A road intelligent inspection method with positioning function

Abstract

The present invention relates to the field of measurement and testing technology, and specifically to an intelligent road inspection method with a positioning function. The method first obtains a road image and three-dimensional point cloud data; clusters the three-dimensional point cloud data to obtain a noise cluster; calculates the probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area in the road surface area; calculates the probability of different bumps during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bump probability; based on the bump probability, motion compensation is performed on the three-dimensional point cloud data of the next frame of the road image; the three-dimensional point cloud data after motion compensation is classified to obtain a disease cluster and corresponding disease parameters. The present invention analyzes the degree of bumping of the monitoring vehicle during the inspection process and performs motion compensation in the vertical direction, thereby making the detection results of road surface diseases more accurate.

Description

Technical Field

[0001] The present invention relates to the field of measurement and testing technology, and in particular to an intelligent road inspection method with a positioning function. Background Art

[0002] Intelligent road inspection is a necessary means to detect the current road disease conditions, and repeated operations are required when the road is long and wide. At present, common intelligent road inspections are to determine road diseases by combining laser radar point cloud data with graphic sensor data. Since the laser radar is in motion during the data collection process, the speed of the monitoring vehicle and the different degrees of bumps on the road surface are prone to motion distortion. In particular, rural roads themselves are not smooth enough, and there are many stones, gravel, and small potholes. The point cloud data collected by the vehicle-mounted laser radar deviates from the actual roadblocks and disease areas, resulting in a serious decrease in the accuracy of the final detection results. The existing vehicle-mounted laser radar point cloud data motion compensation algorithms all default to ignoring the vertical movement of the vehicle body on relatively flat ground, and mainly consider the horizontal motion compensation. However, the monitoring vehicle has different motion distortions in the vertical direction under different degrees of bumps. Considering only the horizontal motion compensation will result in poor motion compensation, which will affect the accuracy of road disease inspection. Summary of the invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a road intelligent inspection method with positioning function, and the technical solution adopted is as follows:

[0004] Acquire road images, and use the vehicle-mounted laser radar to acquire three-dimensional point cloud data collected when collecting road images;

[0005] Clustering the three-dimensional point cloud data to obtain noise clusters; calculating the bumping influence probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area in the road surface area; calculating the bumping probability of different bumps occurring during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bumping influence probability;

[0006] Based on the bump probability, three-dimensional point cloud data of historical frames similar to the three-dimensional point cloud data of the current frame are obtained, the similar three-dimensional point cloud data of the historical frames are input into a Gaussian mixture model, and the Gaussian mixture model is used to perform motion compensation on the three-dimensional point cloud data of the next frame of road image;

[0007] The motion-compensated three-dimensional point cloud data is classified to obtain a defect cluster; and the corresponding defect parameters are obtained by combining the three-dimensional point cloud data in the defect cluster with the corresponding actual road image.

[0008] Preferably, the calculation formula for the probability of turbulence impact of different turbulence occurrences is:

[0009]

[0010] in, is the probability of the turbulence affecting the 3D point cloud data in the noise cluster when a turbulence occurs at the same moment; is the probability of the turbulence of the three-dimensional point cloud data in the noise cluster affecting the secondary turbulence occurring at the same time; is the probability of the three-dimensional point cloud data in the noise cluster being affected by three turbulences occurring at the same time; is the probability of the turbulence effect of the three-dimensional point cloud data in the noise cluster on four turbulences occurring at the same time; is the number of 3D point cloud data belonging to the noise cluster in the zth sub-area in the road surface area of ​​the current frame; The total number of 3D point cloud data that can be accommodated in the road area corresponding to the monitoring vehicle in the current frame; is the number of 3D point cloud data belonging to the noise cluster in the dth sub-area in the road surface area of ​​the current frame; is the number of 3D point cloud data belonging to the noise cluster in the fth sub-area in the road surface area of ​​the current frame; is the number of pairwise combinations in the four sub-areas; The number of combinations of three sub-regions selected from four sub-regions; , , and are the number of 3D point cloud data belonging to the noise cluster in the four sub-regions.

[0011] Preferably, the calculation formula of the bump probability is:

[0012]

[0013] in, To monitor the probability of X bumps occurring when the vehicle passes through a single road area completely; To monitor the probability of X bumps occurring when the vehicle passes through a single road area. is the binomial distribution formula about (n, X); K is the number of any bumps that occur per second; X is the number of times a certain type of event occurs X times in n seconds; n is the time when the monitoring vehicle passes the actual road image corresponding to the 3D point cloud data.

