Obstacle filtering method, device, equipment and medium

Abstract

The invention discloses an obstacle filtering method and device, equipment and a medium, and the method comprises the steps: obtaining the obstacle information of an initial obstacle which is an obstacle in an obstacle set outputted by a visual perception model; determining an initial obstacle meeting a preset filtering condition based on the obstacle information, and taking the initial obstacle as a to-be-filtered obstacle; wherein meeting the preset filtering condition represents that the accuracy of the obstacle output by the visual perception model is insufficient; and filtering the to-be-filtered obstacles from the obstacle set to obtain a residual obstacle set, so that the automatic emergency braking system carries out processing based on the residual obstacle set. In this way, after the visual perception model is output, filtering is carried out, the probability that wrong obstacles are provided for the automatic emergency braking system can be reduced, and safety is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent driving technology, and in particular to an obstacle filtering method, device, equipment and medium. Background Art

[0002] The AEB (Autonomous Emergency Braking) function in ADAS (Advanced Driver Assistance Systems) technology can effectively protect the lives of traffic participants. However, due to the complexity of actual traffic scenarios, the visual perception model cannot fully and accurately identify the attribute information of all traffic participants, and may even cause frequent and large-scale attribute misidentification. Furthermore, the AEB function may make incorrect decisions, causing the vehicle to brake or traffic accidents.

[0003] Therefore, how to reduce the probability of providing incorrect obstacles to the automatic emergency braking system and improve safety is a technical problem that needs to be solved at present. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an obstacle filtering method, device, equipment, and medium. This can reduce the probability of incorrectly providing obstacles to the automatic emergency braking system and improve safety. The specific solution is as follows:

[0005] In a first aspect, the present application provides an obstacle filtering method, comprising:

[0006] Obtaining obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in an obstacle set output by a visual perception model;

[0007] Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered; wherein meeting the preset filtering conditions indicates that the accuracy of the obstacles output by the visual perception model is insufficient;

[0008] The obstacles to be filtered are filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system performs processing based on the remaining obstacle set.

[0009] Optionally, determining initial obstacles that meet preset filtering conditions based on the obstacle information as obstacles to be filtered includes:

[0010] Fitting the lane line transverse coordinate corresponding to the longitudinal distance based on the initial obstacle longitudinal distance and lane line information;

[0011] If the difference between the lateral coordinate of the lane line and the lateral distance of the initial obstacle is greater than a preset distance threshold, it is determined that the initial obstacle meets the preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

[0012] Optionally, determining initial obstacles that meet preset filtering conditions based on the obstacle information as obstacles to be filtered includes:

[0013] Filtering the horizontal distance and the vertical distance of the initial obstacle based on the distance filter coefficient to obtain the current filtered horizontal distance and the current filtered vertical distance of the initial obstacle;

[0014] Obtaining a target historical filtered lateral velocity and a target historical filtered longitudinal velocity of the initial obstacle; wherein the target historical filtered lateral velocity and the target historical filtered longitudinal velocity are historical filtered lateral velocities and historical filtered longitudinal velocities that are at a preset time interval from the current time interval;

[0015] Calculating a predicted lateral distance based on the target's historical filtered lateral velocity and the preset time duration;

[0016] Calculating a predicted longitudinal distance based on the target's historical filtered longitudinal speed and the preset duration;

[0017] If the difference between the lateral predicted distance and the current filtered lateral distance or the difference between the longitudinal predicted distance and the current filtered longitudinal distance meets a preset difference condition, it is determined that the initial obstacle meets the preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

[0018] Optionally, determining initial obstacles that meet preset filtering conditions based on the obstacle information as obstacles to be filtered includes:

[0019] determining a collision time of the initial obstacle;

[0020] When the collision time is less than a preset time threshold, calculating a current expected deceleration of the initial obstacle;

[0021] Based on the historical expected deceleration of the initial obstacle and the current expected deceleration, it is determined whether the initial obstacle meets a preset deceleration fluctuation condition. If the initial obstacle meets the preset deceleration fluctuation condition, it is determined that the initial obstacle meets a preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

[0022] Optionally, also include:

[0023] Sort the obstacles in the remaining obstacle set based on the expected deceleration and the collision time to obtain a sorting result;

[0024] The sorting result is output to an automatic emergency braking system so that the automatic emergency braking system performs processing based on the sorting result.

[0025] Optionally, determining initial obstacles that meet preset filtering conditions based on the obstacle information as obstacles to be filtered includes:

[0026] Comparing the existence probability and category accuracy probability of the initial obstacle with the existence probability threshold and the accuracy probability threshold respectively;

[0027] When the existence probability is less than the existence probability threshold or the category accuracy probability is less than the accuracy probability threshold, it is determined that the initial obstacle meets the preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

[0028] Optionally, determining initial obstacles that meet preset filtering conditions based on the obstacle information as obstacles to be filtered includes:

[0029] Comparing the horizontal and vertical distance variances and the horizontal and vertical speed variances of the initial obstacle with the distance variance threshold and the obstacle speed variance threshold respectively;

[0030] When the lateral and longitudinal distance variance is greater than the distance variance threshold or the lateral and longitudinal speed variance is greater than the obstacle speed variance threshold, it is determined that the initial obstacle meets the preset filtering condition and the initial obstacle is determined as an obstacle to be filtered.

