrail combinations). The sensed detections can also be grouped according to the regularity of the intervals in the longitudinal position relative to the autonomous vehicle 10, and combinations of sensed detections can then be classified according to the range of each combination of the regularity of the intervals between the longitudinal positions relative to the autonomous vehicle. Combinations that satisfy one or more combination regularity selection criteria are also determined (step 704) as candidate object combinations (e.g., guardrail combinations).

[0148] Then, static detections belonging to combinations that satisfy the lateral and longitudinal regularity selection criteria for candidate object combinations can be classified as sensing detections of objects of a predetermined type associated with the lateral position and longitudinal regularity selection criteria (step 706). In other words, if static detection combinations that share the same lateral position relative to the autonomous vehicle in the longitudinal position range are also separated from each other in the longitudinal direction by regular intervals relative to the autonomous vehicle 10, they are likely to be sensing detections of elongated structures with regular structural features (such as guardrails that travel parallel to the path traversed by the autonomous vehicle mentioned above).

[0149] Therefore, based on the determination that the static detection combination meets two selection criteria, the combination as a whole is classified as a static detection guardrail combination, and... Figure 7 In the illustrated embodiment, those stationary detections are removed from the output of preprocessor component 26 (in step 708). If no combination satisfies both selection criteria, the stationary detections may remain in the output (step 710). In some embodiments, one or more adapted selection criteria may be utilized (this is in... Figure 7 (Not shown in the image) Repeat steps 700 to 706.

[0150] exist Figure 7 In the example embodiment of the method shown, stationary sensor detections are first grouped laterally, and then the lateral combinations of stationary detections are analyzed to satisfy longitudinal regularity of the intervals between the sensor detections. However, in other embodiments, it is possible to first analyze the sensor detections to locate combinations formed by the sensor detections that satisfy selection criteria based on their longitudinal positions relative to the autonomous vehicle that satisfy one or more regularity interval criteria, and then regroup these combinations of regularly spaced sensor detections based on their lateral positions relative to the autonomous vehicle.

[0151] However, embodiments that first determine the lateral position can advantageously allow for a more rapid reduction in the number of sensing locations that need to be analyzed for regularity in their longitudinal spacing, as stationary detections located outside the inner and outer lateral boundaries can be discarded. In some embodiments, longitudinal boundaries are also applied to quickly discard stationary detections associated with locations that are too far longitudinally from the autonomous vehicle (e.g., beyond + / - 100 meters).

[0152] An example embodiment of the method for classifying stationary objects (where lateral positions are first grouped before longitudinal intervals) includes determining at least one lateral combination of stationary detections that satisfy one or more lateral position selection criteria based on the lateral position from the direction the autonomous vehicle is facing, including each combination of stationary detections; determining one or more combinations of stationary detections that satisfy lateral selection criteria and combination regularity selection criteria for combinations of stationary detections corresponding to at least one stationary object; and then, for each at least one lateral combination of stationary detections, determining whether the lateral combination satisfies one or more combination regularity selection criteria; and if so, determining that the stationary detections in the lateral combination correspond to at least one stationary object by classifying the stationary detections in the lateral combination as stationary detections of that particular type of stationary object. In some embodiments, further steps may be implemented, such as applying confidence scores and / or filtering out certain combinations of sensed detections based on the fact that certain combinations of sensed detections represent stationary objects and do not need to be tracked.

[0153] Figure 8 An example embodiment of this method for detecting or classifying stationary objects is illustrated, wherein raw sensor data is first collected from sensors, for example, by obtaining radar signals from radar sensors (step 800). Radar sensor 16 may include an array of sensor arrays that, in some embodiments, collectively provide 360-degree coverage around the autonomous vehicle, or in other embodiments, may be provided by an omnidirectional radar array. This allows for the acquisition of almost all or all of the information surrounding the autonomous vehicle's environment. In each iteration, the latest detection received from each sensor is used to cover a 360-degree detection area around the autonomous vehicle (in some embodiments, this may be a near-360-degree detection area).

[0154] Then, in step (802), stationary (or candidate stationary) detections in the sensor data are determined. In some embodiments, stationary radar detections and non-stationary radar detections are distinguished by utilizing the autonomous motion speed compensation range rate, and the lateral and longitudinal positions of each stationary detection relative to the autonomous vehicle are determined (step 804). In some example embodiments, the method further includes a pre-filtering step 806, in which detections with a lateral position (which can be represented in Cartesian autonomous vehicle coordinates, for example by an abs(X) value, where an abs(X) value indicates a lateral position greater than a threshold (e.g., 40 meters) from the origin) are removed from the combination of stationary sensing detections, because these represent stationary objects that are not of high interest to the autonomous vehicle 10. As pre-filtering step 806, some embodiments of the invention also include removing detections with a longitudinal position at that point (e.g., given by an abs(Y) axis value less than a threshold (e.g., 1.5 meters), because this will ensure the removal of any detections in front of or behind the autonomous vehicle, since guardrails, etc., should not be found at such locations.

[0155] The remaining stationary detections 70 are then assigned to one or more combinations based on their lateral position relative to the autonomous vehicle 10 (step 810). In some examples, each of the one or more combinations is associated with a range of non-overlapping lateral positions, meaning that each stationary detection 70 belongs to one and only one combination. In some embodiments, a histogram of lateral distances from the autonomous vehicle is formed from the remaining detections, wherein each combination is represented by a histogram.

