Road pit detection method based on laser radar and related device

Abstract

The invention provides a road pit detection method based on a laser radar and a related device. The road pit detection method comprises the following steps: S101, performing ground filtering by using a ray method, dividing rays for laser radar point cloud according to polar coordinates, analyzing height continuity point by point along each ray, segmenting an underground area, and obtaining underground point cloud area data as intermediate point cloud data; s102, clustering the intermediate point cloud data by using a Euclidean clustering method; s103, taking the clustered obstacle as a target object, calculating a weighted centroid of each target object by taking the point cloud reflection intensity as a weight, and considering the target object with the centroid below the ground as a road pit; and S104, acquiring specific information of the vehicle, predicting the edge of each road pit according to the specific information and the pit information, and acquiring pit edge information. Ray method ground filtering and Euclidean clustering are adopted, and edge prediction is carried out in combination with the reflection intensity weighted centroid and specific information.

Description

Technical Field

[0001] The present application relates to the fields of autonomous driving and pothole detection, and in particular to a road pothole detection method and related devices based on lidar. Background Art

[0002] Currently, pothole detection technology in the autonomous driving field primarily relies on laser-camera fusion or deep learning-based detection methods. While these two technologies offer advantages in accuracy and reliability, they also present numerous challenges in practical applications.

[0003] The laser-camera fusion method combines the precise distance information provided by the lidar with the visual images captured by the camera to identify potholes and other obstacles on the road. However, this method has the following drawbacks: The complex calibration and fusion process is necessary to achieve effective data fusion. This process is not only time-consuming but also requires high-precision specialized equipment and technology. It also has a long deployment cycle, from hardware installation to system debugging, which typically takes days to weeks, increasing the time cost and implementation difficulty of the project.

[0004] Deep learning-based detection methods rely on large, labeled datasets to train models, enabling them to automatically learn features and perform classification and recognition. However, this approach has the following drawbacks: Data annotation and computing resource requirements are high. Training deep learning models requires a large amount of carefully labeled data, making the process time-consuming and labor-intensive. Furthermore, model training often requires the support of high-performance GPU clusters, significantly increasing hardware costs. Deep learning models also consume significant computing resources during actual operation, impacting the responsiveness and overall performance of the autonomous driving system. Research has shown that a complete model training session can take hours to days, and during the inference phase, processing each image takes tens to hundreds of milliseconds, severely impacting real-time detection efficiency.

[0005] Based on this, the present application provides a road pothole detection method and related device based on lidar. Summary of the Invention

[0006] In order to overcome the above-mentioned shortcomings, the present application provides a road pothole detection method and related devices based on lidar.

[0007] The purpose of this application is achieved by the following technical solutions:

[0008] In a first aspect, the present application provides a road pothole detection method based on lidar, comprising the following steps:

[0009] S101, using ray method ground filtering, divides the lidar point cloud into rays according to polar coordinates, analyzes the height continuity point by point along each ray, segments the area below the ground, obtains the point cloud area data below the ground and uses it as the intermediate point cloud data;

[0010] S102, clustering the intermediate point cloud data using the Euclidean clustering method, classifying obstacles from the point cloud data, and obtaining dimension data including length, width, and height of the clustered obstacles;

[0011] S103: The clustered obstacles are regarded as targets, and the weighted centroid of each target is calculated using the point cloud reflection intensity as a weight. Targets whose centroids are below the ground are considered to be potholes. The location data and size data of each pothole are used as pothole information.

[0012] S104, obtaining specific information of the vehicle, wherein the specific information includes the vehicle's current pitch angle data, sensor installation height, and obstacle target height; predicting the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

[0013] The beneficial effects of this technical solution are: ray-based ground filtering effectively separates the point cloud area below the ground, Euclidean clustering accurately identifies obstacles and obtains size data, and then uses weighted centroid calculation to accurately determine potholes. Specific information is integrated to predict the pothole edges. The entire process improves the probability of identifying potholes on the road and more accurately predicts the location of pothole edges, effectively preventing autonomous driving robots from being jolted or even damaged due to not detecting potholes in time, thereby improving their driving safety and reliability. Ray-based ground filtering analyzes the high continuity point by point along each ray, accurately segmenting the point cloud area below the ground and reducing the probability of false detection. The weighted centroid is calculated using the point cloud reflection intensity as the weight, fully considering the characteristics of the point cloud data, avoiding false judgments caused by simple geometric feature judgment, and improving the accuracy of pothole detection. This embodiment abandons camera fusion and deep learning methods, eliminating the need for expensive camera equipment and high-performance computing equipment for model training, greatly reducing hardware costs. Compared with traditional methods, it reduces complex multi-sensor fusion processing and data calibration work, and significantly shortens deployment time.

