A method and system for tracking occluded targets of an autonomous driving vehicle

Abstract

The present invention relates to the field of vehicle intelligent driving technology, and in particular to a method and system for tracking obscured targets of an autonomous driving vehicle. The present invention defines targets such as pedestrians on the road that may cause safety accidents as tracking targets, continuously tracks the position of the tracking targets, and detects potential obstructions on the road that may obstruct the tracking targets; when the tracking target is lost, it is detected and confirmed that the tracking target is obscured by the potential obstruction and stored in memory as an obscured target and a real obstruction. Through the above-mentioned obscured target memory storage strategy and real obstruction marking strategy, even if the autonomous driving system loses the visual information of the tracking target due to obstruction, it can still continue to execute the first driving decision such as slowing down and avoiding pedestrians based on the real obstruction, avoiding the safety problem of misjudging that the tracking target has left and ignoring its existence to update the driving behavior, thereby improving the consistency of autonomous driving target tracking and the safety of driving decisions.

Description

Technical Field

[0001] The present invention relates to the field of vehicle intelligent driving technology, and in particular to a method and system for tracking an obscured target of an autonomous driving vehicle. Background Art

[0002] Object detection and tracking based on real-time visual data is a core function in the environmental perception module of autonomous vehicle systems. Traditional object detection systems generally use a real-time visual data-driven mechanism, relying on continuous video stream input to identify and update the status of dynamic objects such as pedestrians and non-motorized vehicles.

[0003] However, this mechanism has a significant flaw: when a target is briefly obscured by a moving vehicle, static building, or other obstacle, the system, unable to obtain real-time visual information of the target, will immediately determine that the target is lost and invalidate the tracking tag. This processing method not only causes the tracked target to be incorrectly marked as "disappeared" due to the interruption of vehicle visual information, resulting in a high misjudgment rate; it also prevents the vehicle from generating continuous behavior predictions before the target reappears, misjudging road conditions and increasing the risk of accidents. For example, in complex urban scenarios, when a pedestrian is briefly obscured by a bus, the existing vehicle's autonomous driving system will lose and clear the pedestrian label, thereby ignoring the pedestrian's existence and updating driving behavior and route planning, making decisions such as accelerating or accelerating to avoid obstacles. If a pedestrian crosses the road during this period, the risk of a collision is easily increased.

[0004] Therefore, there is an urgent need for a method that can solve the problem of loss of visual information of the vehicle tracking target due to occlusion, and reduce the misjudgment rate and safety hazards of the autonomous driving system caused by this. Summary of the Invention

[0005] The present invention aims to provide a method and system for tracking occluded targets in autonomous vehicles, thereby improving the consistency of autonomous driving target tracking and the safety of driving decisions, and solving the problem of misjudgment of road conditions and safety hazards caused by loss of visual information of the tracking target due to occlusion in existing autonomous driving systems.

[0006] To achieve the above objectives, the present invention provides a method for tracking an obscured target by an autonomous driving vehicle, comprising the following steps:

[0007] Get several tracking targets;

[0008] Acquire potential obstructions based on the plurality of tracking targets, and then acquire an obstruction range corresponding to the potential obstructions;

[0009] For any of the tracking targets, the following tracking strategies are executed:

[0010] In response to a tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder;

[0011] executing a preset first driving decision based on the real obstruction;

[0012] In response to an occlusion target presence signal of the occlusion target, the occlusion target and the real occluder are cleared.

[0013] The aforementioned occluded target tracking method for autonomous vehicles defines pedestrians and other road users who could potentially cause accidents as tracking targets. The method continuously tracks the target's position and detects potential obstructions that could obstruct the target. When the target is lost, the system detects and confirms that the target is obscured by a potential obstruction and stores it in memory as both an occluded target and a real obstruction. This occluded target memory storage strategy and real obstruction marking strategy allow the autonomous vehicle system to continue executing its primary driving decisions, such as slowing down and yielding to pedestrians, based on real obstructions, even if the system loses visual information about the target due to occlusion. This avoids the safety issue of misjudging the target's departure and updating driving behavior based on its presence, thereby improving the consistency of autonomous vehicle target tracking and the safety of driving decisions.

[0014] Furthermore, the obtaining of several tracking targets includes:

[0015] Get the road centerline;

[0016] Acquire a plurality of moving targets, and acquire the moving direction and moving angle of each of the moving targets;

[0017] For any of the moving targets, if the moving direction satisfies a preset direction rule with the road centerline, and the moving angle satisfies a preset angle rule with the road centerline, the moving target is marked as a tracking target.

[0018] In this implementation, the road centerline is used as a reference benchmark. The road centerline is a benchmark that extends forward from the starting point along the center line of the road. Compared with the horizontal coordinate of the traditional Cartesian coordinate system, analysis based on the road centerline can more intuitively judge the relative relationship between the object movement and the center of the road. Specifically, this implementation determines the movement trend of the moving target through the road baseline. When the moving direction of the moving target moves from the outer edge of the road toward the center line of the road, that is, the moving direction of the moving target points from the outer side of the road to the center line of the road; and the angle between the moving angle of the moving target and the road centerline is close to 90°, it is considered that the moving target is at risk of being obstructed by potential obstructions in the road. At this time, the moving target is marked as a tracking target for continuous tracking. Through this implementation, key targets that need to be tracked are screened out from multiple targets in complex road conditions, which reduces the complexity of identifying and tracking obstructed targets and improves the efficiency of tracking obstructed targets.