[0014] Preferably, clustering the three-dimensional point cloud data to obtain noise clusters includes:

[0015] The three-dimensional point cloud data is clustered based on density to obtain multiple clusters; the height mean of each cluster is calculated, and the clusters with a height mean greater than a calibrated road surface height are classified into a noise cluster; the three-dimensional point cloud data includes height.

[0016] Preferably, the acquiring of historical frame three-dimensional point cloud data similar to the current frame three-dimensional point cloud data based on the bump probability includes:

[0017] Calculate the similarity of the bump probabilities corresponding to the current frame and the historical frame; take the historical frame whose similarity is greater than the preset similarity threshold as the historical frame similar to the current frame, and obtain the corresponding historical frame 3D point cloud data similar to the current frame 3D point cloud data.

[0018] Preferably, the step of classifying the motion-compensated three-dimensional point cloud data to obtain the defect clusters includes:

[0019] For the motion-compensated three-dimensional point cloud data, the three-dimensional point cloud data is clustered based on density to obtain multiple clusters; the height mean of each cluster is calculated, and the clusters with a height mean less than the calibrated height of the road surface are classified as disease clusters.

[0020] Preferably, the calculation formula of the similarity is:

[0021]

[0022] in, is the similarity between the current frame 3D point cloud data and the historical frame 3D point cloud data; is the number of historical frames; The probability of X bumps occurring when the monitoring vehicle corresponding to the current frame passes through a single road area completely. To monitor the probability of X bumps occurring when the vehicle passes through a single road area.

[0023] Preferably, the step of obtaining corresponding disease parameters by combining the three-dimensional point cloud data in the disease cluster with the corresponding actual road image includes:

[0024] The disease parameters include the area size of the disease and the volume of the depression; the three-dimensional point cloud data in the disease cluster is projected onto the corresponding actual road image, and the number of pixels corresponding to the three-dimensional point cloud data in the disease cluster on the actual road image is the area size of the disease; the area size of the disease is multiplied by the absolute value of the minimum height of the pixel points corresponding to the three-dimensional point cloud data in the disease cluster to obtain the depression volume.

[0025] The embodiments of the present invention have at least the following beneficial effects:

[0026] The present invention clusters the three-dimensional point cloud data collected by the vehicle-mounted laser radar to obtain noise clusters; calculates the bump impact probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area in the road surface area; calculates the bump probability of different bumps occurring during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bump impact probability; analyzes the bump degree of the monitoring vehicle during the inspection process, and the bump probability reflects the abnormality in the vertical direction, and subsequently performs vertical motion compensation according to the calculated bump probability. It combines the bump probability to obtain the historical frame three-dimensional point cloud data similar to the current frame three-dimensional point cloud data, inputs the similar historical frame three-dimensional point cloud data into the Gaussian mixture model, and uses the Gaussian mixture model to perform motion compensation on the three-dimensional point cloud data of the next frame road image, thereby realizing compensation in the vertical direction under the bumpy road, and avoiding the impact of the bumpy road on the accuracy of the detection of the diseased area; classifying the three-dimensional point cloud data after motion compensation to obtain the disease cluster; combining the three-dimensional point cloud data in the disease cluster with the corresponding actual road image to obtain the corresponding disease parameters, and finally obtaining the disease cluster and disease parameters according to the three-dimensional point cloud data after motion compensation, thereby realizing the disease inspection of the road. The present invention takes into account the degree of bump in the vertical direction, and has a good compensation effect for the motion compensation of the three-dimensional point cloud data of the road environment and the road with poor flatness, thereby making the detection result of the road disease more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or 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. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 A method flow chart of an intelligent road inspection method with positioning function provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of a road intelligent inspection method with positioning function proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0030] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0031] The embodiment of the present invention provides a specific implementation method of a road intelligent inspection method with a positioning function, and the method is applicable to the scene of road intelligent inspection. In this scene, intelligent road inspection is carried out by a monitoring vehicle, and the monitoring vehicle is equipped with a camera and a vehicle-mounted laser radar. The camera is fixed under the monitoring vehicle, and the road image collected by it is mainly used for positioning. In this scene, the road surface disease is determined by combining the laser radar point cloud data with the graphic sensor data. However, since the laser radar is in motion during the data collection process, the speed of the monitoring vehicle and the different degrees of bumps on the road surface are prone to motion distortion, especially the rural roads themselves have poor flatness, and there are generally many stones, gravel, and small potholes. Therefore, in order to solve the problem that the existing vehicle-mounted laser radar point cloud data motion compensation algorithm is defaulted to being able to approximately ignore the vertical movement of the vehicle body on a relatively flat ground, and mainly considers the problem of horizontal motion compensation. The present invention analyzes the degree of bumps of the monitoring vehicle during the inspection process, and performs motion compensation in the vertical direction. It has a good compensation effect for the motion compensation of the point cloud data of the road environment and the road with poor flatness. Since the influence in the vertical direction is considered, the detection result of the road surface disease will be more accurate.