[0014] Preferably, the S102 includes:

[0015] S201, performing spatial search on the intermediate point cloud data by constructing a KD tree;

[0016] S202, selecting any initial point in the point cloud and performing a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters;

[0017] S203: Continue to perform proximity search for all point cloud points in the set that have not been subjected to proximity search according to the set distance threshold, and add the newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold;

[0018] S204: Reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, calculate its length, width and height and use them as size data;

[0019] S205 , classifying the clustered obstacles from the intermediate point cloud data, wherein each cluster corresponds to one obstacle, and the size data thereof is used as feature information of the obstacle.

[0020] The beneficial effect of this technical solution is that, considering that traditional search methods may require comparing the distances between point clouds one by one, the construction of the KD tree significantly improves the efficiency of spatial search. Through its hierarchical structure, the KD tree can quickly locate point clouds within a specific area, greatly reducing the time and computing resources required for the search. Through proximity search and iterative clustering, point cloud points belonging to the same obstacle can be accurately clustered together. At the same time, setting a reasonable distance threshold helps to screen out point cloud clusters that meet the characteristics of the obstacle, thereby improving the accuracy of obstacle identification. At the same time, calculating the length, width, and height of each cluster further enriches the characteristic information of the obstacle and provides more comprehensive data support for subsequent analysis.

[0021] Preferably, the step of calculating the weighted centroid of each target object using the point cloud reflection intensity as a weight includes:

[0022] The coordinates of the weighted centroid are obtained according to the following formula:

[0023]

[0024] c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height.

[0025] The beneficial effect of this technical solution is that by incorporating point cloud reflection intensity as a weight, weighted centroid calculation can more accurately locate the center of the target object. Because objects with different materials and surface characteristics reflect laser light differently, potholes often have different reflective properties from the surrounding normal road surface. Therefore, the weighted centroid can better reflect the true location and characteristics of the target object. Compared with judgment based solely on geometric shape, it can more accurately distinguish potholes from other obstacles and reduce the error rate.

[0026] Preferably, for each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes:

[0027] The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes:

[0028]

[0029]

[0030] Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

[0031] This technical solution has the beneficial effect of taking into account factors such as the vehicle's pitch angle and tilt height, enabling pothole edge prediction results to adapt to changes in the vehicle's posture under different road conditions. Even when the vehicle is traveling uphill or downhill, or on uneven roads, it can accurately predict the pothole edge, avoiding misjudgments or missed detections due to changes in vehicle posture, thereby improving the reliability and adaptability of the detection method.

[0032] In a second aspect, the present application also provides a road pothole detection device based on laser radar, comprising:

[0033] The ground point cloud acquisition module is used to use ray method ground filtering to divide the lidar point cloud into rays according to polar coordinates, analyze the height continuity point by point along each ray, segment the area below the ground, and obtain the point cloud area data below the ground as the intermediate point cloud data;

[0034] The data clustering module is used to cluster the intermediate point cloud data using the Euclidean clustering method, classify obstacles from the point cloud data, and obtain the dimension data of the clustered obstacles including length, width and height;

[0035] The pothole detection module is used to treat clustered obstacles as targets, calculate the weighted centroid of each target using the point cloud reflection intensity as the weight, and identify targets with the centroid below the ground as potholes. The location and size data of each pothole are used as pothole information.

[0036] The edge acquisition module is used to obtain specific information about the vehicle, including the vehicle's current pitch angle data, sensor installation height, and obstacle target height; and predict the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

[0037] Preferably, the data clustering module includes:

[0038] The spatial search unit is used to perform spatial search on the intermediate point cloud data by constructing a KD tree;

[0039] The first set acquisition unit is used to select any initial point in the point cloud and perform a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters;

[0040] The second set acquisition unit is configured to continue to perform proximity search according to the set distance threshold for all point cloud points in the set that have not been subjected to proximity search, and add newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold;

[0041] A cluster acquisition unit is used to reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, its length, width and height are calculated and used as size data;

[0042] The obstacle classification unit is used to classify the clustered obstacles from the intermediate point cloud data. Each cluster corresponds to an obstacle, and its size data is used as the feature information of the obstacle.

[0043] Preferably, the step of calculating the weighted centroid of each target object using the point cloud reflection intensity as a weight includes:

[0044] The coordinates of the weighted centroid are obtained according to the following formula:

[0045]

[0046] c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z iare the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height.

[0047] Preferably, for each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes:

[0048] The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes:

[0049]

[0050]

[0051] Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

[0052] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above-described methods when the computer program is executed by at least one processor.

[0053] In a fourth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by at least one processor, it implements the steps of any of the methods described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present application is further described below with reference to the accompanying drawings and implementation methods.

[0055] Figure 1 This is a flow chart of a road pothole detection method based on lidar provided in an embodiment of the present application.

[0056] Figure 2 This is a flow chart of point cloud data separation of an obstacle provided in an embodiment of the present application.

