Obstacle detection method and device and computer equipment

Abstract

The invention relates to the technical field of image processing, in particular to an obstacle detection method and device and computer equipment, and the method comprises the steps: 1, obtaining a road surface image, marking a road section region corresponding to the road surface image as a monitored road section, dividing the road surface image into a plurality of lane rectangles according to road segmentation marking lines, dividing the lane rectangle into a plurality of unit rectangles along the length direction of the lane; 2, after the vehicle enters the monitored road section, forming a vehicle passing image set by a plurality of acquired road surface images, determining a plurality of passing related lanes based on the vehicle image set, and comparing and analyzing the images corresponding to the passing related lanes to determine a road surface abnormal rectangle; 3, marking the unit rectangle as a first abnormal rectangle or a second abnormal rectangle based on the data when the passing vehicle passes through the monitored road section; 4, comprehensively calculating the abnormal confidence coefficient of the unit rectangle based on the number of marked times of the unit rectangle; and 5, generating different signals based on the abnormal confidence degree of the unit rectangle.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an obstacle detection method, device and computer equipment. Background Art

[0002] The obstacle monitoring method is a core technology that uses multiple sensors and intelligent algorithms to sense and identify obstacles in the environment in real time, providing safe navigation and decision-making basis for autonomous systems. The current mainstream solution uses multi-sensor fusion, combined with Kalman filtering, deep learning and other algorithms, to achieve all-weather, highly robust obstacle detection in the fields of autonomous driving, robotics, and intelligent transportation, taking into account both real-time performance and adaptability to complex scenarios, and promoting the evolution of intelligent systems from collision avoidance warning to fully autonomous decision-making.

[0003] The current mainstream obstacle detection method can achieve safety warnings and route avoidance during vehicle driving through multiple types of sensors carried by the vehicle. However, this type of detection method is limited by the vehicle's own conditions, such as the number of sensors carried by the vehicle, the sensitivity of the vehicle's obstacle monitoring system, and the vehicle's moving speed, resulting in inconsistent obstacle detection effects for different vehicles, affecting road traffic safety. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present invention provides an obstacle detection method, an apparatus and a computer device, which can effectively solve the problem of inconsistent obstacle detection effects of different vehicles in the prior art.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention provides an obstacle detection method, device and computer equipment, which at least include the following steps: Step 1: Obtain a road surface image, record the road section area corresponding to the road surface image as a monitoring section, divide the road surface image into a plurality of lane rectangles according to the road segmentation markings, and divide the lane rectangle into a plurality of unit rectangles along the lane length direction; Step 2: periodically acquire road images and record them as road periodic images. When a vehicle enters a monitored road section, acquire road images and record them as road vehicle images. Multiple road periodic images acquired after acquiring the road vehicle images and the road vehicle images together constitute a vehicle passing image set. Each vehicle passing image set corresponds to a triggering vehicle. Based on the vehicle image set analysis, multiple passing lanes corresponding to the triggering vehicle are obtained, and image comparison analysis is performed on multiple unit rectangles corresponding to the passing lanes, where: Obtaining rectangular images of the unit rectangle before and after the triggering vehicle passes through and comparing them, and determining the abnormal road surface rectangle based on the similarity between the two rectangular images; Step 3: Based on the analysis of the acceleration change and steering wheel torque change when the passing vehicle passes through the monitored road section, the unit rectangle is marked as the first abnormal rectangle or the second abnormal rectangle; Step 4: Based on the number of times the unit rectangle is marked, comprehensively calculate the abnormality confidence of the unit rectangle; Step 5: Compare the abnormality confidence of the unit rectangle with the preset first confidence threshold and second confidence threshold, and generate different signals based on the abnormality confidence of the unit rectangle.

[0006] Furthermore, the vehicle passing image set construction process is as follows: Get the speed limit value of the monitored section , substitute into the formula Calculate the correlation period ,in In order to monitor the length of the road section, the time of collecting road vehicle images is , will be in the time interval The road surface images collected periodically are recorded as The associated images of the road vehicle images at that moment, and the road vehicle images and their associated images together constitute a vehicle passing image set.

