Target detection algorithm based on improved homomorphic filter in haze days
Yanzhi Dong, Lekang Liu, Jiaxin Xu, Guangqi Wan · 2022 Global Reliability and Prognostics and Health Management (PHM-Yantai) · 2022
With the frequent occurrence of smog weather, traffic accidents occur frequently. It poses new challenges to the transportation system. Aiming at the serious fog weather, the paper proposes a YOLO-Demist algorithm based on the homomorphic filter function model which adopted the linear improvements factor em. By using the improved homomorphic filter for image enhancement and the maximum inhibition processing on photographs or videos, the results show that the average detection rate is increased from 66.01% to 83.16%, the detection rate is more stable, and the leakage and error detection rate are significantly improved compared with YOLO-V3. The algorithms run with OpenCV for target detection, and the main objects detected are cars and birds. Compared with HLE (Histogram Equalization), SSR(Single Scale Retinex), and other traditional algorithms, the results show that the algorithm can effectively detect the visual disability of road traffic objects under haze weather and reduce the frequency of traffic accidents under severe weather.