[0032] The specific scheme of the intelligent road inspection method with positioning function provided by the present invention is described in detail below with reference to the accompanying drawings.

[0033] See also Figure 1 , which shows a method flow chart of a road intelligent inspection method with positioning function provided by an embodiment of the present invention, the method comprising the following steps:

[0034] Step S100, acquiring a road image, and using a vehicle-mounted laser radar to acquire three-dimensional point cloud data collected when acquiring the road image.

[0035] The camera on the monitoring vehicle is used to obtain road images, which are used for positioning. The vehicle-mounted laser radar is used to obtain the three-dimensional point cloud data in front of the monitoring vehicle when collecting road images. Laser radar, or laser detection and ranging, consists of a laser transmitter, antenna, receiver, tracking frame, and information processing. It detects objects by transmitting, reflecting, and receiving infrared beams. When the laser beam encounters an object, it is diffusely reflected and returns to the laser receiver. The radar module multiplies the time interval between sending and receiving signals by the speed of light and divides it by 2 to calculate the distance between the transmitter and the object. The road inspection robot can use its array ground-penetrating radar and camera module to detect road defects and obtain three-dimensional point cloud data of the road at different GPS coordinates. The three-dimensional point cloud data includes coordinate points, road distance, color information, and height. Among them, the road distance is the distance between the transmitter and the object.

[0036] Step S200, clustering the three-dimensional point cloud data to obtain noise clusters; calculating the bump impact probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area in the road surface area; calculating the bump impact probability of different bumps occurring during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bump impact probability.

[0037] When the laser beam encounters an obstacle or target object, the lidar will receive the information reflected back by the obstacle or target object. Since roads, especially rural roads, are not smooth enough with stones, gravel, mud, small potholes, etc., the data collection by the monitoring vehicle in motion will be affected by frequent bumps. The point cloud data array formed by the lidar will be offset, resulting in motion distortion, which will lead to distortion of the fitting results of the subsequent point cloud data in the diseased area.

[0038] The existing motion compensation algorithms for 3D point cloud data based on vehicle-mounted LiDAR assume that the vertical motion of the vehicle can be approximately ignored on relatively flat ground, and mainly consider the horizontal motion compensation. However, in different scenarios, such as when the monitoring vehicle is driving on rural roads, the frequency of vertical bumps is very high, and the monitoring vehicle is mainly used to detect road cracks, potholes and other depressions. Therefore, the point cloud data offset caused by the vertical motion of the vehicle cannot be directly ignored.

[0039] The laser radar transmitter is at the head of the monitoring vehicle, and the 3D point cloud data it collects is not synchronized with the road image collected by the camera at the same time. The road image collected by the camera is mainly used for positioning, because the laser radar collects the 3D point cloud data of the road section in front of the monitoring vehicle, and the road image is the image of the monitoring vehicle, so it is necessary to match the 3D point cloud data with its corresponding road image. Therefore, the position of the vehicle body is not synchronized with the coordinate position of the 3D point cloud data collected by the laser radar. If there are many stones, potholes, etc. at the current location of the vehicle body, the monitoring vehicle will be bumpy during dynamic driving due to the light body of the intelligent monitoring vehicle, and the point cloud data collected in the current frame will be offset.

[0040] Furthermore, the actual road image corresponding to the currently collected three-dimensional point cloud data is obtained. When the monitoring vehicle collects the three-dimensional point cloud data corresponding to the next frame of road image, it will travel on the road surface that has been collected in the previous frame, that is, the three-dimensional point cloud data currently collected corresponds to the data of the road surface collected in the next frame. Assuming that the monitoring vehicle collects point cloud data of the road surface in front of it in a certain frame, when the monitoring vehicle completely drives forward to the area of ​​the previously collected point cloud data and is about to collect the next frame of point cloud data, the road condition information of the static position where the monitoring vehicle stops can be obtained from the road surface point cloud data collected at the beginning. Therefore, according to the difference in acquisition time between the three-dimensional point cloud data and the road image, the time difference between the three-dimensional point cloud data and the actual corresponding actual road image is calculated according to the length of the monitoring vehicle and its driving speed.

