For road traffic safety, the pavement damage detection has cost enormous manpower and material resources, whereas the efficiency of dectection has been low. To save detection time while taking advantage of the high accuracy of neural network for object detection, in this paper, we propose a improved single shot multibox detector, that achieves less time and higher accuracy for pavement damage detection. we design the module to predict locations, types and sizes of damage on pavement. In a real test environment, the model can quickly detect and classify road damage. It is of great significant for road maintenance and driving safety.

CICTP