[0057] Figure 3 This is a schematic diagram of the geometric relationship of a laser radar provided in an embodiment of the present application when detecting road potholes.

[0058] Figure 4 This is a schematic structural diagram of a road pothole detection device based on lidar provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] Below, the present application is further described in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments described below or the various technical features can be arbitrarily combined to form new embodiments. The implementation procedures of the present application will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific implementation procedures, and the various details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for the purpose of illustrating the present application, and are not intended to limit the scope of protection of the present application.

[0060] Example 1

[0061] See also Figure 1 , Figure 1 This is a flow chart of a road pothole detection method based on lidar provided in an embodiment of the present application.

[0062] The road pothole detection method provided in the embodiment of the present application includes the following steps:

[0063] S101, using ray method ground filtering, divides the lidar point cloud into rays according to polar coordinates, analyzes the height continuity point by point along each ray, segments the area below the ground, obtains the point cloud area data below the ground and uses it as the intermediate point cloud data;

[0064] After acquiring point cloud data, the LiDAR divides it into rays using polar coordinates and uses predefined radius and angle differential formulas. The system then analyzes height continuity point by point along each ray, segmenting the subsurface point cloud area using predefined ground-level determination criteria. This subsurface point cloud area serves as the intermediate point cloud data, initially screening for areas with potential potholes and reducing the amount of data required for subsequent processing.

[0065] S102, clustering the intermediate point cloud data using the Euclidean clustering method, classifying obstacles from the point cloud data, and obtaining dimension data including length, width, and height of the clustered obstacles;

[0066] A KD tree is constructed for the intermediate point cloud data. An initial point P is selected and a distance threshold is set (e.g., 0.3 meters in urban road scenarios). A proximity search is performed on the initial point P, and all point cloud points within the distance threshold are added to the set Q. The proximity search is then continued for the unsearched points in set Q. This process is repeated until all point clouds are clustered. The length, width, and height data of each obstacle are obtained, and the obstacles are classified from the point cloud data, providing clear target objects for subsequent pothole determination.

[0067] S103: The clustered obstacles are regarded as targets, and the weighted centroid of each target is calculated using the point cloud reflection intensity as a weight. Targets whose centroids are below the ground are considered to be potholes. The location data and size data of each pothole are used as pothole information.

[0068] Using the point cloud's reflection intensity as a weight, the center of mass of each clustered obstacle is calculated using the weighted center of mass formula. If the z-value of the center of mass is below the ground, the obstacle is identified as a pothole. The location and size data of each pothole are used as pothole information, allowing for accurate pothole identification and preventing misidentification.

[0069] S104, obtaining specific information of the vehicle, wherein the specific information includes the vehicle's current pitch angle data, sensor installation height, and obstacle target height; predicting the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

[0070] The system acquires specific information about the vehicle, including the current pitch angle, sensor installation height, and obstacle target height. It predicts the edge of each target pothole and obtains this information to help the autonomous driving robot plan its path, allowing it to know the location of pothole edges in advance and avoid driving risks.

[0071] Thus, ray-based ground filtering effectively separates the subsurface point cloud area, Euclidean clustering accurately identifies obstacles and obtains dimensional data, and weighted centroid calculation is used to accurately identify potholes. Integrating specific information predicts pothole edges. This entire process improves the probability of identifying potholes on the road and more accurately predicts their edges, effectively preventing autonomous driving robots from experiencing bumps or even damage due to not detecting potholes in a timely manner, thereby improving their driving safety and reliability. Ray-based ground filtering analyzes the high continuity of each ray point by point, accurately segmenting the subsurface point cloud area and reducing the probability of false detection. Using point cloud reflection intensity as a weight, weighted centroid calculation fully considers the characteristics of the point cloud data, avoiding false positives caused by purely geometric feature analysis and improving pothole detection accuracy. This embodiment eliminates camera fusion and deep learning methods, eliminating the need for expensive camera equipment and high-performance computing equipment for model training, significantly reducing hardware costs. Compared with traditional methods, it reduces the complex multi-sensor fusion processing and data calibration, significantly shortening deployment time.

[0072] In a specific application, when a LiDAR emits a laser beam and receives a signal reflected from an object's surface, it calculates the distance from that point to the LiDAR based on the laser's time of flight (ToF). Combined with the laser beam's emission angle, it calculates the point's coordinates (x, y, z) in three-dimensional space. The z value below represents the vertical height of the point relative to the LiDAR's installation height.

[0073] See also Figure 2 , Figure 2 This is a flow chart of point cloud data separation of an obstacle provided in an embodiment of the present application.

[0074] In one embodiment, the S102 includes:

[0075] S201, performing spatial search on the intermediate point cloud data by constructing a KD tree;

[0076] The KD tree is an efficient data structure that enables rapid spatial searches. A KD tree is first constructed for the intermediate point cloud data, providing the foundation for subsequent proximity searches. By segmenting the point cloud data along various dimensions, the KD tree forms a tree-like structure, enabling rapid location of point clouds within a specific area during spatial searches, significantly improving search efficiency.