[0019] Furthermore, the acquiring of potential occluders based on the plurality of tracking targets, and then acquiring an occlusion range corresponding to the potential occluders, includes:

[0020] Acquire a first distance interval based on all the tracked targets;

[0021] Acquire a plurality of obstacles and acquire a second distance of each obstacle;

[0022] For any obstacle, if the second distance satisfies the first distance interval, the obstacle is marked as the potential occluder.

[0023] In this implementation, the distance between each tracking target and the autonomous vehicle is obtained, thereby determining the distance interval between the tracking target and the autonomous vehicle. Obstacles located within this distance interval are located between the autonomous vehicle and the tracking target, and therefore pose a potential risk of obstructing the tracking target, causing the autonomous vehicle's visual information to lose sight of the tracking target. This implementation marks obstacles between the autonomous vehicle and the tracking target as potential obstructions, thereby filtering out key targets requiring attention from multiple obstacles in complex road conditions. This reduces the complexity of identifying and tracking obstructed scenes and improves the efficiency of tracking obstructed targets.

[0024] Furthermore, in response to the tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder includes:

[0025] Setting a detection buffer zone outside the shielding range;

[0026] In response to the tracking target losing the signal, acquiring the historical coordinates of the tracking target;

[0027] If the historical coordinates are located in the detection buffer, it is determined that the tracked target enters the occlusion range.

[0028] In this implementation, in order to improve the robustness of the vehicle-occluded target tracking method and reduce the interference of the autonomous vehicle's signal perception and signal acquisition errors on the tracking of the occluded target, an additional buffer zone is set outside the occlusion range. This buffer zone is used to further determine whether the tracked target lost from the vehicle's field of view has indeed entered the corresponding occlusion range. Specifically, if the tracked target is located in the detection buffer zone immediately before it is lost, since the moving direction and angle of the tracked target have been defined in the above steps, it can be confirmed that the tracked target has indeed entered the corresponding occlusion range. On the contrary, if the tracked target is not located in the detection buffer zone immediately before it is lost, the probability that the target signal is lost is due to the signal perception and signal acquisition errors of the driving vehicle, and further judgment needs to be made after the perception signal acquisition is updated.

[0029] Furthermore, before responding to the tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, and then storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder, the method further includes:

[0030] Obtaining several scene frames within a preset time period;

[0031] For any of the scene frames, if the tracking target does not exist in the scene frame, the scene frame is marked as a tracking target lost frame;

[0032] If the number of the tracked target lost frames meets a preset lost frame threshold, the tracked target lost signal is generated.

[0033] In this implementation, the judgment conditions of multi-frame perception are further set to reduce the error rate of the present invention. Specifically, multi-frame perception refers to the analysis and judgment of multiple scene data frames acquired over a period of time. Since the output of such a module such as perception or prediction cannot be completely accurate, when a certain frame is inaccurate due to sensor noise or other reasons, if only relying on the perception and prediction results of a single frame to make a decision, it will cause the decision to jump and oscillate. Therefore, compared with relying solely on the loss of the tracking target in one scene frame for judgment, this implementation method judges whether the tracking target is indeed lost or it is a vehicle signal acquisition error by analyzing the proportion of target loss frames in all scene frames acquired over a period of time. When the proportion of target loss frames in all scene frames acquired over a period of time is high, it is considered that the tracking target has indeed been lost, and the tracking target loss signal is generated.

[0034] Furthermore, after responding to the tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, and then storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder, the method further includes:

[0035] Obtaining the confidence level and real-time marking duration of the real obstruction;

[0036] If the real-time marking duration exceeds a first preset time, lowering the confidence level and executing a preset second driving decision based on the confidence level;

[0037] If the real-time marking duration exceeds a second preset time, the blocking target and the real blocking object are cleared.

[0038] In this implementation, in order to reduce the possible false detections and missed detections in the vehicle signal perception results themselves, this implementation performs a confidence level analysis on the real obstructions detected in the above steps. If the real obstruction has not been cleared for a long time, that is, the obstruction target corresponding to the real obstruction has not reappeared in the vehicle's field of view for a long time. At this time, it is considered that the vehicle system's early perception may have false detections and missed detections, marking the wrong real obstruction, or the obstruction has not been cleared for a long time due to a vehicle system error. At this time, the confidence level of the real obstruction is lowered, and a second driving decision such as slowing down, avoiding pedestrians, etc. is executed in addition to the first driving decision, and the real obstruction is considered to be a false detection or missed detection when the real-time marking time exceeds the second preset time, and it is cleared. Based on this implementation, the present invention can reduce the probability of false detection and missed detection in obstruction target tracking, thereby improving the accuracy of obstruction target tracking.