[0007] Furthermore, the process of obtaining the relevant lanes is as follows: The road surface image in the vehicle passing image set is recorded as the related image, the vehicle outline of the triggering vehicle is marked in the related image, the minimum rectangle containing the vehicle outline is constructed as the vehicle rectangle, and the straight line passing through the midpoints of the two short sides of the vehicle rectangle is constructed as the vehicle body direction straight line; In the relevant image, the lane center lines corresponding to each lane rectangle are drawn and recorded as , f is the serial number of the lane centerline; The intersection points of the vehicle body direction straight lines and the center lines of each lane in the relevant images are recorded as passing intersection points, and the lanes containing the passing intersection points are recorded as passing related lanes.

[0008] Furthermore, the process of marking the abnormal road surface rectangle is as follows: Get the time intervals separately The two road surface comparison images collected most recently before and after are recorded as comparison images; Obtain two rectangular images corresponding to the same unit rectangle in the two comparison images, calculate the similarity of the two rectangular images, and when the similarity of the two rectangles is less than or equal to a preset similarity threshold, record the corresponding unit rectangle as an abnormal rectangular area; The similarities of multiple abnormal rectangular areas are obtained. When the similarities of multiple abnormal rectangular areas are not equal, the abnormal rectangular area is marked as a road surface abnormal rectangle, and the corresponding position is marked as a suspected obstacle position.

[0009] Furthermore, the first abnormal rectangle marking process is as follows: A time limit period is preset, and vehicles passing through the monitored road section within the time limit period are obtained as target vehicles. The driving route of the target vehicle within the time limit period is obtained, and any point on the driving route corresponds to a passing time, and the passing time interval corresponding to the monitored road section is obtained; Obtaining a vertical acceleration change curve of the target vehicle within a time interval, obtaining multiple peak points of the vertical acceleration change curve, presetting an acceleration threshold, and recording the moment corresponding to the peak point where the vertical acceleration value is greater than the acceleration threshold as the first abnormal moment; The vehicle position corresponding to the first abnormal time is obtained and recorded as the first abnormal position, and the unit rectangle including the first abnormal position is marked as the first abnormal rectangle.

[0010] Furthermore, the second abnormal rectangle marking process is as follows: Obtain a steering wheel torque change curve of the target vehicle within a time interval, preset a torque threshold, record the portion of the steering wheel torque change curve that is higher than the torque threshold as a torque abnormality portion, analyze the steering wheel rotation angle within the time interval corresponding to the torque abnormality portion, and obtain a steering abnormality value; The steering abnormality value is compared with a preset steering abnormality threshold value, and when the steering abnormality value is greater than the preset steering abnormality threshold value, the start time of the torque abnormal part is recorded as the second abnormal time; The vehicle position corresponding to the second abnormal moment is obtained and recorded as the second abnormal position, and all unit rectangles corresponding to the lane where the second abnormal position is located are marked as second abnormal rectangles.

[0011] Furthermore, the steering outlier value calculation process is specifically as follows: Assume that the time interval corresponding to the abnormal torque is , get the time interval Angle series of inner steering wheel rotation , the angle sequence contains multiple rotation angles , each rotation angle corresponds to a one-way rotation of the steering wheel; When the one-way rotation direction is clockwise, the rotation angle Positive, when the unidirectional rotation direction is counterclockwise, the rotation angle is negative; Substitute into the formula Calculation is performed in Get the one-way rotation amplitude and the total rotation amplitude ; in: ; Indicates taking The maximum value between The one-way rotation range and the total rotation amplitude Substitute into the formula Calculate in and get the steering anomaly value, where All are preset weight coefficients.

[0012] Furthermore, the calculation process of the abnormality confidence of the unit rectangle is as follows: Obtain the number of times each unit rectangle is marked as a road surface anomaly rectangle and record it as the first impact value , obtain the number of times each unit rectangle is marked as the first abnormal rectangle and record it as the second influence value , obtain the number of times each unit rectangle is marked as the second abnormal rectangle and record it as the third influence value ; Substitute into the formula Calculate the abnormal confidence of each unit rectangle, where: All are preset weight coefficients; m is the number of unit rectangles on each lane rectangle.

[0013] The obstacle detection device includes road surface monitoring equipment arranged on both sides of the road, and the road surface monitoring equipment faces the road surface; It also includes a vehicle-mounted device, which is electrically connected to a three-axis accelerometer and a torque sensor. The three-axis accelerometer is arranged at the front of the vehicle to measure the three-axis acceleration of the vehicle, and the torque sensor is arranged at the steering wheel shaft to measure the torque when the steering wheel is turned.