[0077] S202, selecting any initial point in the point cloud and performing a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters;

[0078] A proximity search is performed with a random initial point in the intermediate point cloud data as the center. A distance threshold of 0.2 to 0.5 meters is determined based on factors such as the potential size of obstacles in the actual application scenario and the density of the point cloud data. This step can initially screen out point cloud points that are spatially close to the initial point, preparing for subsequent clustering.

[0079] S203: Continue to perform proximity search for all point cloud points in the set that have not been subjected to proximity search according to the set distance threshold, and add the newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold;

[0080] For any point cloud points in the set that haven't yet been subjected to a proximity search, the algorithm continues to perform a proximity search based on the set distance threshold, adding the newly found point cloud points to the set. This process repeats until all point cloud points in the set have completed the proximity search. This method gradually expands the clustering range, finding a cluster of points that meets the distance threshold and forming a cluster. This helps group point cloud points belonging to the same obstacle together.

[0081] S204: Reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, calculate its length, width and height and use them as size data;

[0082] Reselect unclustered points in the intermediate point cloud data as new initial points, and repeat steps S202 and S23 until all intermediate point cloud data are clustered to obtain multiple clusters. For each cluster, calculate its length, width, and height to reflect the size and shape of the obstacle.

[0083] S205 , classifying the clustered obstacles from the intermediate point cloud data, wherein each cluster corresponds to one obstacle, and the size data thereof is used as feature information of the obstacle.

[0084] The clustered obstacles are then separated from the intermediate point cloud data. Each cluster corresponds to an obstacle, and its size data is used as the feature information of the obstacle. This allows different obstacles to be distinguished and gives each obstacle a specific feature description, which facilitates further analysis and processing.

[0085] Considering that traditional search methods may require comparing distances between point clouds one by one, the construction of a KD tree significantly improves spatial search efficiency. Its hierarchical structure allows for rapid location of point clouds within a specific area, significantly reducing the time and computational resources required for the search. Through proximity search and iterative clustering, point clouds belonging to the same obstacle can be accurately clustered together. Setting a reasonable distance threshold helps identify point cloud clusters that match obstacle characteristics, thereby improving obstacle identification accuracy. Furthermore, calculating the length, width, and height of each cluster further enriches the characteristic information of the obstacle, providing more comprehensive data support for subsequent analysis.

[0086] In one embodiment, the step of calculating the weighted center of mass of each target object using the point cloud reflection intensity as a weight includes:

[0087] The coordinates of the weighted centroid are obtained according to the following formula:

[0088]

[0089] c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height.

[0090] During pothole detection, after obtaining point cloud clusters corresponding to each target object through ray-based ground filtering and Euclidean clustering, it is necessary to further determine whether these targets are potholes. At this time, the weighted centroid of each target object is calculated using the point cloud reflection intensity as the weight. In the calculation formula, c represents the coordinates of the weighted centroid, which consists of the coordinate values ​​in the x, y, and z directions; w i is the reflection intensity information of the i-th point cloud point, which can be directly obtained through the laser radar, reflecting the reflection ability of the point cloud point to the laser, and is related to the material and surface characteristics of the point cloud point. i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, indicating the specific position of the point cloud point in space; n is the total number of point cloud points contained in the target object, representing the size of the point cloud cluster corresponding to the target object.

[0091] This approach fully accounts for the influence of the reflective properties of the point cloud points on the centroid position. Compared to simple geometric centroid calculation, the weighted centroid better reflects the true location characteristics of the target object, especially when distinguishing obstacles with different materials or surface characteristics, and has higher accuracy. When the calculated centroid z-value is below the ground, the target object is considered a pothole.

[0092] Therefore, by incorporating point cloud reflection intensity as a weight, weighted centroid calculation can more accurately locate the center of the target object. Because objects of different materials and surface characteristics reflect laser light differently, potholes often have different reflective properties than the surrounding normal road surface. Therefore, weighted centroid calculation can better reflect the true location and characteristics of the target object. Compared to judgment based solely on geometric shape, it can more accurately distinguish potholes from other obstacles, reducing the risk of false positives.

[0093] In one embodiment, for each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes:

[0094] The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes:

[0095]

[0096]

[0097] Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

[0098] In specific applications, the vehicle's current pitch angle can be obtained through vehicle positioning information. During the driving process, the vehicle may produce changes in pitch posture due to factors such as road undulations, which will affect the accurate judgment of the pothole edge position. In each cluster (i.e., target object), the three-dimensional coordinates of the point closest to the vehicle body are found. This is considered to be the first position that may affect the vehicle's driving and is used to predict the pothole edge. In each cluster, the z coordinate value of the point closest to the vehicle reflects the position relationship of the pothole relative to the vehicle in the vertical direction. Combining the vehicle's pitch angle α and the x and y coordinates of the nearest point, the vehicle's tilt height is calculated to reflect the impact of vehicle posture changes on the prediction of the pothole edge position.

