Evaluation on Deep Learning Algorithms for Real-Time Road Damage Detection

Maintaining the quality of road infrastructure has always been a crucial task as it is responsible for the economy and the safety of society's transportation. With the advancement of artificial intelligent, more efficient solutions using computer vision technologies are proposed as alternatives for the traditional hard-labour, human-manual approaches. In this paper, we experiment different deep learning algorithms to test out their performance. We will deploy and conduct comparisons on the performance of both one-stage and two-stage object detection algorithms: SSD, YOLO and Faster-RCNN. These models will be trained on the provided dataset from IEEE Crowd sensing-based Road Damage Detection Challenge 2022. Results show that Yolo has the best performance with 62.6% mAP and 17 milliseconds inference speed on YOLOv7.