[0014] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0015] Compared with the known prior art, the technical solution provided by the present invention has the following beneficial effects: 1. Based on the existing road surface monitoring, the present invention can determine the lane through which the triggering vehicle passes by analyzing multiple images collected when the triggering vehicle passes, and then compare and analyze the images before and after the triggering vehicle passes corresponding to each lane to determine whether the triggering vehicle drops objects to form road obstacles. This method utilizes the characteristic that "most road obstacles come from vehicle drops" and effectively improves the efficiency of road surface monitoring in identifying obstacles.

[0016] 2. The present invention integrates vehicle driving data with road monitoring data to comprehensively determine the location information of obstacles. On the one hand, it can reduce the equipment pressure of road monitoring equipment in identifying obstacles. On the other hand, it can make full use of the driving data of different vehicles passing through the monitored road section to analyze the road conditions. The combination of the two can effectively improve the accuracy of obstacle identification. Compared with the existing single-angle obstacle identification method, it is more suitable for road safety management for the driving safety of all vehicles under the current rapid development of the "Internet of Vehicles". BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The present invention will be further described below in conjunction with the embodiments.

[0021] The current mainstream obstacle detection method can achieve safety warning and route avoidance during vehicle driving through multiple types of sensors carried by the vehicle. However, the ability of this type of detection method to identify obstacles is limited by the vehicle's own conditions, such as the number of sensors carried by the vehicle, the sensitivity of the vehicle's obstacle monitoring system, and the vehicle's moving speed. This results in different vehicles having different recognition effects on the same obstacle on the road, causing some vehicles to be unable to detect road obstacles in time, thus causing road hazards. Therefore, there is an urgent need for an obstacle detection method that can call on road monitoring and combine the vehicle's own obstacle recognition system and map navigation to achieve uniform obstacle detection and obstacle avoidance push to ensure the driving safety of all vehicles on the road.

[0022] See also Figure 1 , the obstacle detection method at least comprises the following steps: Step 1: Obtain road images through road monitoring, perform independent analysis on the monitoring area of ​​each road monitoring camera, obtain the road section area corresponding to the road monitoring camera (the road section covered by each road monitoring camera is calibrated at the initial stage of installation to clarify the road section area corresponding to each road monitoring camera) and record it as the monitored road section. Divide the road image corresponding to the road monitoring camera into multiple rectangular areas according to the road dividing markings and record them as lane rectangles. Each lane rectangle corresponds to a lane, and the lane rectangle is further divided into multiple unit rectangles along the length direction of the lane; Step 2: Identify suspected obstacles on the road based on the road image, where: An image fixed acquisition cycle is preset, and road images are regularly acquired at intervals of the image fixed acquisition cycle and recorded as road periodic images. When a vehicle passes through a monitored road section, road images are acquired and recorded as road vehicle images. Vehicles in the road periodic images are identified and their number is counted, and road periodic images without vehicles are recorded as road comparison images (vehicle identification and vehicle passing detection are existing image recognition technologies, which will not be elaborated here); It should be noted that road vehicle images and road periodic images do not conflict with each other. The same road image can be both a road vehicle image and a road periodic image. The difference lies in the different triggering conditions for road image acquisition. Whenever a vehicle enters the monitored section, a road vehicle image will be generated.

[0023] Get the speed limit value of the monitored section , substitute into the formula Calculate the correlation period ,in To monitor the length of the road section, assume that any road vehicle image acquisition time is , will be in the time interval The road surface periodic images collected are recorded as the associated images of the road vehicle images, the road vehicle images and their associated images together form a vehicle passing image set, the vehicle corresponding to the road vehicle image (i.e., the vehicle that triggers the image collection signal) is recorded as the triggering vehicle, and each vehicle passing image set is bound to the corresponding triggering vehicle; It should be noted that the road monitoring equipment in the prior art can intelligently identify the vehicle that triggers the image acquisition signal when collecting road vehicle images. The same function is also applied to radar speed measurement capture equipment and lane change capture equipment. The road vehicle image and its associated images correspond to multiple images of a vehicle traveling in the monitored section. Through these images, the trajectory analysis of vehicles passing through the monitored section can be performed to determine the lanes that the vehicles pass through. In addition, by limiting the association period, the triggering vehicle can be included in the associated image (when the triggering vehicle does not exceed the speed limit).