[0099] Based on the parameters obtained above, substitute them into the formula to calculate the coordinates of the pothole edge (xedg, yedg): This formula can be considered to comprehensively consider the vehicle's own state (pitch angle and tilt height) and the position information of the pothole obstacle, and calculate the coordinates of the pothole edge in the vehicle coordinate system through geometric relationships. Here, hsensor is the sensor installation height. The calculated pothole edge coordinates (xedg, yedg) are the predicted pothole edge position information. This is provided to the control system of the autonomous vehicle so that it can plan the driving path in advance, avoid pothole areas, and ensure driving safety.

[0100] This approach takes into account factors such as the vehicle's pitch angle and tilt height, ensuring that pothole edge predictions are adaptable to varying vehicle postures under varying road conditions. Even when traveling uphill or downhill, or on uneven roads, pothole edge locations can be accurately predicted, avoiding misjudgments or missed detections due to changes in vehicle posture, thus improving the detection method's reliability and adaptability.

[0101] As an example, a road pothole detection method based on laser radar is provided, including steps S101 to S104. The method includes:

[0102] S101, using ray method ground filtering, divides the lidar point cloud into rays according to polar coordinates, analyzes the height continuity point by point along each ray, segments the area below the ground, obtains the point cloud area data below the ground and uses it as the intermediate point cloud data;

[0103] In specific applications, ray division is performed based on the radius differential formula and the angle differential formula. The radius differential formula for ray division is:

[0104]

[0105] The angle differential formula for ray division in this method is:

[0106]

[0107] where d i 、a i To set the threshold, the threshold is determined by a large number of experiments and data analysis based on different road scenes and lidar parameters. As an example, in the urban road scene, after many tests, it was found that d i Take 0.5, a i When the value is 5, the rays can be effectively divided, balancing detection accuracy and efficiency.

[0108] The judgment conditions below the ground are:

[0109] z≤-h sensor +tanΔ r

[0110] where h sensor is the installation height of the laser radar, Δ r is the distance between two adjacent points on a ray. If the z value of a point cloud satisfies the above formula, the point is considered to be below ground level. This method allows for point-by-point analysis of height continuity along each ray, accurately segmenting the subsurface point cloud area while reducing computational effort and the probability of false detection. The z value is the height of each point in the 3D spatial coordinates acquired by the LiDAR, reflecting the vertical position of the laser pulse emitted by the LiDAR.

[0111] S102, clustering the intermediate point cloud data using the Euclidean clustering method, classifying obstacles from the point cloud data, and obtaining dimension data including length, width, and height of the clustered obstacles;

[0112] S201, performing spatial search on the intermediate point cloud data by constructing a KD tree;

[0113] S202, selecting any initial point in the point cloud and performing a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters;

[0114] S203: Continue to perform proximity search for all point cloud points in the set that have not been subjected to proximity search according to the set distance threshold, and add the newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold;

[0115] S204: Reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, calculate its length, width and height and use them as size data;

[0116] S205 , classifying the clustered obstacles from the intermediate point cloud data, wherein each cluster corresponds to one obstacle, and the size data thereof is used as feature information of the obstacle.

[0117] S103: The clustered obstacles are regarded as targets, and the weighted centroid of each target is calculated using the point cloud reflection intensity as a weight. Targets whose centroids are below the ground are considered to be potholes. The location data and size data of each pothole are used as pothole information.

[0118] The calculation of the weighted center of mass of each target object using the point cloud reflection intensity as the weight includes:

[0119] The coordinates of the weighted centroid are obtained according to the following formula:

[0120]

[0121] c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height. Through this formula, the coordinates of each point cloud point will be reflected by its reflection intensity w i The weighted center of mass c of the entire object is calculated. If the z coordinate of this center of mass is below ground level (i.e., z ≤ ground level), the object is considered a pothole. Point cloud reflection intensity information can be directly obtained through LiDAR. If the z value of the weighted center of mass is below ground level, the obstacle is considered a pothole. This method can accurately identify potholes and avoid misidentifying other obstacles as potholes.

[0122] S104, obtaining specific information of the vehicle, wherein the specific information includes the vehicle's current pitch angle data, sensor installation height, and obstacle target height; predicting the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

[0123] For each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes:

[0124] The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes:

[0125]

[0126]

[0127] Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

[0128] This formula can be used to predict the edge of a pothole, providing an important basis for autonomous driving robots to plan their driving paths.