[0156] In some embodiments, one or more of the lateral combinations of stationary detections are then sorted relative to the position of the autonomous vehicle, and the combination of one or more lateral sorts can be selected based on the combined lateral position characteristics that satisfy one or more lateral combination selection criteria (step 812). For example, in some embodiments, a crest factor or average peak value is calculated for each sorted combination of sensed detections assigned to each histogram. In some examples, the mean of the left histogram and the mean of the right histogram are calculated separately. If a histogram has a crest factor exceeding a threshold (e.g., 4.0) and the number of sensed detections in the histogram exceeds a threshold (e.g., 10), then the detections in that histogram are classified as guardrail candidates.

[0157] For example, as described above, the method then determines the regularity of the distances between the stationary detection pairs in the direction faced by the autonomous vehicle 10 in each selected combination (step 814).

[0158] For example, in some embodiments, for each selected histogram where the number of sensed detections exceeds a threshold number and the histogram has a peak factor exceeding the threshold, the selected histogram or combination is considered a candidate combination of sensed guardrail detections. For each selected combination, for all sensed (e.g., radar) detections belonging to that candidate combination or histogram, a histogram of the differences between longitudinal positions (which in some embodiments may be represented as the differences between longitudinal position values ​​along the longitudinal axis of the autonomous vehicle) is created.

[0159] Next, Figure 8 The method shown determines whether the regularity of each selected lateral grouping of stationary detections satisfies one or more selection regularity criteria as described above, such as step 816, by sequentially sorting the sensing detections longitudinally and determining the longitudinal position difference between adjacent stationary detections. This allows for the determination and plotting of the longitudinal position difference between pairs of stationary detections for each lateral grouping. In some embodiments, a histogram is created on the difference between the X values ​​of all radar detections belonging to one of the candidate combinations of one or more lateral selections.

[0160] For example, in some embodiments, if the strongest peak in the second histogram is above a high threshold (e.g., 20), the corresponding detection is classified (step 818) or marked as belonging to a predetermined type of stationary object, such as an elongated object like a roadblock or guardrail (see example illustrating this point). Figure 6 In other words, if most of the sensing detections at a specific lateral distance from the autonomous vehicle present a regular position in the longitudinal direction (i.e., in a direction parallel to the longitudinal axis of the autonomous vehicle), some embodiments of the method disclosed herein for detecting stationary objects of a predetermined type consider this lateral group of stationary detections as a group of stationary objects that may all belong to guardrails or barriers or similar types of objects. In some embodiments, this allows the entire lateral group of stationary detections to be discarded at the group level before any additional analysis is performed on the entire lateral group of stationary detections (e.g., by the main processing unit 28) (step 820). In other words, in some example embodiments of the method, stationary detections that are detected (or classified) as stationary objects of a predetermined type (i.e., barrier or guardrail detections) are removed from the output (step 820).

[0161] In some other example embodiments of the method, instead of simply discarding, a combined confidence score that can be associated with the sensed stillness detection in the combination is determined.

[0162] Figure 9 An example embodiment of the method is shown, wherein the step of determining whether the regularity of each selected combination satisfies one or more regularity selection criteria ( Figure 8Following step 816), a confidence indicator is determined for each stationary detection in the selected combinations that is classified as corresponding to a specific type of stationary object (step 902). Then, the stationary detections in each selected combination are classified as detections of that type of stationary object according to their determined confidence levels (step 904). Then, based on the confidence levels determined by satisfying a second set of regular selection criteria, the stationary sensing detections are marked as candidates for sensing detections of that predetermined type of stationary object for the output (step 906), and the marked candidates are retained in the output (step 908) instead of being discarded.

[0163] In some embodiments, in step 908, not all candidates for marking are retained. For example, if a regularity selection criterion can provide a first threshold, the sensing detection is automatically discarded and removed from the output if it exceeds the first threshold, and a second set of regularity criteria can provide a lower second threshold, the sensing detection is marked as a candidate for sensing detection of a predetermined object of that type if it exceeds the second threshold.

[0164] In some example embodiments, for instance, if the strongest peak in the second histogram (representing the regularity of the intervals of sensing detection in the longitudinal direction) is below a high threshold but still above a low threshold (e.g., 12), such as Figure 6 As shown, the corresponding sensing detection is marked as a candidate sensing detection for the predetermined type of object (i.e., if the object of the predetermined type is a fence, it is marked as a fence candidate detection). This information is still useful in subsequent algorithms even if there is a low confidence level of a real fence. If the peak value is below a lower threshold, no fence marking is performed.

[0165] In some embodiments, within each guardrail detection combination (i.e., within each guardrail perpendicular), the method also classifies dynamic radar detections within a specified longitudinal range (e.g., within ±5 meters) of the guardrail (relative to the autonomous vehicle) as belonging to the guardrail. The reason is that, even if the detection is of a stationary object on the side of the autonomous vehicle, the radar ranging rate may not indicate that it is stationary. However, such detections are highly problematic for tracking algorithms, so classifying them as guardrail detections is important.