[0041]

[0042] in, Represents the time corresponding to the i-th frame of 3D point cloud data; represents the length of the monitoring vehicle; v represents the speed of the monitoring vehicle; It represents the time taken by the monitoring vehicle to travel the L length of road; Represents the time of the ith frame of 3D point cloud data minus the time taken by the monitoring vehicle to travel L length of road; is the time corresponding to the i-1th frame of 3D point cloud data.

[0043] Among them, the time of the i-th frame point cloud data minus the time taken by the monitoring vehicle to travel L length of road , as the time point of the i-th frame point cloud data to compensate the next frame. Therefore, when predicting the motion compensation of the next frame of 3D point cloud data, the reference is the distance before the next frame. Frame of 3D point cloud data. Among them, the length and driving speed of the monitoring vehicle are set by the implementer according to the actual situation. The longer the length of the vehicle, the longer the time difference between the two frames, and the greater the time difference between the corresponding i-1 frame 3D point cloud data and the i-th frame 3D point cloud data; conversely, the shorter the length of the vehicle, the shorter the time difference between the two frames, and the smaller the time difference between the corresponding i-1 frame 3D point cloud data and the i-th frame 3D point cloud data; the greater the speed, the shorter the time difference between the two frames, and the smaller the time difference between the corresponding i-1 frame 3D point cloud data and the i-th frame 3D point cloud data; conversely, the smaller the speed, the longer the time difference between the two frames, and the greater the time difference between the corresponding i-1 frame 3D point cloud data and the i-th frame 3D point cloud data; the time corresponding to the i-th frame 3D point cloud data is also the time corresponding to the i-th frame road image.

[0044] By matching the i-1th frame of 3D point cloud data with the road image captured by the i-th frame camera, the actual 3D point cloud data corresponding to the current frame road image can be obtained, that is, the actual road image corresponding to the currently captured 3D point cloud data can be obtained, and the actual road image is captured one frame later than the corresponding 3D point cloud data. It can also be said that every time the vehicle drives to a road surface that has not been captured by the previous road image, a frame of capture is performed, and the i-1th frame of 3D point cloud data is matched with the road image captured by the i-th frame camera. Among them, the road image is in the same range as when the 3D point cloud data is obtained. The road condition data at the current position of the monitoring vehicle, that is, the 3D point cloud data can be obtained from the previously captured 3D point cloud data, and the current frame corresponding to the 3D point cloud data is obtained according to the length and driving speed of the monitoring vehicle.

[0045] Furthermore, the bump probability corresponding to the 3D point cloud data is calculated, specifically:

[0046] Step 1: Cluster the 3D point cloud data to obtain noise clusters.

[0047] The three-dimensional point cloud data is clustered based on density to obtain multiple clusters; the height mean of each cluster is calculated, and the clusters with a height mean greater than the calibrated height of the road surface are classified into the noise cluster, and the clusters with a height mean less than the calibrated height of the road surface are classified into the defect cluster, and the clusters with a height mean equal to the calibrated height of the road surface are no longer subject to subsequent calculation; the three-dimensional point cloud data includes height. Among them, the data in the noise cluster are the point cloud data clusters of stones, bricks, etc. that are higher than the calibrated height of the road surface; the data in the defect cluster are the point cloud data clusters of cracks, depressions, etc. that are lower than the calibrated height of the road surface.

[0048] Step 2: Calculate the impact probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area in the road surface area.

[0049] Due to the different distribution of noise point cloud data under the location of the monitoring vehicle, the bumping effect on the vehicle body is different. Therefore, the existing method of using all historical frame data to predict the next frame has a large error. The present invention achieves accurate fitting of the motion compensation of the predicted frame point cloud data by slicing the historical frame data. It should be noted that even if the diseased area may cause bumps to the monitoring vehicle, the present invention only considers the bumping effect of noise data on the monitoring vehicle, and then slices are selected from the historical data.