[0024] Analyze the elements in the vehicle passing image set one by one to obtain multiple passing related lanes, among which: The road surface image in the vehicle passing image set is recorded as the related image, the vehicle outline of the triggering vehicle is marked in the related image, the minimum rectangle containing the vehicle outline is constructed as the vehicle rectangle, and the straight line passing through the midpoints of the two short sides of the vehicle rectangle is constructed as the vehicle body direction straight line (the vehicle body direction straight line is used to indicate the driving direction of the vehicle), and the lane center line corresponding to each lane rectangle is drawn in the related image (the lane center line is a length equal to The line segments of , f is the serial number of the lane centerline (corresponding to the serial number of the lane from left to right), obtain the intersection point of the vehicle body direction straight line in the relevant image (limited to the line segment in the relevant image but not including its extension line) and the centerline of each lane as the passing intersection point, and record the lane corresponding to the lane rectangle containing the passing intersection point as the passing related lane.

[0025] It should be noted that the passing lanes correspond to lanes that the triggering vehicle may pass through, and obstacles on the lanes usually come from passing vehicles. Therefore, obstacle recognition and analysis of the road surface in the passing lanes can more efficiently identify obstacles. Each vehicle passing image set corresponds to one or more passing lanes.

[0026] Get the time intervals (corresponding to each vehicle passing through the image set) respectively The two road comparison images collected most recently before and after are recorded as comparison images. Based on the front-to-back changes corresponding to multiple lanes passing through in the comparison images, the suspected position of the obstacle is determined. The analysis process is as follows: Obtain two rectangular images corresponding to the same unit rectangle in the two comparison images, calculate the similarity of the two rectangular images, and when the similarity of the two rectangles is less than or equal to a preset similarity threshold, record the corresponding unit rectangle as an abnormal rectangular area, obtain the similarity of multiple abnormal rectangular areas, and when the similarities of multiple abnormal rectangular areas are not equal, mark the abnormal rectangular area as a road surface abnormal rectangle, and mark the corresponding position as a suspected obstacle position; It should be noted that when a vehicle passes through a puddle, it will leave water marks on the road surface. Such water marks will mislead the image comparison and will also cause the similarity of the unit rectangle to change. By comparing the similarity of multiple abnormal rectangular areas, the influence of water marks can be eliminated, because in general, fallen obstacles will not form multiple abnormal rectangular areas with equal similarity.

[0027] Step 2 can, based on the existing road surface monitoring, determine the lane through which the triggering vehicle passes by analyzing multiple images collected when the triggering vehicle passes, and then compare and analyze the images before and after the triggering vehicle passes corresponding to each lane to determine whether the triggering vehicle drops objects to form road obstacles. This method utilizes the characteristic that "most road obstacles come from dropped vehicles" and effectively improves the efficiency of road monitoring in identifying obstacles.

[0028] Step 3: Record vehicles that pass through the monitored section as passing vehicles (the determination of passing vehicles can be based on the driving route provided by the vehicle navigation software), obtain the driving trajectory of the passing vehicle through the monitored section through the vehicle, and determine the suspected location of the obstacle based on the driving trajectory data of the passing vehicle, where: A time limit period is preset, and vehicles passing through the monitored road section within the time limit period are obtained as target vehicles. The driving route of the target vehicle within the time limit period is obtained, and any point on the driving route corresponds to a passing time, and the passing time interval corresponding to the monitored road section is obtained; Obtain a vertical acceleration change curve of the target vehicle within the time interval (obtained by a triaxial accelerometer inside the vehicle), obtain multiple peak points of the vertical acceleration change curve (i.e., peak points of multiple peaks), preset an acceleration threshold, and record the moment corresponding to the peak point where the vertical acceleration value is greater than the acceleration threshold as the first abnormal moment; Obtain the steering wheel torque change curve of the target vehicle within the time interval (i.e., the force applied by the driver to turn the vehicle, which can be obtained through the sensor in the steering wheel). A torque threshold is preset, and the part of the steering wheel torque change curve that is higher than the torque threshold is recorded as the torque abnormal part. The time interval corresponding to the torque abnormal part is set to , get the time interval Angle series of inner steering wheel rotation , the angle sequence contains multiple rotation angles , each rotation angle corresponds to a one-way rotation of the steering wheel (time interval The inner steering wheel rotation process can be divided into multiple one-way rotations), with clockwise rotation as the positive direction and counterclockwise rotation as the negative direction. When the one-way rotation direction is positive, the rotation angle is is positive, when the one-way rotation direction is reverse, the rotation angle is negative, substitute into the formula Calculate the unidirectional rotation amplitude and the total rotation amplitude ,in: ; Indicates taking The maximum value between The one-way rotation range and the total rotation amplitude Substitute into the formula Calculate in and get the steering anomaly value, where are all preset weight coefficients. When the steering abnormality value is greater than the preset steering abnormality threshold, The moment is recorded as the second abnormal moment; It should be noted that when the driver finds an obstacle and tries to avoid it, he will turn the steering wheel to make an emergency lane change. At this time, the steering wheel will show a significantly abnormal rotation, which is different from the slight rotation during normal driving. Therefore, the one-way angle and the overall angle of the steering wheel during rotation can be used to determine whether the steering wheel is in a condition to avoid obstacles.