[0166] Now back Figure 5 This shows, for example, by Figure 9 The steps shown illustrate an algorithm for classifying stationary detections. Figure 5In region 78a, black dots represent unclassified stationary or moving detections outside the lateral boundary, and black dots 78b represent unclassified stationary or moving detections within the proximal lateral boundary and outside any proximal region longitudinal boundary. White circles represent stationary sensing detections classified as belonging to guardrails with high confidence, and crosses "x" represent stationary sensing detections with lower confidence, which represent guardrails (here, guardrails are predetermined objects of a specific type mentioned in the preceding embodiments and figures). Figure 5 The location of autonomous vehicle 10 is also shown. Figure 5 In the middle, multiple sensors 80 and 82 are clearly aligned longitudinally in the direction parallel to the direction the autonomous vehicle is facing (in other words, they share the same lateral position).

[0167] When detection is grouped laterally (similar to, for example) Figure 4 The tests shown in histograms 13-16 and 36-37 have passed the peak factor check (see also...). Figure 8 In step 812), this detection will produce peaks, making them candidates for guardrails. For each of these six histograms, which include the combination of candidate sensed stationary object detections, a second histogram is created for the regularity check. Figure 4 In the example shown, in step (816), only histograms 13-16 will pass the regularity check, which means that the sensed detections in histograms 13-16 can be batched and discarded as detections of stationary guardrails. Conversely, although histograms 36-37 do not pass step (812), they do pass step 816, and therefore their members are marked as candidate guardrail detections.

[0168] Figure 10A The diagram illustrates an example of sensing and detection by an autonomous vehicle prior to performing a method according to some embodiments of the techniques disclosed herein. Figure 10A As shown, on the left and right sides of the autonomous vehicle 10 are some prominent dummy objects (also called phantom objects or phantoms), which are... Figure 10A The object is shown as a horizontal shadow and is removed after being executed according to the method of the disclosed technique.

[0169] Figure 10B The diagram illustrates how, after performing a method for detecting stationary objects of a predetermined type, such as guardrails or roadblocks, according to some embodiments of the technology disclosed herein, some sensor detections, shown as horizontal shadows, are removed (because they are phantoms, such as...). Figure 10A As shown in the image). Figure 10B As shown, the remaining objects are not phantoms, and this demonstrates how the disclosed method can effectively separate the number of detections that the sensing system 14 must track.

[0170] The present disclosure has been described above with reference to specific embodiments. However, other embodiments besides those described above are also possible and fall within the scope of this disclosure. Method steps different from those described above may be provided within the scope of this disclosure, and the method may be performed by hardware or software. Therefore, according to exemplary embodiments, a non-transitory computer-readable storage medium is provided to store one or more programs configured to be executed by one or more processors of a vehicle control system, the one or more programs including instructions for performing the method according to any of the above embodiments. Alternatively, according to another exemplary embodiment, a cloud computing system may be configured to perform any of the methods presented herein. A cloud computing system may include distributed cloud computing resources that jointly perform the methods presented herein under the control of one or more computer program products.

[0171] Generally, computer-accessible media can include any tangible or non-transitory storage medium or memory medium, such as electronic, magnetic, or optical media (e.g., a disk or CD / DVD-ROM coupled to a computer system via a bus). The terms “tangible” and “non-transitory” as used herein are intended to describe computer-readable storage media (or “memory”) excluding those that transmit electromagnetic signals, but are not intended to otherwise limit the types of physical computer-readable storage devices covered by the terms computer-readable media or memory. For example, the terms “non-transitory computer-readable media” or “tangible memory” are intended to cover types of storage devices that do not necessarily permanently store information, including, for example, random access memory (RAM). Program instructions and data stored in a non-transitory form on a tangible computer-accessible storage medium can be further transmitted via a transmission medium or signals such as electrical signals, electromagnetic signals, or digital signals, which can be transmitted via communication media such as networks and / or wireless links.

[0172] The processor 40 associated with the perception system 14 (and the processor associated with the general control system 12 of the vehicle 10) may be or include any number of hardware components for performing data or signal processing or for executing computer code stored in the memory 42 of the perception system 14 and / or the control system 12.

[0173] Memory 42 may be one or more means for storing data and / or computer code for performing or implementing the various methods described herein. Memory may include volatile or non-volatile memory. Memory may include database components, object code components, script components, or any other type of information structure for supporting the various activities described herein. According to exemplary embodiments, any distributed or local storage device may be used with the systems and methods of this specification. According to exemplary embodiments, memory 42 may be communicatively connected to processor 40 (e.g., via circuitry or any other wired, wireless, or network connection) and includes computer code for performing one or more processes described herein.

[0174] It should be understood that Figure 1B The sensor interface 44 shown may also provide the possibility of acquiring sensor data directly or via dedicated sensor control circuitry in vehicle 10. Data communication interface 46 may include one or more antenna interfaces, which further provide the possibility of transmitting output from sensing system 14 to a remote location (e.g., a remote operator or control center) via vehicle antennas / antenna arrays. Furthermore, some sensors in vehicle 10 may communicate with control system 12 using a local network setup such as CAN bus, I2C, Ethernet, fiber optics, etc. In some example embodiments, communication interface 46 is accordingly arranged to communicate with other control functions of vehicle 10 and also serves as a control interface; however, a separate control interface (not shown) may be provided. Local communication within the vehicle may also be wireless, using technologies such as Wi-Fi, LoRa, Zigbee, Bluetooth, or similar medium / short-range technologies.

