Fast Traffic Accident Identification Method Based on SSD Model

Abstract

Traditional traffic accident identification methods have the problems of complex detection process, poor detection performance and poor real-time performance so far. In this paper, we propose a new type of traffic accident identification method based on target detection algorithm Single Shot MultiBox Detector (SSD). We collect and simulate traffic accident data sets in different scenarios and compare the detection performance of different target detection algorithms, aiming at the problems of traffic accident detection existing in the original SSD, the idea of multi-feature fusion and adaptive default box selection algorithm are proposed to improve it. Finally, we present an evaluation on the collected data, the improved SSD_A method shows considerable performance, which can reach 97% mAP (mean average precision) at the speed of 32 FPS (frames per second).