[0031] In a second aspect, the present application provides an obstacle filtering device, comprising:

[0032] An obstacle information acquisition module, configured to acquire obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in an obstacle set output by a visual perception model;

[0033] a filtered obstacle determination module, configured to determine, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered; wherein meeting the preset filtering conditions indicates that the accuracy of the obstacles output by the visual perception model is insufficient;

[0034] The obstacle filtering module is configured to filter the obstacles to be filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system can perform processing based on the remaining obstacle set.

[0035] In a third aspect, the present application provides an electronic device, including a memory and a processor, wherein:

[0036] The memory is used to store computer programs;

[0037] The processor is configured to execute the computer program to implement the aforementioned obstacle filtering method.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program implements the aforementioned obstacle filtering method when executed by a processor.

[0039] In a fifth aspect, the present application provides a computer program product, comprising a computer program / instruction, which implements the steps of the aforementioned obstacle filtering method when executed by a processor.

[0040] From the above scheme, it can be seen that the present application provides an obstacle filtering method, including: obtaining obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in an obstacle set output by a visual perception model; determining, based on the obstacle information, an initial obstacle that meets a preset filtering condition as an obstacle to be filtered; wherein meeting the preset filtering condition indicates that the obstacle output by the visual perception model is insufficiently accurate; filtering the obstacles to be filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system performs processing based on the remaining obstacle set.

[0041] It can be seen that the beneficial effects of the present application are: after the visual perception model outputs an obstacle set, the present application determines the initial obstacles that meet the preset filtering conditions, that is, the inaccurately identified obstacles, based on the information of the initial obstacles in the obstacle set, as the obstacles to be filtered. After filtering, the remaining obstacle set is provided to the automatic emergency braking system. In this way, after the visual perception model is output, filtering is performed, which can effectively filter out unreasonable obstacles detected by vision, reduce the probability of providing erroneous obstacles to the automatic emergency braking system, and improve safety.

[0042] Correspondingly, the obstacle filtering device, equipment and readable storage medium provided by the present application also have the above-mentioned technical effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0044] Figure 1 A flow chart of an obstacle filtering method provided in an embodiment of the present application;

[0045] Figure 2 This application provides an obstacle filtering solution and a schematic diagram of the upstream and downstream relationship provided in an embodiment of the present application;

[0046] Figure 3 An obstacle filtering flow chart provided in an embodiment of the present application;

[0047] Figure 4 A schematic diagram of driving within a lane provided in an embodiment of the present application;

[0048] Figure 5 A schematic diagram of the mapping relationship between vehicle speed and distance filter coefficient provided in an embodiment of the present application;

[0049] Figure 6 A schematic diagram of a mapping relationship between vehicle speed and speed filter coefficient provided in an embodiment of the present application;

[0050] Figure 7 A schematic structural diagram of an obstacle filtering device provided in an embodiment of the present application;

[0051] Figure 8 A structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0053] Currently, conventional ADAS functions are primarily implemented based on the vehicle's external forward-looking integrated device. This device utilizes a combination of a System on a Chip (SOC) and a Microcontroller Unit (MCU) to implement a range of functions. The SOC houses a visual detection model, while the MCU implements regulatory control algorithms, enabling traditional assisted driving features such as AEB, Adaptive Cruise Control (ACC), and Lane Keeping Assist (LKA). Conventional assisted driving features are actively enabled by the driver, who then controls the vehicle's trajectory and speed based on visual perception and planning logic. Since these features are enabled by the driver, who often maintains a certain level of awareness of the vehicle's posture, AEB, unlike other assistance features, runs constantly in the background of the vehicle controller. The AEB function performs a risk assessment on all obstacles in the visually detected traffic scene to determine whether there is a possibility of collision. However, collisions often occur very quickly in real-world scenarios, which means that the results of visual perception need to have high convergence precision and accuracy, and the time left for the AEB function to perform target selection and risk assessment is also very short. However, the actual traffic environment is too complex, and the results of visual model detection are difficult to achieve accurately and quickly. In addition, the model's detection results are often incorrect, and even obstacles that do not exist in the real world are simulated. This can cause AEB to make incorrect planning results, causing the vehicle to brake, which may lead to rear-end collisions or unnecessary collisions, and a poor driving experience for the driver, posing a huge challenge to the AEB function.

[0054] The AEB function in ADAS technology can effectively protect the lives of traffic participants. However, due to the complexity of actual traffic scenarios, visual perception models cannot fully and accurately identify the attributes of all traffic participants, and may even cause frequent and large-scale attribute misidentification. For example, low visibility weather, the presence of light sources reflected by rain on the road, vehicles of different speeds, the reflection of tall buildings on flat roads, steep roads, posters in underground parking lots, and dense groups of obstacles in downtown areas all cause varying degrees of misidentification. Furthermore, this can cause the AEB function to make incorrect decisions, causing the vehicle to stop or traffic accidents. Although the detection accuracy of the visual model can be improved through repeated training with large amounts of data, solving a difficult engineering and technical problem requires the coordinated efforts of multiple solutions.