[0129] Considering that in the field of autonomous driving, traditional road pothole detection methods mostly rely on LiDAR and camera fusion or deep learning-based solutions, but the former has complex calibration and fusion processing problems, while the latter faces the dilemma of large data labeling and computing resource requirements. In other words, related technologies are difficult to detect road potholes efficiently and accurately, and are costly and time-consuming to deploy. This application adopts ray method ground filtering, performs ray segmentation based on the polar coordinate characteristics of the LiDAR point cloud, and analyzes the height continuity point by point along each ray. It can effectively separate the point cloud area below the ground without the need for complex multi-sensor fusion processing, reducing the workload and error of data calibration and improving data processing efficiency. Specifically, by setting reasonable radius differential formulas and angle differential formulas, the LiDAR point cloud is ray-partitioned, and the point cloud area below the ground is accurately segmented according to the below-ground judgment conditions. Taking the urban road scene as an example, set di = 0.5, ai = 5, the LiDAR installation height hsensor = 1.5 meters, the z value of a point cloud is -1.2 meters, and if other conditions are met, the point is determined to be the ground and below the ground. This method based on polar coordinate characteristics and high continuity analysis makes full use of the characteristics of lidar point cloud data itself, avoids the complex process of multi-sensor fusion, directly extracts useful information from point cloud data, and reduces data processing links.

[0130] Considering that simple clustering methods cannot effectively distinguish obstacles, it is easy to misjudge multiple adjacent obstacles as a whole, or split an obstacle into multiple parts, affecting the subsequent extraction and analysis of obstacle features; at the same time, if the spatial relationship between point clouds is not reasonably utilized, it is easy to miss smaller obstacles or mistakenly regard noise point clouds as obstacles, reducing the accuracy of detection. This application uses the Euclidean clustering method to cluster the intermediate point cloud data. First, a KD tree is constructed, and the initial point P is selected. A neighbor search is performed with a set distance threshold (such as 0.3 meters in urban road scenarios). Points that meet the distance conditions are obtained and added to the set Q, and it is continuously iterated until the clustering of all point clouds is completed to obtain the length, width, height and other size data of each obstacle. In this way, the distance relationship between point clouds can be used to accurately separate obstacles from point cloud data.

[0131] Considering that traditional methods only rely on simple geometric features when determining whether an obstacle is a pothole, which can easily lead to misjudgment, this application uses the point cloud reflection intensity as the weight to calculate the weighted centroid of each target object, fully considering the characteristics of point cloud data. It can more accurately determine whether an obstacle is a pothole, avoiding misjudgment caused by simple geometric feature judgment.

[0132] See also Figure 3 , Figure 3 This is a schematic diagram of the geometric relationship of a LiDAR used in pothole detection, as provided in an embodiment of the present application. The diagram illustrates the LiDAR installation position, how the laser beam emitted by the LiDAR interacts with the ground and potholes, and the geometric relationship between the LiDAR installation height (lidar_height), the ground, and the potholes. As another example, a pothole detection method includes the following steps:

[0133] S1, ray method ground filtering implementation steps: the program divides the laser radar input point cloud into rays according to polar coordinates, and uses the radius differential formula and angle differential formula Perform ray division on the LiDAR point cloud. Assume that in a certain scene, set d i =0.5, a i =5, the program analyzes the height continuity point by point, and determines the condition z≤-h according to the ground below sensor +tanΔ r , get the part of point cloud data below the ground. If the laser radar is installed at a height of h sensor =1.5 meters, and the z value of a point cloud is -1.2 meters, then the point is determined to be the ground and the part below the ground.

[0134] S2, Euclidean clustering implementation steps: Construct a KD tree for the non-ground point cloud data after ground filtering and select an initial point P. For example, at a certain moment, the coordinates of the initial point P are (10, 5, 2). A proximity search is performed using a set distance threshold (such as 0.3 meters in urban road scenarios). Points that meet the distance threshold are added to set Q. The search continues for unsearched points in set Q, and the process continues iterating until all point clouds are clustered and the length, width, and height information of each obstacle are obtained.

[0135] S3, weighted centroid calculation and pothole judgment implementation steps: According to the weighted centroid calculation formula The center of mass of each obstacle is calculated using the point cloud reflection intensity wi as a weight. If the z-value of the center of mass is below the ground, the obstacle is considered a pothole. For example, suppose an obstacle consists of five points, their reflection intensities and coordinates are known, and the calculated z-value of the center of mass is -1.8 meters, which is below the ground level and therefore a pothole.

[0136] S4, fusion positioning prediction pothole edge implementation steps: the program fuses the vehicle positioning information to obtain the vehicle pitch angle α, combined with the sensor installation height h sensor , the obstacle target height, based on the x-value formula of the pothole edge:

[0137] And the formula for the y value of the pothole edge: Predict the edge of the pothole and publish the predicted position. Assume that the coordinate point closest to the vehicle body is (3, 4, -1.8), h sensor The value is 1.5, α is 1 degree, so The value is 0.876. According to the formula, the coordinates of the edge of the pothole are predicted: x=3*0.876=2.628, y=4*0.876=3.504; therefore, the coordinates of the edge of the pothole are (2.628, 3.504, -1.8).