[0039] Furthermore, before clearing the occluding target and the real obstruction in response to the occluding target appearing signal of the occluding target, the method further includes:

[0040] Obtaining several scene frames within a preset time period;

[0041] For any of the scene frames, if the occluding target exists in the scene frame, the scene frame is marked as an occluding target existing frame;

[0042] If the number of frames in which the occluding target exists meets a preset frame threshold, the occluding target occurrence signal is generated.

[0043] In this implementation, the judgment conditions of multi-frame perception are further set to reduce the error rate of the present invention. Specifically, multi-frame perception refers to the analysis and judgment of multiple scene data frames acquired within a period of time. Since the output of such a module of perception or prediction cannot be completely accurate, when a certain frame is inaccurate due to sensor noise or other reasons, if only relying on the perception and prediction results of a single frame to make a decision, it will cause the decision to jump and oscillate. Therefore, compared with relying solely on the appearance of the occluded target in one scene frame for judgment, this implementation method judges whether the occluded target actually appears or it is a vehicle signal acquisition error by analyzing the proportion of frames with occluded targets in all scene frames acquired within a period of time. When the proportion of frames with occluded targets in all scene frames acquired within a period of time is high, it is considered that the occluded target has indeed appeared in the field of view of the vehicle, and the occluded target appearance signal is generated.

[0044] A second aspect of the present invention provides an occluded target tracking system for an autonomous driving vehicle, which includes a tracking target recognition module, a potential occluder recognition module, an occluded target tracking module, and a driving module; wherein:

[0045] The tracking target identification module is used to obtain a plurality of tracking targets;

[0046] The potential obstruction identification module is used to obtain potential obstructions based on a plurality of tracking targets, and then obtain an obstruction range corresponding to the potential obstructions;

[0047] The occluded target tracking module is used to execute the following tracking strategy for any of the tracking targets:

[0048] In response to a tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder; in response to an occlusion target presence signal of the occlusion target, clearing the occlusion target and the real occluder;

[0049] The driving module is used to execute a preset first driving decision according to the real obstruction.

[0050] In this implementation, targets such as pedestrians on the road that could cause safety accidents are defined as tracking targets. The target's position is continuously tracked, and potential obstructions on the road that could obstruct the target are detected. When the target is lost, the system detects and confirms that the target is obscured by a potential obstruction and stores it in memory as an obscured target and a real obstruction. Through this obscured target memory storage strategy and real obstruction marking strategy, even if the autonomous driving system loses visual information about the tracking target due to occlusion, it can still execute the primary driving decision, such as slowing down and avoiding pedestrians, based on the real obstruction. This avoids the safety issue of misjudging the tracking target's departure and updating the driving behavior without noticing its presence, thereby improving the consistency of autonomous driving target tracking and the safety of driving decisions.

[0051] Furthermore, the obtaining of several tracking targets includes:

[0052] Get the road centerline;

[0053] Acquire a plurality of moving targets, and acquire the moving direction and moving angle of each of the moving targets;

[0054] For any of the moving targets, if the moving direction satisfies a preset direction rule with the road centerline, and the moving angle satisfies a preset angle rule with the road centerline, the moving target is marked as a tracking target.

[0055] In this implementation, the road centerline is used as a reference benchmark. The road centerline is a benchmark that extends forward from the starting point along the center line of the road. Compared with the horizontal coordinate of the traditional Cartesian coordinate system, analysis based on the road centerline can more intuitively judge the relative relationship between the object movement and the center of the road. Specifically, this implementation determines the movement trend of the moving target through the road baseline. When the moving direction of the moving target moves from the outer edge of the road toward the center line of the road, that is, the moving direction of the moving target points from the outer side of the road to the center line of the road; and the angle between the moving angle of the moving target and the road centerline is close to 90°, it is considered that the moving target is at risk of being obstructed by potential obstructions in the road. At this time, the moving target is marked as a tracking target for continuous tracking. Through this implementation, key targets that need to be tracked are screened out from multiple targets in complex road conditions, which reduces the complexity of identifying and tracking obstructed targets and improves the efficiency of tracking obstructed targets.

[0056] Furthermore, the acquiring of potential occluders based on the plurality of tracking targets, and then acquiring an occlusion range corresponding to the potential occluders, includes:

[0057] Acquire a first distance interval based on all the tracked targets;

[0058] Acquire a plurality of obstacles and acquire a second distance of each obstacle;

[0059] For any obstacle, if the second distance satisfies the first distance interval, the obstacle is marked as the potential occluder.