[0055] To this end, the present application provides an obstacle filtering solution, which can perform secondary corrections for the visual model's misdetection through a reasonable and effective obstacle filtering strategy, filter out the erroneous obstacles detected by the visual system, further increase the robustness of the visual model, and effectively improve the performance stability of the AEB function and good user feedback, thereby reducing the complaint rate of various end users.

[0056] See also Figure 1 As shown, the embodiment of the present application discloses an obstacle filtering method, comprising:

[0057] Step S11: Obtain obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in an obstacle set output by a visual perception model.

[0058] Among them, obstacle information can include motion state information and accuracy information. The motion state information can include longitudinal distance, longitudinal distance, lateral speed, longitudinal speed, lateral-longitudinal distance variance, lateral-longitudinal speed variance, etc. The lateral-longitudinal distance variance includes lateral distance variance and longitudinal distance variance. The lateral-longitudinal speed variance includes lateral speed variance and longitudinal speed variance. Accuracy information can include existence probability and category accuracy probability. In the embodiment of the present application, the origin of the coordinate system is the center of the rear axle of the vehicle, the direction of the vehicle's movement is the X-axis, i.e., the longitudinal direction, and the Y-axis, i.e., the lateral direction, perpendicular to the direction of the vehicle's movement and to the left is.

[0059] Step S12: determining initial obstacles that meet preset filtering conditions based on the obstacle information as obstacles to be filtered; wherein, meeting the preset filtering conditions indicates that the accuracy of the obstacles output by the visual perception model is insufficient.

[0060] In an optional implementation, it may be determined whether the motion state of the initial obstacle complies with motion laws based on the motion state information of the initial obstacle. If it does not comply with the motion laws, it is determined that a preset filtering condition is satisfied.

[0061] In an optional embodiment, the lateral coordinates of the lane line corresponding to the longitudinal distance of the initial obstacle and the lane line information can be fitted. If the difference between the lateral coordinates of the lane line and the lateral distance of the initial obstacle is greater than a preset distance threshold, it is determined that the initial obstacle meets the preset filtering condition and is determined to be an obstacle to be filtered.

[0062] If the difference between the lane line's lateral coordinate and the initial obstacle's lateral distance is greater than a preset distance threshold, the initial obstacle is not moving along the lane line trajectory, indicating that the initial obstacle's motion state does not conform to the law of motion. Lane line information is the coefficient of the lane line equation. The independent variable in the lane line equation is the longitudinal distance, and the function value is the lateral record. Based on the lane line equation, the lane line equation's coefficient, and the initial obstacle's longitudinal distance, the corresponding lateral distance, i.e., the lane line's lateral coordinate, can be fitted.

[0063] In an optional embodiment, the lateral distance and longitudinal distance of the initial obstacle may be filtered based on a distance filter coefficient to obtain a current filtered lateral distance and a current filtered longitudinal distance of the initial obstacle; a target historical filtered lateral velocity and a target historical filtered longitudinal velocity of the initial obstacle are obtained; wherein the target historical filtered lateral velocity and the target historical filtered longitudinal velocity are historical filtered lateral velocities and historical filtered longitudinal velocities with a preset time interval from the current time; a predicted lateral distance is calculated based on the target historical filtered lateral velocity and the preset time interval; and a predicted longitudinal distance is calculated based on the target historical filtered longitudinal velocity and the preset time interval; and if a difference between the predicted lateral distance and the current filtered lateral distance or between the predicted longitudinal distance and the current filtered longitudinal distance satisfies a preset difference condition, it is determined that the initial obstacle meets a preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

[0064] Among them, the distance filter coefficient can be obtained based on the speed of the vehicle, and in the embodiment of the present application, it can be obtained based on a preset mapping relationship. The preset time length is, for example, 1.5 seconds. In the embodiment of the present application, the current filtered lateral distance and the current filtered longitudinal distance are obtained, and records can be made, retaining the records for at least the preset time length. The filter coefficient corresponding to the lateral speed is obtained based on the lateral speed, and the filter coefficient corresponding to the longitudinal speed is obtained based on the longitudinal speed. In addition, a preset mapping relationship corresponding to the lateral distance filter and a preset mapping relationship corresponding to the longitudinal distance filter can be provided respectively. The preset difference condition can be that the difference is greater than the difference threshold, and the difference threshold can be calculated based on the filtered distance, for example, it can be 60% of the filtered distance. If the preset difference condition is met, it indicates that the motion state of the obstacle does not conform to the law of motion.

[0065] In addition, the embodiment of the present application can also obtain a speed filter coefficient; based on the speed filter coefficient, the lateral speed and longitudinal speed of the initial obstacle are filtered to obtain the filtered lateral speed and filtered longitudinal speed of the initial obstacle. The speed filter coefficient can be obtained based on the mapping relationship between the obstacle distance and the speed filter coefficient. The speed filter coefficient corresponding to the longitudinal distance is obtained based on the longitudinal distance, and a preset mapping relationship corresponding to the lateral speed filter and a preset mapping relationship corresponding to the longitudinal speed filter can be respectively provided. In the embodiment of the present application, the filtered lateral speed and filtered longitudinal speed can be obtained and recorded, and the records can be retained for at least a preset time period.

