Recognition Method of Road Cracks with Lane Lines Based on Deep Learning

Due to the vehicle wheeling and abrasion, the paint on the lane lines usually appears cracked. In the process of automatic detection of the pavement cracks, the paint cracks can be easily misidentified as the road cracks, reducing the recognition accuracy of the pavement cracks. We propose a lane line detection method based on deep learning method, extracting multi-angle and multi-dimensional features of lane lines automatically. A complete dataset has been constructed to solve the problems of uneven illumination, pollution and abrasion. Our method achieves a result of 91.84% precision, 86.67% recall and 87.34% Dice coefficient, which are all about 30% better than the traditional digital image processing techniques. The crack model and the lane line model are superimposed to improve the recognition effect of pavement cracks, which is better than the single crack model.