[0029] The vehicle position corresponding to the first abnormal moment is obtained and recorded as the first abnormal position, and the unit rectangle containing the first abnormal position is marked as the first abnormal rectangle. The vehicle position corresponding to the second abnormal moment is obtained and recorded as the second abnormal position, and all unit rectangles corresponding to the lane where the second abnormal position is located are marked as the second abnormal rectangle.

[0030] By analyzing the driving data of passing vehicles to obtain the first abnormal rectangle and the second abnormal rectangle, it is helpful to locate the area where obstacles may exist, so as to make full use of the driving data of each passing vehicle to further verify the position of the obstacle on the basis of the identification of the road monitoring equipment, thereby improving the diversity of obstacle detection methods and the accuracy of obstacle detection.

[0031] It should be noted that although road monitoring cameras can also capture the trajectory of vehicles passing through the monitored road sections, if the trajectory analysis of passing vehicles is performed through road monitoring cameras, it means that it is necessary to save the passing videos of multiple passing vehicles at the same time and extract the trajectories of these videos and further analyze the obstacles. This has extremely high requirements on the storage capacity and computing power of the monitoring equipment, and the actual implementation cost is too high. In contrast, using the trajectory data provided by the vehicle computer and performing an independent analysis method of "one vehicle, one calculation" can fully share the computing power pressure while avoiding interference from other vehicle routes, thereby improving analysis efficiency and accuracy.

[0032] Step 4: Combine the obstacle position analysis results in steps 2 and 3 to comprehensively analyze the confidence of obstacles at different positions, where: Obtain the number of times each unit rectangle is marked as a road surface anomaly rectangle and record it as the first impact value , obtain the number of times each unit rectangle is marked as the first abnormal rectangle and record it as the second influence value , obtain the number of times each unit rectangle is marked as the second abnormal rectangle and record it as the third influence value , substitute into the formula Calculate the anomaly confidence of each unit rectangle, where are all preset weight coefficients, m is the number of unit rectangles on each lane rectangle; It should be noted that the abnormality confidence of each unit rectangle is proportional to the number of times the unit rectangle is judged as a road surface abnormality rectangle, a first abnormal rectangle, and a second abnormal rectangle. The more times a unit rectangle is judged as an abnormal rectangle (regardless of which type of abnormal rectangle), and it is judged as an abnormal rectangle under multiple judgment methods, it means that there is a high possibility that there is an obstacle in the area corresponding to the unit rectangle. Therefore, the abnormality confidence of the unit rectangle can help staff determine whether there is an obstacle in the area corresponding to the unit rectangle and the possibility of the existence of an obstacle.

[0033] Step 5: A first confidence threshold and a second confidence threshold are preset, and the first confidence threshold is less than the second confidence threshold. When the abnormal confidence of the unit rectangle is greater than or equal to the first confidence threshold and less than the second confidence threshold, a regional abnormal signal is generated, and the road monitoring image corresponding to the unit rectangle is retrieved and the area corresponding to the unit rectangle is marked, so that the staff can check whether there are obstacles in the area in time and clear them in time if there are obstacles. When the abnormal confidence of the unit rectangle is greater than or equal to the second confidence threshold, an area avoidance signal is generated, the area position corresponding to the unit rectangle is obtained and recorded as the obstacle area, the driving routes of multiple passing vehicles are obtained, and the vehicles that do not pass the obstacle area in the current driving route are recorded as avoidance vehicles. When the avoidance vehicle is R meters away from the obstacle area, the navigation software sends an avoidance warning message to the avoidance vehicle. The avoidance warning message can be a sound reminder or the dashboard indicator light flashes. The calculation formula of the distance R is: , The preset reaction time.

