Continuous pavement crack detection using ECA-enhanced instance segmentation of video images

Pavement crack detection is a vital part of road maintenance which is directly impacted by the accuracy and efficiency of detection. Traditional manual visual inspection is influenced by workers' subjectivity and has low efficiency. With the increasing of road coverage, manual detection can no longer meet the demand and needs to be replaced by an accurate and efficient scheme. The development of on-vehicle cameras and edge computing technology has led to the real-time pavement crack detection and measurement using video images. This paper proposes a continuous pavement crack detection using ECA-Enhanced Instance Segmentation of video images. Firstly, the ECA-Enhanced InstNet instance segmentation network is used to detect and extract pavement cracks with high speed. Secondly, the Hungarian algorithm is used to match the detected cracks over video images and associate them together as a whole crack occurrence targets. Then, Inverse Perspective Mapping transformation is employed to measure the geometric properties of the pavement cracks. Experiments show that the algorithm proposed in this paper achieves an AP of 71.7 %. When using 1280 * 720 resolution video for real-time detection, the Frames Per Second can reach 67.6. Additionally, the algorithm can measure the geometric properties of pavement cracks, demonstrating certain application value.