[0060] In this implementation, the distance between each tracking target and the autonomous vehicle is obtained, thereby determining the distance interval between the tracking target and the autonomous vehicle. Obstacles located within this distance interval are located between the autonomous vehicle and the tracking target, and therefore pose a potential risk of obstructing the tracking target, causing the autonomous vehicle's visual information to lose sight of the tracking target. This implementation marks obstacles between the autonomous vehicle and the tracking target as potential obstructions, thereby filtering out key targets requiring attention from multiple obstacles in complex road conditions. This reduces the complexity of identifying and tracking obstructed scenes and improves the efficiency of tracking obstructed targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 1 is a flow chart of a method for tracking an occluded target by an autonomous driving vehicle provided by an embodiment of the present invention;

[0062] Figure 2 1 is a schematic diagram of a scenario of an occluded target tracking method for an autonomous driving vehicle provided by an embodiment of the present invention;

[0063] Figure 3 1 is a schematic diagram of another scenario of an occluded target tracking method for an autonomous driving vehicle provided by an embodiment of the present invention;

[0064] Figure 4 is a schematic diagram of an SL coordinate system provided by an embodiment of the present invention;

[0065] Figure 5 It is a structural diagram of an occluded target tracking system for an autonomous driving vehicle provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. It should be noted that the following detailed descriptions are all exemplary descriptions and are intended to provide further detailed descriptions of the present invention. Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those generally understood by those skilled in the art to which this application belongs; the terms used herein in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the description of the above-mentioned drawings, as well as any variations thereof, are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order.

[0067] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps and they may be executed in other orders. Furthermore, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed and completed at the same time, but may be executed at different times. Their execution order is not necessarily sequential, but may be executed in rotation or alternation with other steps or at least a portion of their sub-steps or stages.

[0068] Before describing the present application in detail with reference to the accompanying drawings and in combination with the embodiments, the technical background involved in the present application will first be described.

[0069] In the fields of intelligent transportation systems and autonomous driving, real-time status tracking and tag management of traffic participants are core components of environmental perception. Existing systems typically rely on visual sensor data to dynamically tag objects such as vehicles and pedestrians, and then update status information such as the object's position and speed based on continuous detection results.

[0070] However, traditional dynamic tag management systems have the following technical bottlenecks:

[0071] Label failure in occlusion scenarios: Mainstream object detection systems use a purely data-driven mechanism, and their label generation relies entirely on real-time perception data. When a target is temporarily obscured by a physical obstacle, the system loses visual information and immediately determines that the target is gone. This causes the label database to mistakenly remove valid target information, significantly increasing the risk of missed detections.

[0072] Lack of state persistence logic: Existing label update algorithms lack the memory and reasoning capabilities that mimic human driving cognition. When a target is invisible, the system cannot deduce a reasonable position based on its historical trajectory, nor can it maintain its prediction of the target's behavior patterns, resulting in the misidentification of reappearing targets as new objects.

[0073] Specifically, in complex urban scenarios, when a pedestrian is briefly blocked by a bus, the existing vehicle autonomous driving system will lose and clear the pedestrian label, thereby ignoring the pedestrian's existence and updating driving behavior and route planning, making decisions such as accelerating or accelerating around obstacles. If a pedestrian crosses the road during this period, the risk of a collision accident is likely to increase.

[0074] In order to solve the above technical problems, see Figure 1 , Figure 2 and Figure 3A first aspect of an embodiment of the present invention provides a method for tracking an obstructed target by an autonomous driving vehicle. The method is implemented by an autonomous driving vehicle and includes the following steps:

[0075] S1. Obtain several tracking targets;

[0076] S2. Obtaining potential obstructions based on the plurality of tracking targets, and then obtaining an obstruction range corresponding to the potential obstructions;

[0077] The blocking range is located between the blocking entrance and the blocking exit of the potential blocking object.

[0078] S3. For any of the tracking targets, execute the following tracking strategy:

[0079] S31, in response to a tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder;

[0080] S32, executing a preset first driving decision based on the real obstruction;

[0081] S33: In response to an occlusion target appearance signal of the occlusion target, clear the occlusion target and the real occlusion object.

[0082] Specifically, the tracking targets include but are not limited to pedestrians, animals, and small motor vehicles. When a real obstruction tag is in effect, the vehicle is forced to slow down or stop in the detection buffer zone around the obstruction entrance and exit, rather than accelerating to bypass it. This avoids ghosting scenarios, thereby improving the consistency of autonomous driving target tracking and driving decision safety.

[0083] The aforementioned occluded target tracking method for autonomous vehicles defines pedestrians and other road users who could potentially cause accidents as tracking targets. The method continuously tracks the target's position and detects potential obstructions that could obstruct the target. When the target is lost, the system detects and confirms that the target is obscured by a potential obstruction and stores it in memory as both an occluded target and a real obstruction. This occluded target memory storage strategy and real obstruction marking strategy allow the autonomous vehicle system to continue executing its primary driving decisions, such as slowing down and yielding to pedestrians, based on real obstructions, even if the system loses visual information about the target due to occlusion. This avoids the safety issue of misjudging the target's departure and updating driving behavior based on its presence, thereby improving the consistency of autonomous vehicle target tracking and the safety of driving decisions.

[0084] It should be noted that in a possible embodiment of the present invention, when multiple tracking targets are lost at the same time and multiple tracking targets enter the occlusion range of the same potential occluder, for each tracking target, the potential occluder is repeatedly determined to be a real occluder of the tracking target, that is, multiple real occluders corresponding to different tracking targets overlap at the position of the same potential occluder. When the occlusion target corresponding to a real occluder reappears in the vehicle's field of view, that is, when an occlusion target appearance signal is generated, the occlusion target and the real occluder corresponding to the occlusion target are cleared, that is, one real occluder mark is reduced at the position of the same potential occluder. When all occlusion targets that have entered the occlusion range have reappeared and been cleared, that is, all real occluder marks at the potential occluder position are also completely cleared, at which point the vehicle system can assume that there is no occlusion at that position. When a new tracking target enters the occlusion range, a real occluder mark is added at the corresponding potential occluder position.