[0034] The output instructions based on the abnormal confidence are adjusted according to the pre-set first confidence threshold and the second confidence threshold, so that the method in the present invention can adjust the final output strategy according to the different confidence levels, so as to avoid the impact of the misidentification of obstacles on the traffic order as much as possible without affecting the obstacle processing speed (when an obstacle is misidentified, the confidence level is often not high, and at this time no warning instructions will be sent to all vehicles passing through the area), thereby ensuring normal road traffic on the monitored section.

[0035] The present invention can combine the existing road monitoring and identification technology with the data recorded by the vehicle navigation software, so as to detect and locate the position of road obstacles from different angles, and further build confidence by combining the results of analysis in different ways, so as to improve the accuracy of the final output result. Compared with the single-angle detection and identification technology in the prior art, the present invention has weaker dependence on the number and type of the vehicle's own sensors, and is suitable for all vehicles, so as to ensure the uniform obstacle avoidance effect of all vehicles.

[0036] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above method when executing the computer program.

[0037] The obstacle detection device includes road surface monitoring equipment arranged on both sides of the road, and the road surface monitoring equipment faces the road surface; It also includes a vehicle-mounted device, which is electrically connected to a three-axis accelerometer and a torque sensor. The three-axis accelerometer is arranged at the front of the vehicle to measure the three-axis acceleration of the vehicle, and the torque sensor is arranged at the steering wheel shaft to measure the torque when the steering wheel is turned.

[0038] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An obstacle detection method, characterized in that: The following steps are involved: Step 1: Obtain a road surface image, record the road section area corresponding to the road surface image as a monitoring section, divide the road surface image into a plurality of lane rectangles according to the road segmentation markings, and divide the lane rectangle into a plurality of unit rectangles along the lane length direction; Step 2: periodically acquire road images and record them as road periodic images. When a vehicle enters a monitored road section, acquire road images and record them as road vehicle images. Multiple road periodic images acquired after acquiring the road vehicle images and the road vehicle images together constitute a vehicle passing image set. Each vehicle passing image set corresponds to a triggering vehicle. Based on the vehicle image set analysis, multiple passing lanes corresponding to the triggering vehicle are obtained, and image comparison analysis is performed on multiple unit rectangles corresponding to the passing lanes, where: Obtaining rectangular images of the unit rectangle before and after the triggering vehicle passes through and comparing them, and determining the abnormal road surface rectangle based on the similarity between the two rectangular images; Step 3: Based on the analysis of the acceleration change and steering wheel torque change when the passing vehicle passes through the monitored road section, the unit rectangle is marked as the first abnormal rectangle or the second abnormal rectangle; Step 4: Based on the number of times the unit rectangle is marked, comprehensively calculate the abnormality confidence of the unit rectangle; Step 5: Compare the abnormality confidence of the unit rectangle with the preset first confidence threshold and second confidence threshold, and generate different signals based on the abnormality confidence of the unit rectangle.

2. The obstacle detection method according to claim 1, characterized in that: The process of constructing the vehicle passing image set is as follows: Get the speed limit value of the monitored section , substitute into the formula Calculate the correlation period ,in In order to monitor the length of the road section, the time of collecting road vehicle images is , will be in the time interval The road surface images collected periodically are recorded as The associated images of the road vehicle images at the moment, the road vehicle images and the corresponding associated images together form a vehicle passing image set.

3. The obstacle detection method according to claim 1, characterized in that: The process of obtaining the relevant lanes is as follows: The road surface image in the vehicle passing image set is recorded as the related image, the vehicle outline of the triggering vehicle is marked in the related image, the minimum rectangle containing the vehicle outline is constructed as the vehicle rectangle, and the straight line passing through the midpoints of the two short sides of the vehicle rectangle is constructed as the vehicle body direction straight line; In the relevant image, the lane center lines corresponding to each lane rectangle are drawn and recorded as , f is the serial number of the lane centerline; The intersection points of the vehicle body direction straight lines and the center lines of each lane in the relevant images are recorded as passing intersection points, and the lanes containing the passing intersection points are recorded as passing related lanes.