[0050] Furthermore, the probability of different times of turbulence impact formed by the three-dimensional point cloud data in the noise cluster is calculated:

[0051] Since the monitoring vehicle is bumped due to foreign objects such as stones and bricks on the road surface cushioned by the tires, such foreign objects are uniformly referred to as three-dimensional point cloud data in the noise cluster in the three-dimensional point cloud data collected by the present invention. The minimum circumscribed rectangle corresponding to the monitoring vehicle is taken as the road section where the monitoring vehicle is located, which is called the road surface area corresponding to the monitoring vehicle. The road surface area is divided into four areas, namely the upper left corner area, the lower left corner area, the upper right corner area and the lower right corner area. These four areas are collectively referred to as sub-areas. The noise point cloud data in each sub-area only has a bumpy effect on the tires in that area.

[0052] The monitoring vehicle has four tires, and each tire is bumped at most once at the same time. Therefore, one, two, three, or four tires may be bumped at the same time, resulting in 1, 2, 3, or 4 bumps. The probability of the four situations occurring is calculated. That is, the bump impact probability of different bumps is calculated based on the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area of ​​the road surface area.

[0053] The calculation formula for the probability of turbulence impact with different turbulence occurrences is:

[0054]

[0055] in, is the probability of the turbulence affecting the 3D point cloud data in the noise cluster when a turbulence occurs at the same moment; is the probability of the turbulence of the three-dimensional point cloud data in the noise cluster affecting the secondary turbulence occurring at the same time; is the probability of the three-dimensional point cloud data in the noise cluster being affected by three turbulences occurring at the same time; is the probability of the turbulence effect of the three-dimensional point cloud data in the noise cluster on four turbulences occurring at the same time; is the number of 3D point cloud data belonging to the noise cluster in the zth sub-area in the road surface area of ​​the current frame; The total number of 3D point cloud data that can be accommodated in the road area corresponding to the monitoring vehicle in the current frame; is the number of 3D point cloud data belonging to the noise cluster in the dth sub-area in the road surface area of ​​the current frame; is the number of 3D point cloud data belonging to the noise cluster in the fth sub-area in the road surface area of ​​the current frame; is the number of pairwise combinations in the four sub-areas; The number of combinations of three sub-regions selected from four sub-regions; , , and are the number of 3D point cloud data belonging to the noise cluster in the four sub-regions.

[0056] Among them, z, d and f are different from each other. Among them, the 3D point cloud data in the noise cluster has an impact probability of a bump occurring at the same time , that is, any one of the four tires is affected by foreign objects such as stones and produces a bump; Represents the total amount of 3D point cloud data that can be accommodated in the road area ; Represents the ratio of the amount of noise point cloud data in any of the four divided sub-areas to the total amount of point cloud data that can be accommodated in the divided rectangular area; Represents the sum of the noise point cloud data proportions of the four divided rectangular areas; where z represents any one of the four divided areas, d represents any one of the remaining three areas based on the former, and f represents any one of the remaining two areas based on the former two. It represents the product of the proportion of 3D point cloud data belonging to the noise cluster in any two sub-regions. Represents the number of pairwise combinations in the four sub-areas, It represents the average sum of the products of the proportions of the three-dimensional point cloud data belonging to the noise cluster in any two divided areas; Similarly, the average is the sum of the products of the proportions of the three-dimensional point cloud data belonging to the noise cluster in all three arbitrary sub-regions; The probability that the three-dimensional point cloud data in the noise cluster affects the four turbulences occurring at the same time is directly multiplied by the proportion of the three-dimensional point cloud data belonging to the noise cluster in each of the four sub-areas. .

[0057] That is, the probability of turbulence impact caused by different turbulence occurrences is obtained.

[0058] Step three, calculating the bump probability of different bumps occurring during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bump influence probability.

[0059] Assume that the monitoring vehicle takes n seconds to pass through the circumscribed rectangular road below it, and K bumps occur per second due to road noise. In n seconds, the minimum number of bumps per second is 0 and the maximum is 4. Then, the limit formula is used to express the probability value of the event "K bumps occur at a single moment" in this road section. Calculate the probability of any number of bumps occurring during the monitoring vehicle's passage through the circumscribed rectangular road section below it, as the representation value of the road condition interference. That is, according to the bump influence probability, calculate the bump probability of different bumps occurring during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data, and use the bump probability as the road condition interference.