[0085] See also Figure 2 and Figure 3 In another possible embodiment of the present invention, after all tracking targets are screened, all tracking targets and all corresponding potential occluders are packaged together into a fixed structure data in the system, known as a "potential occlusion scene combination" structure, which contains information such as the location and identification data of the tracking targets and potential occluders. In response to scenarios such as target loss and occlusion target appearance for the tracking targets in the potential occlusion scene combination structure, the labels of each object in the structure are updated in real time, for example, marking the tracking target as an occluder and the potential occluder as a real occluder.

[0086] Furthermore, when the exit of the real obstruction detects targets that are the same number as the corresponding obstruction targets, the label of the real obstruction is released; when the exit of the real obstruction detects targets that are less than the corresponding number of obstruction targets, the label of the real obstruction is maintained and the number of remaining obstruction targets remaining in the obstruction range is updated.

[0087] It should be understood that the solution proposed in the present invention solves the problem of occlusion perception loss in specific scenarios. For example, in a scenario where vehicles give way to pedestrians, after the pedestrian is obstructed by the vehicle in front, the autonomous driving main vehicle loses the ability to track the pedestrian. Subsequently, because the vehicle in front stops to give way, the vehicle in front is treated as a general static obstacle by the autonomous driving main vehicle and bypassed, resulting in a collision with the obstructed pedestrian from behind. The vehicle occlusion target tracking method proposed in the present invention solves the problem of perception loss when the pedestrian is obstructed. When the pedestrian is obstructed, it can continue to execute the first driving decision when the pedestrian is not obstructed, such as giving way to pedestrians, slowing down, etc. based on the real obstruction, thereby enhancing the consistency of vehicle driving decisions when the tracking target is lost. It can be considered that the present invention simulates the "short-term memory" ability of human drivers by storing obstructed targets and marking real obstructions, thereby improving the rationality of autonomous driving decisions and avoiding the risk of "ghosting".

[0088] Furthermore, the obtaining of several tracking targets includes:

[0089] Get the road centerline;

[0090] Acquire a plurality of moving targets, and acquire the moving direction and moving angle of each of the moving targets;

[0091] For any of the moving targets, if the moving direction satisfies a preset direction rule with the road centerline, and the moving angle satisfies a preset angle rule with the road centerline, the moving target is marked as a tracking target.

[0092] Specifically, in one possible embodiment, before acquiring the moving target, the autonomous driving vehicle is first used as the center line and the road conditions within a certain range of the autonomous driving vehicle are analyzed. Figure 4 The SL coordinate system shown in the figure is used as a reference. Targets within the range of ±4.5m in the horizontal direction and 30m in the vertical direction of the autonomous driving vehicle in the SL coordinate system are considered potential moving targets. Targets outside this range are considered too far away and can be ignored.

[0093] Specifically, in a possible embodiment, if the absolute value of the angle between the moving direction of the pedestrian and the center line of the road is about 90 degrees and the error is within 30 degrees, the pedestrian is considered to be a tracking target.

[0094] In this implementation, the road centerline is used as a reference benchmark. The road centerline is a benchmark line extending forward from the starting point along the center line of the road. Compared with the horizontal coordinate of the traditional Cartesian coordinate system, analysis based on the road centerline can more intuitively judge the relative relationship between the object movement and the center of the road.

[0095] It should be noted that, see Figure 4 , Figure 4It is a schematic diagram of the SL coordinate system. Specifically, the SL coordinate system is different from the traditional Cartesian coordinate system with the x-axis as the horizontal coordinate and the y-axis as the vertical coordinate. As shown in the figure, the SL coordinate system uses the road centerline as the reference line, S represents the direction along the road centerline, and L represents the direction perpendicular to the road centerline. Figure 4 In the figure, the centerline of the road is displayed as a dotted line in the middle of the SL coordinate system. The SL coordinate system can more conveniently represent the relative positional relationship between objects in the road and the road. This implementation method determines the movement trend of the moving target through the road baseline. When the moving direction of the moving target moves from the outer edge of the road toward the centerline of the road, that is, the moving direction of the moving target points from the outer side of the road to the centerline of the road; and the angle between the moving angle of the moving target and the centerline of the road is close to 90°, it is considered that the moving target is at risk of being obstructed by potential obstructions in the road. At this time, the moving target is marked as a tracking target for continuous tracking. This implementation method filters out key targets that need to be tracked from multiple targets in complex road conditions, reduces the complexity of identifying and tracking obstructed targets, and improves the efficiency of tracking obstructed targets.

[0096] Furthermore, the acquiring of potential occluders based on the plurality of tracking targets, and then acquiring an occlusion range corresponding to the potential occluders, includes:

[0097] Acquire a first distance interval based on all the tracked targets;

[0098] Acquire a plurality of obstacles and acquire a second distance of each obstacle;

[0099] For any obstacle, if the second distance satisfies the first distance interval, the obstacle is marked as the potential occluder.