4. The obstacle detection method according to claim 2, characterized in that: The process of marking the road surface abnormality rectangle is as follows: Get the time intervals separately The two road surface comparison images collected most recently before and after are recorded as comparison images; Obtain two rectangular images corresponding to the same unit rectangle in the two comparison images, calculate the similarity of the two rectangular images, and when the similarity of the two rectangles is less than or equal to a preset similarity threshold, record the corresponding unit rectangle as an abnormal rectangular area; The similarities of multiple abnormal rectangular areas are obtained. When the similarities of multiple abnormal rectangular areas are not equal, the abnormal rectangular area is marked as a road surface abnormal rectangle, and the corresponding position is marked as a suspected obstacle position.

5. The obstacle detection method according to claim 1, characterized in that: The first abnormal rectangle marking process is as follows: A time limit period is preset, and vehicles passing through the monitored road section within the time limit period are obtained as target vehicles. The driving route of the target vehicle within the time limit period is obtained, and any point on the driving route corresponds to a passing time, and the passing time interval corresponding to the monitored road section is obtained; Obtaining a vertical acceleration change curve of the target vehicle within a time interval, obtaining multiple peak points of the vertical acceleration change curve, presetting an acceleration threshold, and recording the moment corresponding to the peak point where the vertical acceleration value is greater than the acceleration threshold as the first abnormal moment; The vehicle position corresponding to the first abnormal time is obtained and recorded as the first abnormal position, and the unit rectangle including the first abnormal position is marked as the first abnormal rectangle.

6. The obstacle detection method according to claim 5, characterized in that: The second abnormal rectangle marking process is as follows: Obtain a steering wheel torque change curve of the target vehicle within a time interval, preset a torque threshold, record the portion of the steering wheel torque change curve that is higher than the torque threshold as a torque abnormality portion, analyze the steering wheel rotation angle within the time interval corresponding to the torque abnormality portion, and obtain a steering abnormality value; The steering abnormality value is compared with a preset steering abnormality threshold value, and when the steering abnormality value is greater than the preset steering abnormality threshold value, the start time of the torque abnormal part is recorded as the second abnormal time; The vehicle position corresponding to the second abnormal moment is obtained and recorded as the second abnormal position, and all unit rectangles corresponding to the lane where the second abnormal position is located are marked as second abnormal rectangles.

7. The obstacle detection method according to claim 6, characterized in that: The steering outlier value calculation process is specifically as follows: Assume that the time interval corresponding to the abnormal torque is , get the time interval Angle series of inner steering wheel rotation , the angle sequence contains multiple rotation angles , each rotation angle corresponds to a one-way rotation of the steering wheel; When the one-way rotation direction is clockwise, the rotation angle Positive, when the unidirectional rotation direction is counterclockwise, the rotation angle is negative; Substitute into the formula Calculation is performed in Get the one-way rotation amplitude and the total rotation amplitude ; in: ； Indicates taking The maximum value between The one-way rotation range and the total rotation amplitude Substitute into the formula Calculate in and get the steering anomaly value, where All are preset weight coefficients.

8. The obstacle detection method according to claim 1, characterized in that: The calculation process of the abnormality confidence of the unit rectangle is as follows: Obtain the number of times each unit rectangle is marked as a road surface anomaly rectangle and record it as the first impact value , obtain the number of times each unit rectangle is marked as the first abnormal rectangle and record it as the second influence value , obtain the number of times each unit rectangle is marked as the second abnormal rectangle and record it as the third influence value ; Substitute into the formula Calculate the abnormal confidence of each unit rectangle, where: All are preset weight coefficients; m is the number of unit rectangles on each lane rectangle.

9. The obstacle detection device is characterized in that: The device comprises: Road surface monitoring equipment is arranged on both sides of the road, facing the road surface and used to obtain road surface images; It also includes a vehicle-mounted device, which is electrically connected to a three-axis accelerometer and a torque sensor. The three-axis accelerometer is arranged at the front of the vehicle to measure the three-axis acceleration of the vehicle, and the torque sensor is arranged at the steering wheel shaft to measure the torque when the steering wheel is turned.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images