[0175] Therefore, it should be understood that parts of the described solution can be implemented in a vehicle, in a system located outside the vehicle, or in a combination of inside and outside the vehicle; for example, in a server communicating with the vehicle, a so-called cloud solution. For example, sensor data can be sent to an external system, and this system performs steps that define the cost of transitioning from one state to another. Different features and steps of the embodiments can be combined in combinations other than those described.

[0176] exist Figure 1A The main processing unit 28 and the preprocessing unit 26 shown in the above description of the embodiments of the present invention can be implemented at least partially by circuits and can share some or all of the circuits, or can be implemented by separate circuits.

[0177] Preferred exemplary embodiments of the methods, computer-readable storage media, control devices, and vehicles are set forth in summary in the following clauses:

[0178] Clause A. A control system 12 for an autonomous vehicle 10 is configured to implement a method comprising:

[0179] Receives 700,800 sensor signal data, including stationary and non-stationary detections of the surrounding environment of the autonomous vehicle;

[0180] Based on the lateral position of each stationary detection from the direction facing the autonomous vehicle 10, at least one combination 74, 76 of the stationary detections 70 that satisfy one or more lateral position selection criteria is determined.

[0181] Based on the regularity of the position difference between stationary detection pairs sequentially positioned in the combination of directions faced by the autonomous vehicle, 704 determines at least one stationary detection combination that satisfies one or more combination regularity selection criteria.

[0182] Determine 706 one or more static detection combinations 74 that satisfy the lateral selection criteria and combination regularity selection criteria used for combinations of static detections corresponding to at least one static object; and

[0183] 708 Removes stationary detection corresponding to at least one predetermined type of stationary object from sensor signal data of one or more data processing units 26, 28 output to control system 12.

[0184] Clause B. The control system 12 of Clause A includes control circuits 24, 26, and 28, wherein the control circuits are configured to implement the method.

[0185] Clause C. The control system 12 of Clause B includes control circuits 24, 26, and 28, wherein the control circuits are reconfigurable using computer program software to implement the method.

[0186] Clause D. A control system under any of the preceding clauses, wherein the autonomous vehicle is located in a lane of a road segment.

[0187] Clause E. A control system of any of the preceding clauses, wherein an autonomous vehicle moves along a lane in a direction substantially parallel to the direction of the road segment.

[0188] Clause F. A control system of any of Clauses D and E, wherein the control system is configured to implement the method of any of Clauses A through C if the curvature of the road does not exceed a threshold such that the road segment is substantially straight, and if the longitudinal axis of the autonomous vehicle is aligned with the longitudinal axis of the substantially straight road segment as the autonomous vehicle moves along the road segment.

[0189] Clause G. The control system of Clause F, wherein if the curvature of the road exceeds a threshold such that the road segment is no longer substantially straight, or if the longitudinal axis of the autonomous vehicle is no longer aligned with the longitudinal axis of the substantially straight road segment as the autonomous vehicle moves along the road segment, the control system is configured to stop implementing any of the methods in Clauses A through E.

[0190] Clause H. In a control system pursuant to any of the foregoing clauses, removal step 706 includes a filtering step.

[0191] Clause I. A control system of any of the preceding clauses, wherein determining (706) one or more combinations of static detections that satisfy a lateral selection criterion and a combination regularity selection criterion for combinations of static detections corresponding to at least one stationary object, includes:

[0192] Based on the lateral position of each combination of stationary detections including those from the direction facing the autonomous vehicle, determine (702) at least one lateral combination of stationary detections that satisfies one or more lateral position selection criteria; and

[0193] For each of at least one lateral combination of static detections, determine (704) whether the lateral combination satisfies one or more combination regularity selection criteria, and if so, determine that the static detection in the lateral combination corresponds to at least one static object.

[0194] Clause J. A control system of Clause I, wherein the method first identifies lateral combinations of objects, and then performs a rule-based check only on pairs of objects in the selected lateral combinations.

[0195] Clause K. In any of the preceding clauses, the control system wherein determining (704) at least one static detection combination that satisfies one or more combinational regularity selection criteria includes:

[0196] For each stationary detection in the at least one combination, determine the distance of the stationary detection location from the autonomous vehicle in the direction facing the autonomous vehicle;

[0197] Determine (814) the distance difference sets between sequentially arranged pairs of positions relative to the autonomous vehicle; and

[0198] Determine whether the distance difference group determined by (816) satisfies at least one distance regularity criterion.

[0199] Clause L. A control system for any of the foregoing clauses, wherein the distance regularity criterion includes at least a distance difference below a predetermined threshold.

[0200] Clause M. Clause L's control system, wherein zero distance difference represents a perfectly regular interval between sensor detections.

[0201] Clause N. A control system for any of the foregoing clauses, wherein the distance rule standard includes at least a difference within a predetermined difference range.

[0202] Clause O. The control system of any of the preceding clauses, wherein the control circuit of control system 12 is configured to determine (702) a combination of stationary detections that satisfy one or more lateral position selection criteria based on the lateral position of each stationary detection from the direction faced by the autonomous vehicle.

[0203] Clause P. The control system of any of the preceding clauses, wherein the control circuit is configured to:

[0204] Determine (804) the lateral position of the stationary detection relative to the direction facing the autonomous vehicle from the received signal;

[0205] The stationary detections are assigned (806) to the lateral combinations based on their lateral positions;

[0206] Lateral sorting (810) lateral combination is based on the lateral position of the lateral combination from the direction faced by the autonomous vehicle of each combination.

