Research On Intelligent Pavement Pothole Spray Repair Technology Based On Machine Vision

The current asphalt pavement pothole repair equipment is of low informatization,which relies heavily on the experience and operation of the repair personnel.Under the background of “Made in China 2025”,intelligent pothole repair is the future development trend.Therefore,in this dissertation,under the support of the Key Scientific Research Platform Construction Subplan-level improvement project(“Green Intelligent Pavement Maintenance Robot”,No.310825173314),the intelligent pavement pothole spraying repair technology based on machine vision was studied,which includes pavement pothole detection and segmentation,pothole depth image denoising and inpainting,aggregate spraying characteristics and accumulation model analysis,and pothole spray repair path planning.The goal is to automate the process of pothole detection and pothole spray repair.The main research contents are as follows:(1)The automatic pothole detection and segmentation method based on machine vison was studied.Two different methods were proposed and compared.For the first method,the shape feature and standard deviation of the binary image were extracted to detecte the pothole,then,the texture feature vectors of GLCM were dimensionally reduced by PCA algorithm,and the FCM clustering result of texture feature and binarization result of gray image were integrated to segment the pothole region.For the second method,the wavelet coefficient of the second layer and third layer and the original gray pixel was used to construct the wavelet highfrequency energy field,then,the shape feature of the energy field was extracted to detect the pothole,and the wavelet energy field was used as the marking field of the Markov Random Field model to segment the pothole region.Compared with the existing vision-based pothole detection and segmentation methods,the constructed wavelet energy field can effectively fuse the grayscale information and texture information of the pavement image,which help to achieve better pothole detection results and solve the problem that the existing vision-based methods can not accurately segment the pothole region.For pothole recognition,the recall was 87.5%,the precision was 83.3%,and the accuracy was 86.7%.For pothole segmentation,the overlap degree between the segmented pothole region and the original pothole region was mostly more than 85%,which accounted for 88.6% of all the detected pavement pothole images.(2)The denoising and inpainting method of pothole depth image captured during the spray repair process was studied.As for denoising of the pothole depth image,a denoising method combining MSER and improved confidence analysis was proposed,of which the MSER detection was used to eliminate the regional noise and the improved confidence analysis through mathematical morphology operations was used to eliminate the high frequency noise.As for hole inpainting of the pothole depth image,a multi-resolution TV inpainting model was proposed,of which the downsampled low resolution image was used to reduce the size of the hole to increase the inpainting speed,and the value-preserving operation in the upsampling process was used to ensure the inpainting quality of the TV inpainting model.Compared with the Navier_Stokes equation based inpainting method and the FMM based inpainting method,the image restoration result of the multi-resolution TV inpainting model was better,of which the RMSE value was 3.97 and the error relative to the original data was only 2.65%.The monitoring error of pothole depth information,plane length information and volume information was less than 3%,and the average time required for monitoring was 0.61 s,which can meet the monitoring accuracy and rapidity requirements of pothole repair process.(3)The aggregate spraying characteristics and round trip aggregate accumulation model of pothole spray repair were analyzed.The Runge-Kutta method was used to solve and analyze the motion differential equation of the aggregate particle swarm,and the working efficiency model suitable for the pneumatic conveying system of the pothole spray repair vehicle was put forward,and the changing characteristics of aggregate conveying efficiency were analyzed with multi-variable method.The results showed that the pipeline inclination angle and the drag coefficient should be small,the inner diameter of the pipeline only needs to meet the aggregate flow requirement,and the aggregate-air mass mixing ratio should be large,but not more than 10,and the airflow speed should be in the range of 15~25m/s preferably.Then,the aggregate accumulation model along the straight line on the plane was established.The optimal overlap ratio between the adjacent filling lines on the plane was solved by fibonacci method and the value was 0.3078 times the spray radius.The actual spacing between the adjacent filling lines inside the pothole was analysed through approaching the optimal crossover degree.The matching relationship between the acceleration and the speed of the nozzle was obtained through analyzing the aggregate accumulation amount at the end of the filling line,and the value was a=v2/2R.The above analysis lays a foundation for spray repair path planning and spray repair experiment.(4)The pavement pothole spray repair path planning based on depth image was studied,and the intelligent spray repair test bench was built to analyze the experiment results.As for the pre-spraying problem,the platform leveling method was proposed based on directional difference minimization,the pothole positioning method was proposed based on maximum value searching,and the pavement surface depth calculation method was also proposed.As for the spray repair path planning problem,the slice that to be repaired was obtained by the proposed double flood fill method,the connection path betweent the nozzle and the initial repair point of the slice was seeked by the proposed path connection method assistanted by the contour,and the spray repair path of the slice was planed by the proposed contour guided round-trip straight line repair path planning method,which could solve the problem that the existing filling method can not reasonably and uniformly plan the spray path.The machine vision based intelligent pavement pothole spray repair test bench was built,which mainly includes repairing material conveying system and intelligent spraying control system.The experiment results of aggregates spraying repair showed that the average spacing error of the spraying repair path of four kinds of pothole was 7.62%,7.79%,7.25%,and 6.58% respectively,the average time required to plan the spraying path was 1.571 s,and the average difference between the mean square error of the repaired road surface and the original road surface was only 1.83 mm,which indicates that the proposed method can achieve good path planning accuracy,path planning speed and pothole repair flatness,and can be used for the intellegient repair of different potholes.