[0100] Specifically, in one possible embodiment, after the tracking target is screened out, the SL coordinate system is used to detect whether there is a low-speed dynamic vehicle or other static obstacle between the autonomous driving main vehicle and the tracking target. The obstacle screening range is: centered on the autonomous driving main vehicle, 4.5m horizontally in the SL coordinate system, and within the first distance interval vertically.

[0101] In this implementation, the distance between each tracking target and the autonomous vehicle is obtained, thereby determining the distance interval between the tracking target and the autonomous vehicle. Obstacles located within this distance interval are located between the autonomous vehicle and the tracking target, and therefore pose a potential risk of obstructing the tracking target, causing the autonomous vehicle's visual information to lose sight of the tracking target. This implementation marks obstacles between the autonomous vehicle and the tracking target as potential obstructions, thereby filtering out key targets requiring attention from multiple obstacles in complex road conditions. This reduces the complexity of identifying and tracking obstructed scenes and improves the efficiency of tracking obstructed targets.

[0102] refer to Figure 2 and Figure 3 Further, in response to the tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, and then storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder, includes:

[0103] Setting a detection buffer zone outside the shielding range;

[0104] In response to the tracking target losing the signal, acquiring the historical coordinates of the tracking target;

[0105] If the historical coordinates are located in the detection buffer, it is determined that the tracked target enters the occlusion range.

[0106] In this implementation, in order to improve the robustness of the vehicle-occluded target tracking method and reduce the interference of the autonomous vehicle's signal perception and signal acquisition errors on the tracking of the occluded target, an additional buffer zone is set outside the occlusion range. This buffer zone is used to further determine whether the tracked target lost from the vehicle's field of view has indeed entered the corresponding occlusion range. Specifically, if the tracked target is located in the detection buffer zone immediately before it is lost, since the moving direction and angle of the tracked target have been defined in the above steps, it can be confirmed that the tracked target has indeed entered the corresponding occlusion range. On the contrary, if the tracked target is not located in the detection buffer zone immediately before it is lost, the probability that the target signal is lost is due to the signal perception and signal acquisition errors of the driving vehicle, and further judgment needs to be made after the perception signal acquisition is updated.

[0107] Furthermore, before responding to the tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, and then storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder, the method further includes:

[0108] Obtaining several scene frames within a preset time period;

[0109] For any of the scene frames, if the tracking target does not exist in the scene frame, the scene frame is marked as a tracking target lost frame;

[0110] If the number of the tracked target lost frames meets a preset lost frame threshold, the tracked target lost signal is generated.

[0111] In this implementation, the judgment conditions of multi-frame perception are further set to reduce the error rate of the present invention. Specifically, multi-frame perception refers to the analysis and judgment of multiple scene data frames acquired over a period of time. Since the output of such a module such as perception or prediction cannot be completely accurate, when a certain frame is inaccurate due to sensor noise or other reasons, if only relying on the perception and prediction results of a single frame to make a decision, it will cause the decision to jump and oscillate. Therefore, compared with relying solely on the loss of the tracking target in one scene frame for judgment, this implementation method judges whether the tracking target is indeed lost or it is a vehicle signal acquisition error by analyzing the proportion of target loss frames in all scene frames acquired over a period of time. When the proportion of target loss frames in all scene frames acquired over a period of time is high, it is considered that the tracking target has indeed been lost, and the tracking target loss signal is generated.

[0112] Specifically, in one embodiment, the multi-frame perception results of the autonomous driving master vehicle refer to the perception results of all scene frames within a period of time. For example, the perception output can be recorded as the first scene frame at t=0.1s, the second scene frame at t=0.2s, and so on. All the perception results of the scene frames within a certain period of time are then sequentially recorded in a storage container within an autonomous driving master vehicle data system. The purpose of recording the perception results of all scene frames within a period of time is to subsequently analyze historical data for the entire period by reading the storage container, thus providing a holistic view of the scene. Otherwise, since the output of a perception or prediction module cannot be completely accurate, if a frame is inaccurate due to sensor noise or other reasons, relying solely on the perception and prediction results of a single frame to make decisions will lead to jumps and fluctuations in the decision-making process. Therefore, this method of recording multiple frames can filter out some inaccurate data to a certain extent, ensuring that the analysis of the present invention provides stable results based on multiple frames of data.

[0113] Furthermore, after responding to the tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, and then storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder, the method further includes:

[0114] Obtaining the confidence level and real-time marking duration of the real obstruction;

[0115] If the real-time marking duration exceeds a first preset time, lowering the confidence level and executing a preset second driving decision based on the confidence level;

[0116] If the real-time marking duration exceeds a second preset time, the blocking target and the real blocking object are cleared.