[0207] Determine whether the number of stationary detections in each lateral combination associated with the lateral position range meets a member number threshold; determine peak selection conditions for each lateral combination, such as the average peak value or peak factor of adjacent combinations in the lateral sorting combination; and

[0208] Determine whether the crest factor meets the crest factor threshold.

[0209] Clause Q. The control system of any of the preceding clauses, wherein removing (708) the stationary detection corresponding to a stationary object includes:

[0210] Each stillness detection (in the combination) is classified (818) as corresponding to a stillness object of a predetermined type; and

[0211] Stationary detection (combination) is filtered (820) from sensor signal data output to one or more data processing units (22) of the autonomous vehicle control system (12) after classification.

[0212] Clause R. Clause Q's control system, wherein the control circuit of the control system is configured as follows:

[0213] Determine (902) the confidence level of the regularity of the static detection corresponding to a static object of a predetermined type; and

[0214] Based on the confidence that the stationary detection corresponds to a stationary object of a predetermined type, the stationary detection (818, 910) is removed from the sensor signal data output to one or more data processing units (22) of the autonomous vehicle control system (12).

[0215] Clause S. A control system of any of the preceding clauses, wherein the control circuitry of the control system is configured to classify each stationary detection based on a determined confidence corresponding to a stationary object of a predetermined type (904).

[0216] Clause T. The control system of the foregoing clause, wherein the method further includes outputting an indication of the classification of stationary detection to other processing components of the control system of vehicle 10.

[0217] Clause U. The control system of any of the preceding clauses, wherein the stationary object of the predetermined type includes a road barrier or guardrail or another elongated roadside structure or a similar structure located adjacent to a road segment in the direction faced by the autonomous vehicle, the similar structure having a substantially uniform repeating structure capable of producing appropriate regularity in the sensing of stationary detection.

[0218] Clause V. The control system of any of the preceding clauses, wherein the control circuitry includes preprocessing units 24, 26 configured to provide outputs to one or more other data processing units 28 of the control system 12.

[0219] Clause W. The control system of the foregoing clause, wherein one or more other data processing components of the control system 12 include one or more components of the following types:

[0220] Sensor data fusion processing component;

[0221] Object tracking processing component; and

[0222] Road estimation component.

[0223] Clause X. The control system of any of the preceding clauses, wherein the sensor signals include radar signals generated by an array of one or more radar sensors (16) of the autonomous vehicle (10).

[0224] Clause Y. A vehicle (10) comprising:

[0225] Speed ​​determination device, used to monitor vehicle speed;

[0226] Sensing system 14 includes at least one sensor for monitoring the environment surrounding the vehicle; and

[0227] The control system 12 according to any one of the preceding claims.

[0228] Clause Z. Vehicle (10) according to Clause Y, wherein the speed determination device includes an inertial measurement unit 20 and a positioning system 22.

[0229] Clause AA: A method implemented by the control system (12) of an autonomous vehicle (10), the method comprising:

[0230] Receive sensor signal data (700, 800) from the surrounding environment of the autonomous vehicle, including stationary and non-stationary detection.

[0231] Based on the lateral position of each stationary detection from the direction facing the autonomous vehicle, determine (702) at least one combination of stationary detections that satisfies one or more lateral position selection criteria;

[0232] Based on the regularity of the position difference between the stationary detection pairs that are sequentially positioned in the direction faced by the autonomous vehicle in the combination, at least one stationary detection combination that satisfies one or more combination regularity selection criteria is determined (704).

[0233] Determine (706) one or more combinations of static detections that satisfy the lateral selection criteria and the combination regularity selection criteria for combinations of static detections corresponding to at least one static object; and

[0234] Remove (708) stationary detection corresponding to at least one predetermined type of stationary object from sensor signal data of one or more data processing units (22) of the output to the control system (12).

[0235] According to the method of the foregoing clauses, determining (706) one or more combinations of static detections that satisfy the lateral selection criteria and the combination regularity selection criteria for combinations of static detections corresponding to at least one static object includes:

[0236] Based on the lateral position of each combination including stationary detections from the direction facing the autonomous vehicle, determine (702) at least one combination of lateral stationary detections that satisfies one or more lateral position selection criteria; and

[0237] For each of at least one lateral combination of static detections, determine (704) whether the lateral combination satisfies one or more combination regularity selection criteria, and if so, determine that the static detection in the lateral combination corresponds to at least one static object.

[0238] Clause AC. According to any of the preceding method clauses, determining (704) at least one static detection combination that satisfies one or more combination regularity selection criteria includes:

[0239] For each stationary detection in the at least one combination, determine the distance of the stationary detection location from the autonomous vehicle in the direction facing the autonomous vehicle;

[0240] Determine (814) the distance difference sets between sequentially arranged pairs of positions relative to the autonomous vehicle; and

[0241] Determine whether the distance difference group determined by (816) satisfies at least one distance regularity criterion.