[0060] The calculation formula for the bump probability is:

[0061]

[0062] in, To monitor the probability of X bumps occurring when the vehicle passes through a single road area completely; To monitor the probability of X bumps occurring when the vehicle passes through a single road area. is the binomial distribution formula about (n, X); K is the number of any bumps that occur per second; X is the number of times a certain type of event occurs X times in n seconds; n is the time when the monitoring vehicle passes the actual road image corresponding to the 3D point cloud data.

[0063] Among them, n is the time it takes for the monitoring vehicle to pass through the circumscribed rectangular section of the road below it, that is, the time it takes for the monitoring vehicle to pass through the road surface area. The number of random bumps per second K = (1, 2, 3, 4). Represents the probability that no K turbulence events occur at a single moment. is the binomial distribution formula for (n, X), that is, the event of K turbulence occurring every second in n seconds is independent, and there are only two possibilities, either K turbulence occurs in this second, or K turbulence does not occur. The probability of the event of K turbulence occurring every second occurring X times in n seconds is , It represents the probability of "K bumps at a single moment" occurring X times when the monitoring vehicle completely passes through the road section. It can represent the probability of any number of bumps at a single moment occurring any number of times. Therefore, it can also represent the road condition interference of the road section. It should be noted that a single moment refers to one second, and K bumps at a single moment means K bumps per second due to road noise.

[0064] Step S300, based on the bump probability, obtain historical frame 3D point cloud data similar to the current frame 3D point cloud data, input the similar historical frame 3D point cloud data into a Gaussian mixture model, and use the Gaussian mixture model to perform motion compensation on the 3D point cloud data of the next frame of road image.

[0065] The bump probability of the current frame is used to perform similarity matching on the historical frame 3D point cloud data, and the historical frame slices are screened and input into the Gaussian mixture model for motion compensation. The similarity between the current frame and the historical frame road condition data is calculated to screen out the historical frame data with higher similarity as the historical frame time series slice. In other words, the bump probability of the current frame is used to perform similarity matching on the historical frame point cloud data to obtain the historical frame time series slice with higher similarity.

[0066] The calculation formula for similarity is:

[0067] in, is the similarity between the current frame 3D point cloud data and the historical frame 3D point cloud data; is the number of historical frames; The probability of X bumps occurring when the monitoring vehicle corresponding to the current frame passes through a single road area completely. To monitor the probability of X bumps occurring when the vehicle passes through a single road area.

[0068] Among them, v represents the current frame, and c represents any frame in the historical frame data; Represents the ratio of the bump probability between the current frame and the historical frame. represents the sum of the ratios. If the traffic data of the current frame is more similar to that of the historical c-th frame, The closer the result is to 1.

[0069] After calculating the similarity of the bump probabilities corresponding to the current frame and the historical frame, the historical frame with a similarity greater than the preset similarity threshold is regarded as a historical frame similar to the current frame, and the corresponding historical frame 3D point cloud data similar to the current frame 3D point cloud data is obtained. In the embodiment of the present invention, the preset similarity threshold is set to 0.9, and in other embodiments, the implementer can adjust the value according to the actual situation. The 3D point cloud data of all historical frames in the historical frame data that are similar to the road condition data of the current frame are obtained by the preset similarity threshold as the historical data slice for predicting the next frame.

[0070] Then, the 3D point cloud data of this historical frame is used as the historical data time series slice for predicting the Gaussian mixture function of the next frame motion compensation, and the point cloud data in the slice frame is input into the Gaussian mixture model. The three parameters of the mixture model are updated: the mixture coefficient, the mean vector, and the covariance matrix, and the posterior probability of the mixture component of each point cloud in the frame is calculated, and the model parameters are updated in sequence; then the Gaussian sorting function is inserted to sort the mixture coefficients of the point cloud, and the feature points are classified in sequence. If the Gaussian model to which the feature point belongs is the background, the feature point is classified as the background, otherwise the feature point is classified as a non-background feature point; finally, according to the height difference between the background feature point and the calibrated road surface, the vertical motion distortion of the current frame when collecting the point cloud data can be obtained, and then the state of the next frame of point cloud data can be predicted to achieve motion compensation. It should be noted that the use of Gaussian mixture model for laser radar motion compensation is a well-known technology for those skilled in the art and will not be elaborated on here. For details, please refer to the literature Pang Zhengya, Zhou Zhifeng, Wang Lirui and Ye Juelei, "Motion compensation method based on laser radar", Laser and Optoelectronics Progress, Vol. 57, No. 2, January 2020, 021106-4, which discloses an improved motion compensation algorithm based on Gaussian mixture model.