[0066] In an optional embodiment, a collision time of the initial obstacle may be determined; if the collision time is less than a preset time threshold, a current expected deceleration of the initial obstacle may be calculated; and based on the historical expected deceleration of the initial obstacle and the current expected deceleration, a determination is made as to whether the initial obstacle meets a preset deceleration fluctuation condition. If the initial obstacle meets the preset deceleration fluctuation condition, the initial obstacle is determined to meet a preset filtering condition and is identified as an obstacle to be filtered.

[0067] If the collision time is relatively short, the risk is relatively high, and further judgment is performed. The preset deceleration fluctuation condition can be that the difference between the current expected deceleration and any two adjacent expected decelerations in the historical expected deceleration record is greater than a preset fluctuation threshold. The historical expected deceleration record is the expected deceleration record of the preset duration closest to the current time. After the embodiment of the present application obtains the current expected deceleration and completes the judgment of the preset deceleration fluctuation condition, the current expected deceleration can be added to the historical expected deceleration record. The historical expected deceleration record retains a preset number of expected decelerations. If the expected deceleration fluctuates significantly, it indicates that it does not conform to the law of motion.

[0068] In an optional embodiment, the lateral and longitudinal distance variances and the lateral and longitudinal speed variances of the initial obstacle may be compared with a distance variance threshold and an obstacle speed variance threshold, respectively. If the lateral and longitudinal distance variances are greater than the distance variance thresholds or the lateral and longitudinal speed variances are greater than the obstacle speed variance thresholds, it is determined that the initial obstacle meets the preset filtering condition and the initial obstacle is determined as an obstacle to be filtered.

[0069] The lateral and longitudinal distance variances include both the lateral and longitudinal distance variances. If either the lateral or longitudinal distance variance is greater than the distance variance threshold, then both the lateral and longitudinal distance variances are greater than the distance variance threshold. The lateral and longitudinal speed variances include both the lateral and longitudinal speed variances. If either the lateral or longitudinal speed variance is greater than the speed variance threshold, then both the lateral and longitudinal speed variances are greater than the speed variance threshold. Variance is an important factor in determining whether an obstacle is experiencing a sudden change; severe sudden changes do not conform to the laws of motion.

[0070] In an optional embodiment, the existence probability and category accuracy probability of the initial obstacle can be compared with an existence probability threshold and an accuracy probability threshold, respectively. If the existence probability is less than the existence probability threshold or the category accuracy probability is less than the accuracy probability threshold, the initial obstacle is determined to meet the preset filtering condition and is determined to be an obstacle to be filtered. In other words, less accurate obstacles output by the visual perception model are directly filtered based on the probability threshold and the accuracy probability threshold.

[0071] That is, in the present application, the preset filtering condition may include multiple sub-conditions, and the multiple sub-conditions may be used individually or in combination.

[0072] Step S13: filtering the obstacles to be filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system performs processing based on the remaining obstacle set.

[0073] In an embodiment of the present application, obstacles in the remaining obstacle set can be sorted based on the expected deceleration and collision time to obtain a sorting result; the sorting result can be output to the automatic emergency braking system so that the automatic emergency braking system can process based on the sorting result.

[0074] The sorting principle is based on expected deceleration. The lower the expected deceleration, the closer it is to the front. Then, the collision time is considered: the shorter the collision time, the closer it is to the front. Expected deceleration is a negative value. This prioritizes the riskiest vehicles, ensuring safety.

[0075] It can be seen that in the embodiment of the present application, after the visual perception model outputs an obstacle set, the initial obstacles that meet the preset filtering conditions, i.e., inaccurately identified obstacles, are determined based on the information of the initial obstacles in the obstacle set, as the obstacles to be filtered. After filtering, the remaining obstacle set is provided to the automatic emergency braking system. In this way, after the visual perception model is output, filtering is performed, which can effectively filter out unreasonable obstacles detected by vision, reduce the probability of providing incorrect obstacles to the automatic emergency braking system, and improve safety.

[0076] For further information, see Figure 2 As shown, Figure 2The present application provides an obstacle filtering solution and a schematic diagram of the upstream and downstream relationships. The visual model is arranged in the SOC, and the image recognized by the camera is input into the visual model frame by frame. The relevant information is then detected, and all obstacle information is sent to the MCU. Considering the transmission delay from the SOC to the MCU, the filtering solution proposed in this application is arranged in the MCU, at the front end of the AEB planning model. This allows unreliable obstacles to be filtered out first, reducing the computing power consumption of the AEB planning module. This also complies with the overall code design concept (continuously filtering useless obstacles).