[0242] Clause AD. According to the method described in the foregoing clause, the distance regularity criteria include at least the following criteria: in each selected lateral combination, for example, in the longitudinal direction relative to the autonomous vehicle, the difference in longitudinal distance between the sensed detections is less than a predetermined threshold. For example, consider a combination of stationary sensed detections all located within 3.0 meters and 3.05 meters laterally from the longitudinal axis of the autonomous vehicle. In this combination, the sensed detections are located at 10.00 meters, 12.01 meters, 14.05 meters, 16.03 meters, 18.30 meters, 20 meters, 22 meters, and 24.1 meters from the starting point along the longitudinal path. The longitudinal differences between adjacent detection pairs are 2.01 meters, 2.04 meters, 3.99 meters, 2.27 meters, 1.7 meters, 2 meters, and 2.01 meters. The number of pairs with longitudinal positional differences within 2.01 meters and 2.05 meters is 4 out of 7. Therefore, if the regularity selection threshold is more than 50% of the sensed detections sharing the regularity threshold, the combination meets the selection criteria, and all sensed detections in the combination will be considered associated with a large, stationary object such as a guardrail or road barrier track (or a similar stationary structure). If the threshold is 75% or more of the sensed detections are spaced between 2 meters and 2.05 meters, the combination cannot be discarded in a single batch preprocessing step.

[0243] Clause AE. According to the aforementioned methodological clauses AC or AD, the distance regulation standard includes at least the difference within a predetermined distance range. For example, in real-world scenarios, barrier track support structures may be spaced at certain intervals according to local, regional, national, and / or international regulations, and therefore the expected distance intervals for battery supports can be specified within a certain range, such as not less than 1 meter and not more than 10 meters. This means that if the example combination includes sensing detections at 8.2m, 9.1m, 10.00m, 11m, 12.01m, 13.3m, 14.05m, 15.02m, 16.03m, 17.10m, and 19.30m, the distance intervals will be b: 0.9m, 0.9m, 1m, 1.01m, 1.02m, 1.01m, 0.99m, 0.98m, 1.07m, and 2.2m. Although 80% of the combinations generate very regular sensing detections, the combination cannot be safely batched because it contains a first interval that exceeds the tolerance range of greater than 1m and less than 10m, as it only contains sensing detections of large, slender objects.

[0244] Clause AF. According to any of the preceding method clauses, wherein determining (702) a combination of stationary detections satisfying one or more lateral position selection criteria based on the lateral position of each stationary detection from the direction facing the autonomous vehicle including the control circuitry includes:

[0245] Determine (804) the lateral position of the stationary detection relative to the direction facing the autonomous vehicle from the received signal;

[0246] The stationary detections are assigned (806) to the lateral combinations based on their lateral positions;

[0247] Lateral sorting (810) lateral combination is based on the lateral position of the lateral combination from the direction faced by the autonomous vehicle of each combination.

[0248] Determine whether the number of stationary detections in each lateral combination associated with the lateral position range satisfies the membership threshold;

[0249] For each horizontally ordered static detection group that meets the minimum member threshold, determine whether the number of static detections in the group satisfies the peak threshold condition compared to the average number of static detections in one or more adjacent horizontally ordered groups.

[0250] Clause AG. According to any of the foregoing method clauses, wherein removing (708) the stillness detection corresponding to the still object includes:

[0251] Each stationary detection in the group that meets the selection criteria is classified (818) as a stationary sensing detection corresponding to a predetermined type of stationary object; and

[0252] The static detection combination is filtered (820) out of the sensor signal data output to one or more data processing units (28) of the autonomous vehicle control system (12) after classification.

[0253] Clause AH. The method according to any of the foregoing method clauses further includes:

[0254] Determine (902) the confidence level of the regularity of the static detection corresponding to a static object of a predetermined type;

[0255] Based on the confidence assessment corresponding to the stationary detection and the predetermined type of stationary object, remove (818, 910) the stationary detection from the sensor signal data output to one or more data processing units (22) of the autonomous vehicle control system (12); and

[0256] Outputs an indication of the classification and / or associated confidence assessment for each candidate stationary object sensing detection.

[0257] Terms AI. According to any of the foregoing method terms, the stationary object of a predetermined type is located adjacent to a road segment along the direction faced by the autonomous vehicle and includes one or more of the following:

[0258] Slender structures; and / or

[0259] Roadblocks; and / or

[0260] Guardrail.

[0261] Clause AJ. The method according to any of the preceding method clauses, wherein the method is implemented by a preprocessing unit (26) configured to provide output to one or more other sensor signal processing units (28) of the control system (12) of the autonomous vehicle (10).

[0262] Clause AK. According to any of the foregoing method clauses, one or more other data processing components of the control system (12) include one or more components of the following types:

[0263] Sensor data fusion processing component;

[0264] Object tracking processing component; and

[0265] Road estimation component.

[0266] Clause AL. The method according to any of the preceding method clauses, wherein the sensor signal includes radar signal generated by an array of one or more radar sensors (16) of the autonomous vehicle (10).

[0267] Clause AM. A control system configured to implement a method for detecting a stationary object according to any of the preceding clauses, the control system comprising:

[0268] A means for receiving (700) sensor signal data, including stationary and non-stationary detection, from the surrounding environment of an autonomous vehicle;

[0269] A means for determining (702) at least one combination of stationary detections that satisfies one or more lateral position selection criteria based on the lateral position of each stationary detection in the direction facing the autonomous vehicle;

[0270] A means for determining (704) at least one static detection combination that satisfies one or more combination regularity selection criteria based on the regularity of the position difference between the static detection pairs that are sequentially positioned in the direction facing the autonomous vehicle in the combination.