[0117] In this implementation, in order to reduce the possible false detections and missed detections in the vehicle signal perception results themselves, this implementation performs a confidence level analysis on the real obstructions detected in the above steps. If the real obstruction has not been cleared for a long time, that is, the obstruction target corresponding to the real obstruction has not reappeared in the vehicle's field of view for a long time. At this time, it is considered that the vehicle system's early perception may have false detections and missed detections, marking the wrong real obstruction, or the obstruction has not been cleared for a long time due to a vehicle system error. At this time, the confidence level of the real obstruction is lowered, and a second driving decision such as slowing down, avoiding pedestrians, etc. is executed in addition to the first driving decision, and the real obstruction is considered to be a false detection or missed detection when the real-time marking time exceeds the second preset time, and it is cleared. Based on this implementation, the present invention can reduce the probability of false detection and missed detection in obstruction target tracking, thereby improving the accuracy of obstruction target tracking.

[0118] refer to Figure 2 and Figure 3 Furthermore, in response to the occlusion target occurrence signal of the occlusion target, before clearing the occlusion target and the real occlusion object, the method further includes:

[0119] Obtaining several scene frames within a preset time period;

[0120] For any of the scene frames, if the occluding target exists in the scene frame, the scene frame is marked as an occluding target existing frame;

[0121] If the number of frames in which the occluding target exists meets a preset frame threshold, the occluding target occurrence signal is generated.

[0122] In this implementation, the judgment conditions of multi-frame perception are further set to reduce the error rate of the present invention. Specifically, multi-frame perception refers to the analysis and judgment of multiple scene data frames acquired within a period of time. Since the output of such a module of perception or prediction cannot be completely accurate, when a certain frame is inaccurate due to sensor noise or other reasons, if only relying on the perception and prediction results of a single frame to make a decision, it will cause the decision to jump and oscillate. Therefore, compared with relying solely on the appearance of the occluded target in one scene frame for judgment, this implementation method judges whether the occluded target actually appears or it is a vehicle signal acquisition error by analyzing the proportion of frames with occluded targets in all scene frames acquired within a period of time. When the proportion of frames with occluded targets in all scene frames acquired within a period of time is high, it is considered that the occluded target has indeed appeared in the field of view of the vehicle, and the occluded target appearance signal is generated.

[0123] See also Figure 5In a second aspect, the present invention provides an occluded target tracking system for an autonomous driving vehicle, comprising a tracking target recognition module 100, a potential occluder recognition module 200, an occluded target tracking module 300, and a driving module 400; wherein:

[0124] The tracking target identification module 100 is used to obtain a number of tracking targets;

[0125] The potential obstruction identification module 200 is used to obtain potential obstructions based on a plurality of tracking targets, and then obtain an obstruction range corresponding to the potential obstructions;

[0126] The occluded target tracking module 300 is configured to execute the following tracking strategy for any of the tracked targets:

[0127] In response to a tracking target loss signal of the tracking target, confirming that the tracking target enters the occlusion range, storing the tracking target as an occlusion target, and marking the potential occluder as a real occluder; in response to an occlusion target presence signal of the occlusion target, clearing the occlusion target and the real occluder;

[0128] The driving module 400 is configured to execute a preset first driving decision according to the real obstruction.

[0129] In this implementation, targets such as pedestrians on the road that could cause safety accidents are defined as tracking targets. The target's position is continuously tracked, and potential obstructions on the road that could obstruct the target are detected. When the target is lost, the system detects and confirms that the target is obscured by a potential obstruction and stores it in memory as an obscured target and a real obstruction. Through this obscured target memory storage strategy and real obstruction marking strategy, even if the autonomous driving system loses visual information about the tracking target due to occlusion, it can still execute the primary driving decision, such as slowing down and avoiding pedestrians, based on the real obstruction. This avoids the safety issue of misjudging the tracking target's departure and updating the driving behavior without noticing its presence, thereby improving the consistency of autonomous driving target tracking and the safety of driving decisions.

[0130] Furthermore, the obtaining of several tracking targets includes:

[0131] Get the road centerline;

[0132] Acquire a plurality of moving targets, and acquire the moving direction and moving angle of each of the moving targets;

[0133] For any of the moving targets, if the moving direction satisfies a preset direction rule with the road centerline, and the moving angle satisfies a preset angle rule with the road centerline, the moving target is marked as a tracking target.

[0134] In this implementation, the road centerline is used as a reference benchmark. The road centerline is a benchmark that extends forward from the starting point along the center line of the road. Compared with the horizontal coordinate of the traditional Cartesian coordinate system, analysis based on the road centerline can more intuitively judge the relative relationship between the object movement and the center of the road. Specifically, this implementation determines the movement trend of the moving target through the road baseline. When the moving direction of the moving target moves from the outer edge of the road toward the center line of the road, that is, the moving direction of the moving target points from the outer side of the road to the center line of the road; and the angle between the moving angle of the moving target and the road centerline is close to 90°, it is considered that the moving target is at risk of being obstructed by potential obstructions in the road. At this time, the moving target is marked as a tracking target for continuous tracking. Through this implementation, key targets that need to be tracked are screened out from multiple targets in complex road conditions, which reduces the complexity of identifying and tracking obstructed targets and improves the efficiency of tracking obstructed targets.