[0077] For further information, see Figure 3 As shown, Figure 3 This is a flowchart of an obstacle filtering process provided in an embodiment of the present application. Specifically, it may include the following steps:

[0078] Step 1: Filter the 32 obstacles received from the sensor based on the obstacle presence probability and obstacle category accuracy probability output by the visual model. This step primarily filters out highly inaccurate obstacles. The obstacle presence probability threshold and obstacle category accuracy threshold in this strategy are set to 0.5 and 0.6, respectively. The corresponding probability range of the sensor output is [0, 1]. To increase the range of obstacle filtering, the thresholds in this strategy can be modified accordingly, but the threshold setting should be determined based on the characteristics of visual perception. If the number of obstacles received is less than the set threshold, the obstacle is filtered out and the next obstacle is filtered. If the threshold is met, the next obstacle is filtered.

[0079] Step 2: Obstacles that have passed the previous step are filtered based on the variance of their horizontal and vertical distances and their horizontal and vertical speeds, as detected by the visual model. The variance of obstacle attributes is a key factor in determining whether these attributes have experienced sudden changes. Therefore, this step can simply filter out obstacles with sudden changes in distance or speed. The obstacle distance variance threshold and the obstacle speed variance threshold in this strategy are set to 15 and 10, respectively. The corresponding variance range of the perception output is [0,100]. Considering the possibility of sudden deceleration or acceleration of obstacles, the thresholds cannot be set too low. To increase the range of obstacle filtering, the thresholds in this strategy can be modified accordingly, but the threshold settings should be based on the characteristics of visual perception. If a received obstacle exceeds the set threshold, it is filtered out and the next obstacle is filtered. If the threshold is met, the next obstacle is filtered.

[0080] Step 3: Obstacles that have passed the previous step are filtered based on the relative relationship between the obstacles and the lane lines. Lane lines are an important factor in traffic scenes. Both motor vehicles and pedestrians are driving along the trajectories of lane lines. Therefore, the relevant information of the lane lines sensed and recognized is first fitted. This invention uses a cubic polynomial to fit the lane lines. The specific formula is as follows:

[0081] ;

[0082] Where X represents the coordinate of the X-axis in the current coordinate system. Represents the lateral distance between the vehicle and the lane line (i.e., the lateral distance), Represents the tangent direction of the lane line, which can be expressed by the vehicle's heading angle. represents the curvature of the road, represents the rate of change of the curvature of the road, and y represents the y coordinate fitted under the current X coordinate, see Figure 4 As shown, Figure 4 A schematic diagram of driving within a lane provided in an embodiment of the present application. The strategy in the third step is mainly to filter out some unreasonable obstacles traveling in the opposite direction. First, after obtaining the lane information, the X-direction distance of the obstacle can be obtained and substituted into the fitting formula to obtain the Y-direction distance. By comparing the Y-direction distance of the obstacle traveling in the opposite direction with the fitted Y value, if it is found that the difference between the Y value of the obstacle and the fitted road boundary value is greater than 1.2m, it is considered that the vehicle traveling in the opposite direction is not traveling according to the trajectory of the lane line, so it can be filtered. When the received obstacle does not meet the set threshold, the obstacle is filtered out and the next obstacle is filtered. If it meets the threshold, the next filter is performed.

[0083] Step 4: Visual model detection is prone to jitter, and the attributes of the obstacles it detects also experience some unreasonable fluctuations. Furthermore, the accuracy of the visual model is easily affected by the vehicle's speed and distance. Therefore, the attributes of the detected obstacles can be corrected and optimized. For obstacles that have passed the previous step, the strategy for correcting and optimizing the attributes of the obstacles based on the motion attributes sent by the visual system and the vehicle's motion state is as follows:

[0084] (1) Determine the filter weights of the horizontal and vertical distances of obstacles based on the vehicle speed, i.e., the distance filter coefficients: Considering that the principle of visual detection is to extract image information frame by frame, its detection accuracy will inevitably be affected by the vehicle speed. Therefore, the filter weights of the horizontal and vertical distances of different obstacles based on the vehicle speed are designed, see Figure 5 As shown, Figure 5A schematic diagram of the mapping relationship between vehicle speed and distance filter coefficients provided in an embodiment of the present application. The upper limit of the longitudinal distance filter for obstacles is 0.8, and the lower limit is 0.2. Similarly, the upper limit of the lateral filter for obstacles is 0.75, and the lower limit is 0.25.

[0085] (2) Based on the historical horizontal and vertical distances of the obstacle, calculate the horizontal and vertical distances at the current moment; the calculation formula for the horizontal and vertical distances is as follows:

[0086] ;

[0087] in, Indicates the obstacle distance after filtering, horizontal distance or longitudinal distance, weight represents the weight obtained based on the vehicle speed table, that is, the distance filter coefficient. Represents the distance at the previous moment, Represents the distance observed at the current moment. Calculate the horizontal distance after filtering, then 、 、 are all horizontal distances, and the vertical distance after filtering is calculated. 、 、 All are vertical distances.

[0088] (3) Determine the obstacle's horizontal and vertical speed filter coefficients based on the obstacle distance. Similar to the obstacle's horizontal and vertical distances, the obstacle's horizontal and vertical speeds are also filtered accordingly to improve the accuracy of the obstacle's attributes. The upper limit of the obstacle's longitudinal speed filter is 0.63, and the lower limit is 0.27. Similarly, the upper limit of the obstacle's lateral speed filter is 0.77, and the lower limit is 0.23. Figure 6 As shown, Figure 6 A schematic diagram of the mapping relationship between vehicle speed and speed filter coefficient provided in an embodiment of the present application.