[0071] Step S400, classifying the motion compensated three-dimensional point cloud data to obtain a defect cluster; and obtaining corresponding defect parameters according to the combination of the three-dimensional point cloud data in the defect cluster and the corresponding actual road image.

[0072] The motion-compensated three-dimensional point cloud data is classified, and its classification method is similar to the clustering method in step S200. Specifically: for the motion-compensated three-dimensional point cloud data, the three-dimensional point cloud data is clustered based on density to obtain multiple clusters; the height mean of each cluster is calculated, and the clusters with a height mean less than the road surface calibration height are classified into the disease cluster. The coordinate information of the three-dimensional point cloud data in the disease cluster is obtained, and then the GPS positioning is used as a guide to locate it at the road surface position corresponding to the three-dimensional point cloud data, and the actual road image corresponding to the disease cluster is obtained, that is, the road image corresponding to the disease cluster at the GPS coordinate position is taken by the vehicle-mounted camera, and the three-dimensional point cloud data is matched and combined with the corresponding actual road image data, and the disease cluster obtained by the laser radar of the monitoring vehicle is projected onto the corresponding actual road image, that is, the third-dimensional depth information is added on the basis of the original two-dimensional actual road image, so the three-dimensional point cloud data is combined with the corresponding actual road image to simulate a complete disease 3D image and obtain the corresponding disease parameters. It should be noted that the combination of point cloud data and image is a well-known technology for those skilled in the art, and will not be repeated here. The damage parameters include parameter information such as the size of the damage area and the volume of the depression, so as to obtain a more accurate repair plan and complete the road inspection work. The corresponding damage parameters are obtained, such as the method of obtaining the size of the damage area: obtaining the number of pixels corresponding to the three-dimensional point cloud data in the corresponding damage cluster on the actual road image, and using the number of pixels as the size of the damage area; the method of obtaining the depression volume of the damage is: multiplying the size of the damage area by the absolute value of the minimum height of the pixels corresponding to the three-dimensional point cloud data in the damage cluster to obtain the depression volume of the damage. It should be noted that the purpose of directly multiplying the absolute value of the minimum height is to be prepared for subsequent repairs, to determine the volume of the dent with the minimum height, and to avoid the problem of less patching material carried during repairs; multiplying by the absolute value of the minimum height is because, taking the calibrated height of the pavement as the benchmark, the value corresponding to the calibrated height of the pavement is 0, the height value above the calibrated height of the pavement is a positive number, and the height value below the calibrated height of the pavement is a negative number, so in the case of a dent, the minimum height is a negative value, and what we want to calculate here is the volume of the dent of the disease, so the area of ​​the disease is multiplied by the absolute value of the minimum height, and the multiplication is the absolute value of the minimum height, then the resulting dent volume is a positive number.

[0073] In summary, the embodiments of the present invention relate to the field of measurement and testing technology. The method first obtains a road image, and uses a vehicle-mounted laser radar to obtain a road image and three-dimensional point cloud data; clusters the three-dimensional point cloud data to obtain a noise cluster; calculates the probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area of ​​the road surface area; calculates the probability of different bumps during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bump influence probability; based on the bump probability, obtains historical frame three-dimensional point cloud data similar to the current frame three-dimensional point cloud data, inputs the similar historical frame three-dimensional point cloud data into a Gaussian mixture model, and uses the Gaussian mixture model to perform motion compensation on the three-dimensional point cloud data of the next frame of the road image; classifies the motion-compensated three-dimensional point cloud data to obtain a disease cluster; and obtains corresponding disease parameters according to the combination of the three-dimensional point cloud data in the disease cluster and the corresponding actual road image. The present invention has a good compensation effect on the point cloud data motion compensation of the road environment and the road with poor flatness, thereby making the detection result of the road disease more accurate.