[0135] Furthermore, the acquiring of potential occluders based on the plurality of tracking targets, and then acquiring an occlusion range corresponding to the potential occluders, includes:

[0136] Acquire a first distance interval based on all the tracked targets;

[0137] Acquire a plurality of obstacles and acquire a second distance of each obstacle;

[0138] For any obstacle, if the second distance satisfies the first distance interval, the obstacle is marked as the potential occluder.

[0139] In this implementation, the distance between each tracking target and the autonomous vehicle is obtained, thereby determining the distance interval between the tracking target and the autonomous vehicle. Obstacles located within this distance interval are located between the autonomous vehicle and the tracking target, and therefore pose a potential risk of obstructing the tracking target, causing the autonomous vehicle's visual information to lose sight of the tracking target. This implementation marks obstacles between the autonomous vehicle and the tracking target as potential obstructions, thereby filtering out key targets requiring attention from multiple obstacles in complex road conditions. This reduces the complexity of identifying and tracking obstructed scenes and improves the efficiency of tracking obstructed targets.

[0140] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0141] The "embodiment" mentioned in this document means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0142] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make several improvements and substitutions without departing from the scope of the present application, and such improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for tracking an obscured target by an autonomous driving vehicle, characterized in that: include: Get the road centerline; Acquire a plurality of moving targets, and acquire the moving direction and moving angle of each of the moving targets; For any moving target, if the moving direction satisfies a preset direction rule with the road centerline, and the moving angle satisfies a preset angle rule with the road centerline, the moving target is marked as a tracking target; Acquire a first distance interval based on all the tracked targets; Acquire a plurality of obstacles and acquire a second distance of each obstacle; For any obstacle, if the second distance satisfies the first distance interval, the obstacle is marked as a potential occluder; Obtaining an occlusion range corresponding to the potential occluder; For any of the tracking targets, the following tracking strategies are executed: Setting a detection buffer zone outside the shielding range; In response to a tracking target loss signal of the tracking target, acquiring historical coordinates of the tracking target; If the historical coordinates are located in the detection buffer, it is determined that the tracked target has entered the occlusion range, and the tracked target is stored as an occlusion target, and the potential occluder is marked as a real occluder; executing a preset first driving decision based on the real obstruction; In response to an occlusion target presence signal of the occlusion target, the occlusion target and the real occluder are cleared.

2. The method for tracking an obstructed target by an autonomous driving vehicle according to claim 1, wherein: Before obtaining the historical coordinates of the tracked target in response to the tracked target loss signal of the tracked target, the method further includes: Acquire several scene frames within a preset time period; For any of the scene frames, if the tracking target does not exist in the scene frame, the scene frame is marked as a tracking target lost frame; If the number of the tracked target lost frames meets a preset lost frame threshold, the tracked target lost signal is generated.

3. The method for tracking an obstructed target by an autonomous driving vehicle according to claim 1, wherein: If the historical coordinates are located in the detection buffer, then it is determined that the tracked target has entered the occlusion range, the tracked target is stored as an occlusion target, and the potential occluder is marked as a real occluder, the method further includes: Obtaining the confidence level and real-time marking duration of the real obstruction; If the real-time marking duration exceeds a first preset time, lowering the confidence level and executing a preset second driving decision based on the confidence level; If the real-time marking duration exceeds a second preset time, the blocking target and the real blocking object are cleared.

4. The method for tracking an obstructed target by an autonomous driving vehicle according to claim 1, wherein: Before clearing the occlusion target and the real obstruction in response to the occlusion target occurrence signal of the occlusion target, the method further includes: Acquire several scene frames within a preset time period; For any of the scene frames, if the occluding target exists in the scene frame, the scene frame is marked as an occluding target existing frame; If the number of frames in which the occluding target exists meets a preset frame threshold, the occluding target occurrence signal is generated.

5. An autonomous driving vehicle occluded target tracking system, characterized in that: It includes a tracking target recognition module, a potential obstruction recognition module, an obstruction target tracking module and a driving module; wherein: The tracking target identification module is used to perform the following operations: Get the road centerline; Acquire a plurality of moving targets, and acquire the moving direction and moving angle of each of the moving targets; For any moving target, if the moving direction satisfies a preset direction rule with the road centerline, and the moving angle satisfies a preset angle rule with the road centerline, the moving target is marked as a tracking target; The potential obstruction identification module is configured to perform the following operations: Acquire a first distance interval based on all the tracked targets; Acquire a plurality of obstacles and acquire a second distance of each obstacle; For any obstacle, if the second distance satisfies the first distance interval, the obstacle is marked as a potential occluder; Obtaining an occlusion range corresponding to the potential occluder; The occluded target tracking module is used to execute the following tracking strategy for any of the tracking targets: Setting a detection buffer zone outside the shielding range; In response to a tracking target loss signal of the tracking target, acquiring historical coordinates of the tracking target; If the historical coordinates are located in the detection buffer, it is determined that the tracked target has entered the occlusion range, the tracked target is stored as an occlusion target, and the potential occluder is marked as a real occluder; in response to an occlusion target occurrence signal of the occlusion target, the occlusion target and the real occluder are cleared; The driving module is used to execute a preset first driving decision according to the real obstruction.

Images