Research And Implementation Of Real-Time Road Crack Detection For Automatic Driving

Rapid and accurate automatic identification of cracks is of great importance for ensuring traffic safety.In recent years,with the great increase of traffic flow,the amount of cracks in the road has also increased in the same way.Pavement cracks not only make the vehicle bump,but also impact on the structure of the vehicle,affect the integrity of the vehicle structure,thus endanger the safety of the driver.The research on crack detection in academia has existed for many years,and various methods for pavement crack detection have been put forward,which has effectively helped highway maintenance departments to maintain pavement.But the maintenance is not on time,so the vehicle itself should have the ability to find cracks and avoid.For the traditional vehicle,the driver has the ability to identify cracks and make corresponding operations.For automatic driving vehicles,identifying the cracks quickly and automatically is the premise of subsequent evade operation.The traditional pavement crack detection technology has the problems of low detection accuracy and slow detection speed,which can't meet the needs of automatic driving scene.In this paper,a fast and accurate recognition method is proposed,which can meet the needs of real-time crack detection in automatic driving scenario,and is of great importance for ensuring the safety of automatic driving.Most of the existing crack recognition technologies are based on manual extraction of local image features or edge detection methods.However,the coverage of these features is limited,which may not cope with the complex pavement images with a lot of noise.In recent years,convolution neural network has attracted wide attention due to its ability to automatically extract image feature representation,and work has been tried to apply it to crack detection.But the existing work still only extracts the features of the local area of the image,and can't make use of the slit and continuous spatial structure characteristics of the cracks.In this paper,the network structure is improved,neural network has large sensing area,on the one hand this enlarge the effective use of context information in area surrounding the local features,on the other hand it let the model learn patterns from the large amount of data,which is a good way to deal with the complex situation.Experimental results show that the proposed method is more robust to image noises,and the recognition accuracy is better than traditional methods,and the recognition results have good structural integrity,which is very close to the level of artificial recognition.