| With the rapid development of China’s social economy,China’s highway mileage continues to grow.As the link of cultural exchange and social development,highway plays an extremely important role in China’s transportation system.Crack is one of the most common diseases of highway,the water will gradually infiltrate into the roadbed along the crack of pavement,resulting in the softness of roadbed,the decline of support,affecting its life and overall stability,bringing huge economic losses to the country,and also laying hidden dangers for the safety of transportation.Therefore,a comprehensive,rapid and accurate method is urgently needed to detect the cracks on the pavement and record the location information of cracks.The highway maintenance personnel can repair the cracks on the pavement in time when the cracks are detected,so as to avoid the further development of pavement cracks and provide a strong guarantee for highway traffic safety.However,most current pavement crack detection solutions are based on two-dimensional pavement images,which are susceptible to factors such as lighting,shooting angles and the wear and tear effect of patch strips,leaving pavement cracks to be missed.As a result,there are still many cracks left on the pavement,which continue to affect the safety of transportation.Based on the above engineering application background and current problems,this paper uses high-speed industrial camera,high-power laser transmitter,GPS,photoelectric encoder,electric car and other equipment to independently build an experimental testing car.The experimental vehicle is used to collect 2D laser image and 3D depth image of pavement,and record their latitude and longitude information.After the 2D laser image and 3D depth image of the pavement are collected,the 2D laser image should be pre-processed for image noise reduction and then the subsequent recognition.Since the image quality of 2D pavement laser image is mainly affected by fixed pattern noise,it is not feasible to reduce the fixed mode noise by disassembling the camera directly.Therefore,this paper proposes a fixed pattern noise denoising algorithm based on combined filtering,which avoids the complex operation of camera dismantling and circuit modification,and can complete the fixed pattern noise denoising of two-dimensional laser images,laying a data foundation for subsequent crack detection.Then,based on QT graphical interface development framework,this paper uses C++ and Python language combined with OpenCV and Pytorch deep learning framework to develop a software that can automatically identify pavement cracks.The software includes two image recognition modes:1.The convolutional neural network model YOLOV5 S is used to train pavement cracks model in 2D laser images,and then the deployment and use of crack detection model is completed by TensorRT.In this mode,the software only uses the two-dimensional laser images collected,so its recognition speed is fast and can meet the needs of rapid detection.However,the pavement cracks are not obvious in 2D laser images due to the shooting angle and the existence of patch strips on the surface of cracks.Thus,the recall rate is low,resulting in the phenomenon of missed cracks.2.According to the deficiencies in Mode 1,corresponding improvements have been made in Mode 2.Firstly,the HOG feature is extracted from the two-dimensional laser image of the pavement by the traditional algorithm,and the SIFT feature is extracted from the three-dimensional depth image of the pavement by gray compression.Then support vector machine(SVM)is used to train the two features respectively,and the training model results are obtained.Finally,this paper proposes a decision recognition process based on HOG feature and SIFT feature.On the basis of two-dimensional laser image detection,it gives full play to the advantages of three-dimensional depth image and fully combines the different advantages of the two multi-source images to improve the recall rate of recognition process and reduce the phenomenon of missing cracks.In this paper,the automatic detection of highway cracks is completed by using the self-built detection car and the self-developed software.Compared with the traditional solution of using two-dimensional pavement image for crack detection,the solution proposed in this paper also combines the three-dimensional depth information of the pavement on the basis of the traditional solution,which has higher recall rate and higher robustness.It has certain research significance and practical application value for the intelligent detection and maintenance of the pavement. |