[0138] This application uses ray-based ground filtering, performing ray segmentation based on the polar coordinate characteristics of the lidar point cloud. Through high-continuity analysis, it can effectively separate the point cloud area below the ground. Compared to traditional methods, this method eliminates the need for complex multi-sensor fusion processing, reduces the workload and error of data calibration, and improves data processing efficiency. The Euclidean clustering method uses the distance relationship between point clouds to cluster, accurately identifying obstacles and providing clear target objects for subsequent pothole detection. The weighted centroid is calculated using the point cloud reflection intensity as a weight, fully considering the characteristics of the point cloud data. It can more accurately determine whether an obstacle is a pothole, avoiding misjudgments caused by simple geometric feature judgment. The integration of specific information to predict the edge of the pothole and, combined with the vehicle's own state information, improves the accuracy of pothole location prediction, providing reliable data support for autonomous driving robots to plan safe driving paths in advance. The overall technical solution closely coordinates all links from data processing to pothole detection to edge prediction, effectively solving the problems of long deployment time, high cost, and large computing resource consumption in related technologies.

[0139] Example 2

[0140] See also Figure 4 , Figure 4 This is a schematic structural diagram of a road pothole detection device based on lidar provided in an embodiment of the present application.

[0141] This embodiment provides a road pothole detection device based on laser radar. The specific embodiments and technical effects achieved by the embodiments are consistent with those described in the above method embodiments, and some details are not repeated here. The road pothole detection device includes:

[0142] The ground point cloud acquisition module is used to use ray method ground filtering to divide the lidar point cloud into rays according to polar coordinates, analyze the height continuity point by point along each ray, segment the area below the ground, and obtain the point cloud area data below the ground as the intermediate point cloud data;

[0143] The data clustering module is used to cluster the intermediate point cloud data using the Euclidean clustering method, classify obstacles from the point cloud data, and obtain the dimension data of the clustered obstacles including length, width and height;

[0144] The pothole detection module is used to treat clustered obstacles as targets, calculate the weighted centroid of each target using the point cloud reflection intensity as the weight, and identify targets with the centroid below the ground as potholes. The location and size data of each pothole are used as pothole information.

[0145] The edge acquisition module is used to obtain specific information about the vehicle, including the vehicle's current pitch angle data, sensor installation height, and obstacle target height; and predict the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

[0146] In one embodiment, the data clustering module includes:

[0147] The spatial search unit is used to perform spatial search on the intermediate point cloud data by constructing a KD tree;

[0148] The first set acquisition unit is used to select any initial point in the point cloud and perform a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters;

[0149] The second set acquisition unit is configured to continue to perform proximity search according to the set distance threshold for all point cloud points in the set that have not been subjected to proximity search, and add newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold;

[0150] A cluster acquisition unit is used to reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, its length, width and height are calculated and used as size data;

[0151] The obstacle classification unit is used to classify the clustered obstacles from the intermediate point cloud data. Each cluster corresponds to an obstacle, and its size data is used as the feature information of the obstacle.

[0152] In one embodiment, the step of calculating the weighted center of mass of each target object using the point cloud reflection intensity as a weight includes:

[0153] The coordinates of the weighted centroid are obtained according to the following formula:

[0154]

[0155] c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height.

[0156] In one embodiment, for each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes:

[0157] The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes:

[0158]

[0159]

[0160] Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

[0161] Example 3

[0162] This embodiment provides a computer-readable storage medium, the specific embodiments of which are consistent with the embodiments and technical effects described in the above method embodiments, and some contents are not repeated here. The computer program is stored on the computer-readable storage medium, and when the computer program is executed by at least one processor, the steps of any of the above methods are implemented.

[0163] Example 4

[0164] This embodiment provides a computer program product, the specific embodiments of which are consistent with the embodiments and technical effects described in the above method embodiments, and some contents are not repeated here. The computer program product includes a computer program that, when executed by at least one processor, implements the steps of any of the above methods.

[0165] It should be noted that, in the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can represent: a, b, c, a and b, a and c, b and c or a and b and c, where a, b and c can be single or multiple. It is worth noting that "at least one" can also be interpreted as "one or more items".

[0166] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are configured to distinguish similar objects and are not necessarily configured to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0167] This application is explained from the perspectives of purpose of use, effectiveness, progress and novelty, and has complied with the functional enhancement and use requirements emphasized by the Patent Law. The above description and drawings of this application are only preferred embodiments of this application and are not intended to limit this application. Therefore, all structures, devices, features, etc. that are similar or identical to those of this application, that is, all equivalent replacements or modifications made in accordance with the scope of the patent application of this application, should fall within the scope of protection of the patent application of this application.