[0074] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0075] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A road intelligent inspection method with positioning function, characterized in that: The method comprises the following steps: Acquire road images, and use the vehicle-mounted laser radar to acquire three-dimensional point cloud data collected when collecting road images; Clustering the three-dimensional point cloud data to obtain noise clusters; calculating the bumping influence probability of different bumps according to the number of three-dimensional point cloud data belonging to the noise cluster in each sub-area in the road surface area; calculating the bumping probability of different bumps occurring during the time when the monitoring vehicle passes through the actual road image corresponding to the three-dimensional point cloud data according to the bumping influence probability; Based on the bump probability, three-dimensional point cloud data of historical frames similar to the three-dimensional point cloud data of the current frame are obtained, the similar three-dimensional point cloud data of the historical frames are input into a Gaussian mixture model, and the Gaussian mixture model is used to perform motion compensation on the three-dimensional point cloud data of the next frame of road image; Classify the motion-compensated 3D point cloud data to obtain a defect cluster; and obtain corresponding defect parameters by combining the 3D point cloud data in the defect cluster with the corresponding actual road image; The method for obtaining the noise cluster is as follows: clustering the three-dimensional point cloud data based on density to obtain multiple clusters; calculating the height mean of each cluster, and classifying the clusters whose height mean is greater than the calibrated height of the road surface into the noise cluster; the three-dimensional point cloud data includes height; Among them, the method for obtaining the defect cluster is: for the motion-compensated three-dimensional point cloud data, the motion-compensated three-dimensional point cloud data is clustered based on density to obtain multiple clusters; the height mean of each cluster is calculated, and the clusters whose height mean is less than the calibrated height of the road surface are classified into the defect cluster.

2. The method for intelligent road inspection with positioning function according to claim 1, characterized in that: The calculation formula for the probability of turbulence impact of different turbulence occurrences is: in, is the probability of the turbulence affecting the 3D point cloud data in the noise cluster when a turbulence occurs at the same moment; is the probability of the turbulence of the three-dimensional point cloud data in the noise cluster affecting the secondary turbulence occurring at the same time; is the probability of the three-dimensional point cloud data in the noise cluster being affected by three turbulences occurring at the same time; is the probability of the turbulence effect of the three-dimensional point cloud data in the noise cluster on four turbulences occurring at the same time; is the number of 3D point cloud data belonging to the noise cluster in the zth sub-area in the road surface area of ​​the current frame; The total number of 3D point cloud data that can be accommodated in the road area corresponding to the monitoring vehicle in the current frame; is the number of 3D point cloud data belonging to the noise cluster in the dth sub-area in the road surface area of ​​the current frame; is the number of 3D point cloud data belonging to the noise cluster in the fth sub-area in the road surface area of ​​the current frame; is the number of pairwise combinations in the four sub-areas; The number of combinations of three sub-regions selected from four sub-regions; , , and are the number of 3D point cloud data belonging to the noise cluster in the four sub-regions.

3. The intelligent road inspection method with positioning function according to claim 1, characterized in that: The calculation formula of the bump probability is: in, To monitor the probability of X bumps occurring when the vehicle passes through a single road area. To monitor the probability of X bumps occurring when the vehicle passes through a single road area. is the binomial distribution formula about (n, X); K is the number of any bumps that occur per second; X is the number of times a certain type of event occurs X times in n seconds; n is the time when the monitoring vehicle passes the actual road image corresponding to the 3D point cloud data.

4. The intelligent road inspection method with positioning function according to claim 1, characterized in that: The acquiring, based on the bump probability, historical frame three-dimensional point cloud data similar to the current frame three-dimensional point cloud data comprises: Calculate the similarity of the bump probabilities corresponding to the current frame and the historical frame; take the historical frame whose similarity is greater than the preset similarity threshold as the historical frame similar to the current frame, and obtain the corresponding historical frame 3D point cloud data similar to the current frame 3D point cloud data.

5. The intelligent road inspection method with positioning function according to claim 4, characterized in that: The calculation formula of the similarity is: in, is the similarity between the current frame 3D point cloud data and the historical frame 3D point cloud data; is the number of historical frames; The probability of X bumps occurring when the monitoring vehicle corresponding to the current frame passes through a single road area completely. To monitor the probability of X bumps occurring when the vehicle passes through a single road area.

6. The intelligent road inspection method with positioning function according to claim 1, characterized in that: The corresponding disease parameters are obtained by combining the three-dimensional point cloud data in the disease cluster with the corresponding actual road image, including: The disease parameters include the area size of the disease and the volume of the depression; the three-dimensional point cloud data in the disease cluster is projected onto the corresponding actual road image, and the number of pixels corresponding to the three-dimensional point cloud data in the disease cluster on the actual road image is the area size of the disease; the area size of the disease is multiplied by the absolute value of the minimum height of the pixel points corresponding to the three-dimensional point cloud data in the disease cluster to obtain the depression volume.

Images