[0089] (4) Based on the current horizontal and vertical distances and the historical horizontal and vertical speeds, calculate the horizontal and vertical speeds of the obstacle at the current moment. The speed calculation formula is as follows:

[0090] ;

[0091] ;

[0092] in, Represents the obstacle speed after filtering, weight represents the weight obtained based on the vehicle speed table, Represents the speed at the previous moment, Represents the speed calculated at the current moment, Represents the distance at the previous moment, Represents the timestamp of the last cycle when the obstacle was detected. Represents the timestamp of the current cycle. The operating cycle can be in milliseconds, so it must be divided by 1000 to convert to meters per second. For example, the operating cycle is 25ms.

[0093] Step 5: For obstacles that have passed the previous step, the rationality is determined based on the previously calculated obstacle attributes, such as longitudinal and lateral speeds, and distances. The specific strategy is to record the obstacle's longitudinal distance, longitudinal speed, lateral distance, and lateral speed using containers at 150ms intervals. The recording is repeated for 1.5s, with a maximum container size of 10. The data recorded in this container is equivalent to the obstacle's historical trajectory information, which can be used for filtering. If the historical distance and speed information are seriously inconsistent, the obstacle is considered unreasonable. First, based on the obstacle's attribute information 1.5 seconds prior, the obstacle's position and distance 1.5 seconds later can be estimated. The difference between the predicted position and distance (i.e., the product of the recorded velocity and time) and the observed position and distance (i.e., the recorded filtered distance) can be compared. If the error exceeds 60%, the obstacle's motion is considered inconsistent and the obstacle is considered unreasonable. Furthermore, obstacle prediction uses a uniform velocity model, which can introduce errors. Therefore, a 60% error threshold is allowed. This error threshold is also affected by the time lag between visual perception and the regulatory control link. If a received obstacle does not meet the set threshold, it is filtered out and the next obstacle is filtered. If it meets the threshold, the next obstacle is filtered.

[0094] Step 6: For obstacles that have passed the previous step, the horizontal and vertical collision times are calculated based on the speed of the vehicle and the relative motion relationship with the obstacle. The collision time is calculated by dividing the relative distance by the relative speed. A simple potential collision risk judgment can be obtained. Obstacles with a collision time greater than 5s are marked but not filtered out. The potential risk is relatively low and they will be placed at the end of the obstacle list. They will be sorted according to the size of the collision time. For obstacles with a collision time less than 5s, the next additional calculation will be performed. In addition, for cases where there is no solution to the collision time, the collision time is uniformly set to 20s to ensure that obstacles with real potential risks are not filtered out. The collision time calculation formula is as follows:

[0095] ;

[0096] ;

[0097] in, represents the longitudinal collision time, Represents the longitudinal obstacle distance, Represents the wheelbase of the vehicle, represents the speed of the vehicle, represents the speed of the obstacle, represents the lateral collision time, Represents the horizontal obstacle distance, Represents the width of the vehicle, Represents the length of the obstacle.

[0098] Step 7: The occurrence of a collision event will not only be affected by the speed and distance of the obstacle, but also if the acceleration of the obstacle changes drastically, a collision event will occur. Therefore, acceleration-related information can be further taken into account in the evaluation of the collision event. If the motion state of an obstacle is not self-consistent, it means that the expected deceleration to avoid collision calculated when a collision event occurs is difficult to be continuous, and there may be great jitter. Therefore, by calculating the expected deceleration of each obstacle with a collision risk, this value can reflect the risk relationship between the vehicle and the obstacle from the side. If its size repeatedly jitters abnormally, it means that a collision event may or may not occur, which shows that the existence of the obstacle may be unreasonable. Therefore, for the obstacles that have passed the previous step, the corresponding expected deceleration is calculated based on the kinematic relationship. The specific calculation is as follows:

[0099] ;

[0100] Where, Represents the deceleration of each obstacle to avoid collision, that is, the expected deceleration, represents the vehicle speed, represents the obstacle speed, represents the longitudinal distance of the obstacle, Represents the wheelbase of the vehicle.

[0101] After passing the obstacle in the previous step, the expected deceleration of each obstacle can be obtained according to the deceleration calculation formula. At this time, the expected deceleration of the obstacle under the historical 1.5s trajectory is recorded through the container. The deceleration value is rolled back in a loop to ensure the historical state of 1.5s at all times. If the deceleration fluctuates by more than If the deceleration fluctuates significantly, and the number of fluctuations exceeds one, this may indicate an obstacle or abnormal position and speed changes of the vehicle. This means that the obstacle is considered low-reliability and can be filtered out. This threshold should not be set too low, as it can easily miss risky targets. Obstacles that meet the requirements are recorded and sorted.

[0102] Step 8: Sort the remaining obstacles from steps 6 and 7, prioritizing obstacles passed in step 7 and then those passed in step 6. The sorting principle is that the smaller the expected deceleration, the closer it is to the front, followed by the potential collision time. The smaller the collision time, the closer it is to the front. After sorting, there will be no more than 32 obstacles.