Claims

1. A road pothole detection method based on laser radar, characterized in that: The following steps are involved: S101, using ray method ground filtering, divides the lidar point cloud into rays according to polar coordinates, analyzes the height continuity point by point along each ray, segments the area below the ground, obtains the point cloud area data below the ground and uses it as the intermediate point cloud data; S102, clustering the intermediate point cloud data using the Euclidean clustering method, classifying obstacles from the point cloud data, and obtaining dimension data including length, width, and height of the clustered obstacles; S103: The clustered obstacles are regarded as targets, and the weighted centroid of each target is calculated using the point cloud reflection intensity as a weight. Targets whose centroids are below the ground are considered to be potholes. The location data and size data of each pothole are used as pothole information. S104, obtaining specific information of the vehicle, wherein the specific information includes the vehicle's current pitch angle data, sensor installation height, and obstacle target height; predicting the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

2. The road pothole detection method according to claim 1, characterized in that: The S102 includes: S201, performing spatial search on the intermediate point cloud data by constructing a KD tree; S202, selecting any initial point in the point cloud and performing a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters; S203: Continue to perform proximity search for all point cloud points in the set that have not been subjected to proximity search according to the set distance threshold, and add the newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold; S204: Reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, calculate its length, width and height and use them as size data; S205 , classifying the clustered obstacles from the intermediate point cloud data, wherein each cluster corresponds to one obstacle, and the size data thereof is used as feature information of the obstacle.

3. The road pothole detection method according to claim 1, characterized in that: The calculation of the weighted center of mass of each target object using the point cloud reflection intensity as the weight includes: The coordinates of the weighted centroid are obtained according to the following formula: c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height.

4. The road pothole detection method according to claim 1, characterized in that: For each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes: The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes: Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

5. A road pothole detection device based on laser radar, characterized in that: include: The ground point cloud acquisition module is used to use ray method ground filtering to divide the lidar point cloud into rays according to polar coordinates, analyze the height continuity point by point along each ray, segment the area below the ground, and obtain the point cloud area data below the ground as the intermediate point cloud data; The data clustering module is used to cluster the intermediate point cloud data using the Euclidean clustering method, classify obstacles from the point cloud data, and obtain the dimension data of the clustered obstacles including length, width and height; The pothole detection module is used to treat clustered obstacles as targets, calculate the weighted centroid of each target using the point cloud reflection intensity as the weight, and identify targets with the centroid below the ground as potholes. The location and size data of each pothole are used as pothole information. The edge acquisition module is used to obtain specific information about the vehicle, including the vehicle's current pitch angle data, sensor installation height, and obstacle target height; and predict the edge of each road pothole based on the specific information and pothole information to obtain pothole edge information.

6. The road pothole detection device according to claim 5, characterized in that: The data clustering module includes: The spatial search unit is used to perform spatial search on the intermediate point cloud data by constructing a KD tree; The first set acquisition unit is used to select any initial point in the point cloud and perform a proximity search with the initial point as the center to obtain all point cloud points within a set distance threshold and add them to the set; the set distance threshold is set to a value of 0.2 meters to 0.5 meters; The second set acquisition unit is configured to continue to perform proximity search according to the set distance threshold for all point cloud points in the set that have not been subjected to proximity search, and add newly found point cloud points to the set until the proximity search for all point cloud points in the set is completed, thereby obtaining a cluster of points that meet the set distance threshold; A cluster acquisition unit is used to reselect the initial points that have not been clustered in the intermediate point cloud data, and repeat S202 and S203 until all the intermediate point cloud data are clustered to obtain multiple clusters; for each cluster, its length, width and height are calculated and used as size data; The obstacle classification unit is used to classify the clustered obstacles from the intermediate point cloud data. Each cluster corresponds to an obstacle, and its size data is used as the feature information of the obstacle.

7. The road pothole detection device according to claim 5, characterized in that: The calculation of the weighted center of mass of each target object using the point cloud reflection intensity as the weight includes: The coordinates of the weighted centroid are obtained according to the following formula: c represents the coordinates of the weighted centroid, w i is the point cloud reflection intensity information, x i 、y i 、z i are the coordinates of the i-th point cloud point in three-dimensional space, n is the total number of point cloud points contained in the target object, and hsensor is the sensor installation height.

8. The road pothole detection device according to claim 5, characterized in that: For each target object, the step of predicting and obtaining pothole edge information based on the specific information and pothole information includes: The pothole edge is predicted based on the pothole edge prediction formula. The pothole edge prediction formula includes: Among them, (xedg, yedg) is the coordinate of the edge of the pothole, α is the pitch angle of the vehicle, and (xi, yi, zi) is the coordinate of the point closest to the vehicle body in each cluster. is the z value of the point closest to the vehicle in the pothole obstacle, The vehicle's tilt height is calculated using the vehicle's up and down tilt angles and the x and y coordinates of the nearest point.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by at least one processor, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by at least one processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images