[0271] A means for determining (706) one or more combinations of static detections that satisfy lateral selection criteria and combination regularity selection criteria for combinations of static detections corresponding to at least one static object; and

[0272] A means for removing (708) stationary detection corresponding to at least one stationary object from sensor signal data of one or more data processing units (22) of the output to the control system (12).

[0273] Clause AN. A vehicle (10) comprising:

[0274] The control system (12) according to any of the foregoing control system clauses, wherein the control system (12) includes:

[0275] A speed determination device (20, 22) for monitoring the speed of a vehicle (10) includes an inertial monitoring unit (20) and a positioning system (22);

[0276] A perception system (14) for monitoring the surrounding environment (32) of the vehicle (10) includes at least one sensor (16, 18).

[0277] Clause AO. A method for detecting guardrails implemented by a control system (12) of an autonomous vehicle (10), the method comprising:

[0278] Receive (700) sensor signal data from the surrounding environment of the autonomous vehicle, including stationary and non-stationary detection;

[0279] Based on the lateral position of each stationary detection from the direction facing the autonomous vehicle, determine (702) at least one combination of stationary detections that satisfies one or more lateral position selection criteria;

[0280] Based on the regularity of the position difference between the stationary detection pairs that are sequentially positioned in the direction faced by the autonomous vehicle in the combination, at least one stationary detection combination that satisfies one or more combination regularity selection criteria is determined (704).

[0281] Determine (706) one or more static detection combinations that satisfy the lateral selection criteria and the combination regularity selection criteria as a combination of static detections corresponding to at least one guardrail; and

[0282] Remove (708) the stationary detection corresponding to the guardrail from the sensor signal data of one or more data processing units (22) of the output to the control system (12).

[0283] Clause AP. A computer-readable storage medium storing one or more programs configured to be executed by one or more processors of a vehicle control system, the programs including instructions for performing the methods according to Clause AO.

[0284] Clause AQ. The method according to any of the foregoing method clauses, wherein the method is performed when the autonomous vehicle is traveling or stationary on a road, in a lane on a road, in a lane on a multi-lane road, or in an off-road vehicle.

[0285] Clause AS. The method according to any of the preceding clauses, wherein sensor detection is the detection of one or more objects or a portion thereof, such as guardrails, barriers or similar structures, which are located parallel or substantially parallel to the path being traversed and / or recently traversed by the moving and / or stationary autonomous vehicle.

[0286] Clause AT. A method for detecting stationary objects implemented by a control system (12) of an autonomous vehicle (10), the method comprising: receiving (700) sensor signal data including stationary and non-stationary detections from the surrounding environment of the autonomous vehicle; determining (702) at least one combination of stationary detections satisfying one or more lateral position selection criteria based on the lateral position of each stationary detection in a direction facing the autonomous vehicle; determining (704) at least one combination of stationary detections satisfying one or more combination regularity selection criteria based on the regularity of position differences between pairs of stationary detections sequentially positioned in the direction facing the autonomous vehicle in the combination; determining (706) one or more combinations of stationary detections satisfying the lateral selection criteria and the combination regularity selection criteria for combinations of stationary detections corresponding to at least one stationary object; and removing (708) the stationary detections corresponding to at least one stationary object from sensor signal data output to one or more data processing units (22) of the control system (12). The method may be implemented by a perception system (14) of the control system (12) of the autonomous vehicle (10).

[0287] It should be noted that the word "comprising" does not exclude the presence of other elements or steps besides those listed, and the word "a" preceding an element does not exclude the presence of a plurality of such elements. It should also be noted that no reference numerals in the drawings limit the scope of the claims, the invention can be implemented at least in part by hardware and software, and certain "devices" or "units" can be represented by the same hardware.

[0288] While the accompanying drawings may show a specific order of method steps, the order of steps may differ from the order described. Furthermore, two or more steps may be performed simultaneously or partially simultaneously. For example, the step of receiving signals including information about movement and information about the current road scene may be interchanged based on a specific implementation. This variation will depend on the chosen software and hardware system and the designer's choice. All these variations are within the scope of this disclosure. Similarly, the software implementation can be accomplished using standard programming techniques with rule-based logic and other logic to implement various connection steps, processing steps, comparison steps, and decision steps. The embodiments mentioned and described above are given by way of example only and should not be construed as limiting the invention. Other solutions, uses, objects, and functions within the scope of the invention claimed in the patent embodiments described below should be apparent to those skilled in the art.

Claims

1. A method implemented by a control system (12) of an autonomous vehicle (10), the method comprising: Receive (700, 800) sensor signal data from the surrounding environment of the autonomous vehicle, including stationary and non-stationary detection. Based on the lateral position of each stationary detection from the direction facing the autonomous vehicle, determine (702) at least one combination of stationary detections that satisfies one or more lateral position selection criteria; Based on the regularity of the position difference between the stationary detection pairs that are sequentially positioned in the direction facing the autonomous vehicle in the combination, at least one stationary detection combination that satisfies one or more combination regularity selection criteria is determined (704). One or more static detection combinations that satisfy the lateral position selection criteria and the combination regularity selection criteria are determined (706) as static detection combinations corresponding to at least one predetermined type of static object; and Remove (708) the static detection combination corresponding to the at least one predetermined type of static object from the sensor signal data output to one or more data processing units (28) of the control system (12).