[0103] Finally, this strategy is executed in each cycle to correct and filter the original obstacles, and the filtered results are output to the planning module of the AEB system for further judgment.

[0104] This application first filters out extremely unreasonable obstacles, then modifies obstacle attributes based on factors such as vehicle speed and obstacle distance. Finally, it combines lane information and the vehicle's motion state to repeatedly confirm the rationality of the obstacle from different perspectives, including driver intent and the authenticity of the collision event. This effectively filters out unreasonable obstacles detected by visual detection, improving AEB performance while also reducing reliance on the accuracy of visual model detection.

[0105] See also Figure 7 As shown, an embodiment of the present application provides an obstacle filtering device, comprising:

[0106] The obstacle information acquisition module 11 is used to obtain obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in the obstacle set output by the visual perception model;

[0107] The obstacle filtering determination module 12 is configured to determine, based on the obstacle information, initial obstacles that meet a preset filtering condition as obstacles to be filtered; wherein, meeting the preset filtering condition indicates that the accuracy of the obstacles output by the visual perception model is insufficient;

[0108] The obstacle filtering module 13 is configured to filter the obstacles to be filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system can perform processing based on the remaining obstacle set.

[0109] In an optional embodiment, the obstacle filtering determination module 12 may be configured to fit the lane line lateral coordinate corresponding to the longitudinal distance of the initial obstacle and the lane line information. If the difference between the lane line lateral coordinate and the lateral distance of the initial obstacle is greater than a preset distance threshold, the initial obstacle is determined to meet a preset filtering condition and is determined as an obstacle to be filtered.

[0110] In an optional embodiment, the obstacle filtering determination module 12 may be configured to filter the lateral distance and longitudinal distance of the initial obstacle based on a distance filtering coefficient to obtain a current filtered lateral distance and a current filtered longitudinal distance of the initial obstacle; obtain a target historical filtered lateral velocity and a target historical filtered longitudinal velocity of the initial obstacle, wherein the target historical filtered lateral velocity and the target historical filtered longitudinal velocity are historical filtered lateral velocities and historical filtered longitudinal velocities with a preset time interval from the current time interval; calculate a predicted lateral distance based on the target historical filtered lateral velocity and the preset time interval; calculate a predicted longitudinal distance based on the target historical filtered longitudinal velocity and the preset time interval; and if a difference between the predicted lateral distance and the current filtered lateral distance or between the predicted longitudinal distance and the current filtered longitudinal distance satisfies a preset difference condition, determine that the initial obstacle satisfies a preset filtering condition, and determine the initial obstacle as an obstacle to be filtered.

[0111] In an optional embodiment, the obstacle filtering determination module 12 may be configured to determine a collision time of the initial obstacle; if the collision time is less than a preset time threshold, calculate a current expected deceleration of the initial obstacle; and determine whether the initial obstacle meets a preset deceleration fluctuation condition based on the historical expected deceleration of the initial obstacle and the current expected deceleration. If the initial obstacle meets the preset deceleration fluctuation condition, the initial obstacle is determined to meet a preset filtering condition and is determined as an obstacle to be filtered.

[0112] The device further comprises:

[0113] A sorting module is used to sort the obstacles in the remaining obstacle set based on the expected deceleration and the collision time to obtain a sorting result;

[0114] The result output module is configured to output the sorting result to an automatic emergency braking system so that the automatic emergency braking system performs processing based on the sorting result.

[0115] In an optional embodiment, the obstacle filtering determination module 12 can be used to compare the existence probability and category accuracy probability of the initial obstacle with an existence probability threshold and an accuracy probability threshold, respectively; when the existence probability is less than the existence probability threshold or the category accuracy probability is less than the accuracy probability threshold, it is determined that the initial obstacle meets the preset filtering condition and the initial obstacle is determined as an obstacle to be filtered.

[0116] In an optional embodiment, the obstacle filtering determination module 12 may be configured to compare the lateral and longitudinal distance variances and the lateral and longitudinal speed variances of the initial obstacle with a distance variance threshold and an obstacle speed variance threshold, respectively; and if the lateral and longitudinal distance variances are greater than the distance variance thresholds or the lateral and longitudinal speed variances are greater than the obstacle speed variance thresholds, determine that the initial obstacle meets the preset filtering condition and determine the initial obstacle as an obstacle to be filtered.

[0117] It can be seen that in the embodiment of the present application, after the visual perception model outputs an obstacle set, the initial obstacles that meet the preset filtering conditions, i.e., inaccurately identified obstacles, are determined based on the information of the initial obstacles in the obstacle set, as the obstacles to be filtered. After filtering, the remaining obstacle set is provided to the automatic emergency braking system. In this way, after the visual perception model is output, filtering is performed, which can effectively filter out unreasonable obstacles detected by vision, reduce the probability of providing incorrect obstacles to the automatic emergency braking system, and improve safety.

[0118] See also Figure 8 As shown, an embodiment of the present application discloses an electronic device 20, including a processor 21 and a memory 22; wherein the memory 22 is used to store a computer program; the processor 21 is used to execute the computer program, the obstacle filtering method disclosed in the above embodiment.

