Foggy Image Detection Based on DehazeNet with improved SSD

Abstract

In order to improve the ability of pedestrian detection in foggy scenes, a method is proposed to improve the performance of pedestrian detection in foggy scenes. DehazeNet convolution is combined with the improved SSD target detection algorithm to realize vehicle and pedestrian detection in foggy scene. Target detection model training was carried out by using the fog images after fog removal treatment and the original fog images, and vehicle and pedestrian detection was carried out in traffic environment with different fog concentration levels. The results showed that the mAP value of DehazeNet with SSD network could reach 79.7%, 5.4% higher than the mAP value of SSD algorithm.