2. The method according to claim 1, wherein, Determining (706) one or more static detection combinations that satisfy the lateral position selection criterion and the combination regularity selection criterion for static detection combinations corresponding to at least one static object includes: Based on the lateral position of each combination of stationary detections including those in the direction faced by the autonomous vehicle, determine (702) at least one lateral combination of stationary detections that satisfies one or more lateral position selection criteria; and For each of at least one lateral combination of the static detection, determine (704) whether the lateral combination satisfies the one or more combination regularity selection criteria, and if so, determine that the static detection in the lateral combination corresponds to the at least one static object.

3. The method according to claim 1, wherein, Determining (704) at least one static detection combination that satisfies one or more combinational regularity selection criteria includes: For each stationary detection in the at least one combination, determine the distance of the position of the stationary detection from the autonomous vehicle in the direction facing the autonomous vehicle; Determine (814) the distance difference sets between the position pairs sequentially arranged from the positions of the autonomous vehicle; and Determine whether the distance difference group determined by (816) satisfies at least one distance regularity criterion.

4. The method according to claim 3, wherein, The distance regularity criterion includes at least a criterion that the distance difference between the sensed detections and the distance detected relative to the autonomous vehicle in each selected lateral combination in the longitudinal direction is less than a predetermined threshold.

5. The method according to claim 3 or 4, wherein, The distance regularity standard includes at least the distances within a predetermined distance range.

6. The method according to any one of claims 1 to 4, wherein, Based on the lateral position detected from each stationary position in the direction faced by the autonomous vehicle, including the control circuitry, a combination of stationary detections satisfying one or more lateral position selection criteria is determined (702), including: Determine (804) the lateral position of the stationary detection relative to the direction faced by the autonomous vehicle from the received signal; Based on the lateral position of the stationary detection, the stationary detection is assigned (806) to the lateral combination; Based on the lateral position of the lateral combination from the direction faced by the autonomous vehicle of each combination, the lateral combination is laterally sorted (810). Determine whether the number of stationary detections in each lateral combination associated with the lateral position range meets the member number threshold; For each horizontally ordered static detection combination that satisfies the member number threshold, determine whether the number of static detections in the combination satisfies the peak threshold condition compared to the average number of static detections in one or more adjacent horizontally ordered combinations.

7. The method according to any one of claims 1 to 4, wherein, Removing (708) the stillness detection combination corresponding to the at least one still object includes: Each static detection in the combination that meets the selection criteria is classified (818) as a static sensing detection corresponding to a static object of a predetermined type; and The stationary detection combination is filtered (820) out of the sensor signal data output to one or more data processing units (28) of the autonomous vehicle control system (12) after classification.

8. The method of claim 7, further comprising: Determine (902) the confidence level of the regularity of the stillness detection corresponding to the predetermined type of still object; Based on the confidence assessment corresponding to the stationary object of the predetermined type, the stationary detection is removed (820) from the sensor signal data output to one or more data processing units (28) of the autonomous vehicle control system (12); and Output an indication of the classification of each candidate stationary object sensing detection and / or the confidence assessment of the association of the classification.

9. The method according to any one of claims 1 to 4, wherein, The at least one stationary object (52, 58, 68) is positioned adjacent to a road segment along the direction faced by the autonomous vehicle and includes an elongated structure.

10. The method according to any one of claims 1 to 4, wherein, The method is implemented by a preprocessing unit (26) configured to provide the output to one or more data processing units (28) of the control system (12) of the autonomous vehicle (10).

11. The method according to any one of claims 1 to 4, wherein, The one or more data processing units (28) of the control system (12) include one or more of the following types of units: Sensor data fusion processing component; Object tracking processing component; and Road estimation component.

12. The method according to any one of claims 1 to 4, wherein, The sensor signals include radar signals generated by an array of one or more radar sensors (16) of the autonomous vehicle (10).

13. A control system configured to implement the method for detecting a stationary object according to any one of the preceding claims, the control system comprising: A means for receiving (700) sensor signal data, including stationary and non-stationary detection, from the surrounding environment of the autonomous vehicle; A means for determining (702) at least one combination of stationary detections that satisfies one or more lateral position selection criteria based on the lateral position of each stationary detection from the direction facing the autonomous vehicle; A means for determining (704) at least one static detection combination that satisfies one or more combination regularity selection criteria based on the regularity of the positional differences between the static detection pairs that are sequentially positioned in the direction facing the autonomous vehicle in the combination. A means for determining (706) one or more static detection combinations that satisfy the lateral position selection criterion and the combination regularity selection criterion for static detection combinations corresponding to at least one static object; as well as A means for removing (708) the static detection combination corresponding to the at least one stationary object from the sensor signal data output to one or more data processing units (28) of the control system (12).

14. A vehicle (10), comprising: The control system (12) according to claim 13, wherein the control system (12) comprises: The speed determination device (20, 22) for monitoring the speed of the vehicle (10) includes an inertial monitoring unit (20) and a positioning system (22). A perception system (14) for monitoring the surrounding environment (32) of the vehicle (10) includes at least one sensor (16, 18).

15. A computer-readable storage medium storing one or more programs configured to be executed by one or more processors of a vehicle control system, said one or more programs including instructions for performing the method according to any one of claims 1 to 12.

Images