[0119] For the specific process of the above obstacle filtering method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, which will not be described in detail here.

[0120] Furthermore, the memory 22 as a carrier for resource storage may be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., and the storage method may be temporary storage or permanent storage.

[0121] In addition, the electronic device 20 also includes a power supply 23, a communication interface 24, an input / output interface 25 and a communication bus 26; wherein the power supply 23 is used to provide an operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and an external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input / output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0122] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the obstacle filtering method disclosed in the aforementioned embodiment.

[0123] For the specific process of the above obstacle filtering method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, which will not be described in detail here.

[0124] An embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the obstacle filtering method disclosed in the aforementioned embodiment.

[0125] For the specific process of the above obstacle filtering method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, which will not be described in detail here.

[0126] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0127] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0128] The above is a detailed introduction to the obstacle filtering method, device, equipment and medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core ideas. At the same time, for general technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An obstacle filtering method, characterized in that: include: Obtaining obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in an obstacle set output by a visual perception model; Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered; wherein meeting the preset filtering conditions indicates that the accuracy of the obstacles output by the visual perception model is insufficient; The obstacles to be filtered are filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system performs processing based on the remaining obstacle set.

2. The obstacle filtering method according to claim 1, characterized in that: Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered, includes: Fitting the lane line transverse coordinate corresponding to the longitudinal distance based on the initial obstacle longitudinal distance and lane line information; If the difference between the lateral coordinate of the lane line and the lateral distance of the initial obstacle is greater than a preset distance threshold, it is determined that the initial obstacle meets the preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

3. The obstacle filtering method according to claim 1, characterized in that: Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered, includes: Filtering the horizontal distance and the vertical distance of the initial obstacle based on the distance filter coefficient to obtain the current filtered horizontal distance and the current filtered vertical distance of the initial obstacle; Obtaining a target historical filtered lateral velocity and a target historical filtered longitudinal velocity of the initial obstacle; wherein the target historical filtered lateral velocity and the target historical filtered longitudinal velocity are historical filtered lateral velocities and historical filtered longitudinal velocities that are at a preset time interval from the current time interval; Calculating a predicted lateral distance based on the target's historical filtered lateral velocity and the preset time duration; Calculating a predicted longitudinal distance based on the target's historical filtered longitudinal speed and the preset duration; If the difference between the lateral predicted distance and the current filtered lateral distance or the difference between the longitudinal predicted distance and the current filtered longitudinal distance meets a preset difference condition, it is determined that the initial obstacle meets the preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

4. The obstacle filtering method according to claim 1, characterized in that: Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered, includes: determining a collision time of the initial obstacle; When the collision time is less than a preset time threshold, calculating a current expected deceleration of the initial obstacle; Based on the historical expected deceleration of the initial obstacle and the current expected deceleration, it is determined whether the initial obstacle meets a preset deceleration fluctuation condition. If the initial obstacle meets the preset deceleration fluctuation condition, it is determined that the initial obstacle meets a preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

5. The obstacle filtering method according to claim 4, characterized in that: Also includes: Sort the obstacles in the remaining obstacle set based on the expected deceleration and the collision time to obtain a sorting result; The sorting result is output to an automatic emergency braking system so that the automatic emergency braking system performs processing based on the sorting result.

6. The obstacle filtering method according to claim 1, characterized in that: Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered, includes: Comparing the existence probability and category accuracy probability of the initial obstacle with the existence probability threshold and the accuracy probability threshold respectively; When the existence probability is less than the existence probability threshold or the category accuracy probability is less than the accuracy probability threshold, it is determined that the initial obstacle meets the preset filtering condition, and the initial obstacle is determined as an obstacle to be filtered.

7. The obstacle filtering method according to any one of claims 1 to 6, characterized in that: Determining, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered, includes: Comparing the horizontal and vertical distance variances and the horizontal and vertical speed variances of the initial obstacle with the distance variance threshold and the obstacle speed variance threshold respectively; When the lateral and longitudinal distance variance is greater than the distance variance threshold or the lateral and longitudinal speed variance is greater than the obstacle speed variance threshold, it is determined that the initial obstacle meets the preset filtering condition and the initial obstacle is determined as an obstacle to be filtered.

8. An obstacle filtering device, characterized in that: include: An obstacle information acquisition module, configured to acquire obstacle information of an initial obstacle, wherein the initial obstacle is an obstacle in an obstacle set output by a visual perception model; a filtered obstacle determination module, configured to determine, based on the obstacle information, initial obstacles that meet preset filtering conditions as obstacles to be filtered; wherein meeting the preset filtering conditions indicates that the accuracy of the obstacles output by the visual perception model is insufficient; The obstacle filtering module is configured to filter the obstacles to be filtered from the obstacle set to obtain a remaining obstacle set, so that the automatic emergency braking system can perform processing based on the remaining obstacle set.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program to implement the obstacle filtering method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program, wherein when the computer program is executed by a processor, the obstacle filtering method according to any one of claims 1 to 7 is implemented.

Images