ENHANCED TRAFFIC JAM RECOGNITION BASES ON IMAGE PROCESSING AND DEEP LEARNING FUSION

Abstract

In today's urban life, traffic jam is one of the painful situations for traffic participants in general and drivers in particular. In recent years, many scientific and technological methods have been proposed to improve this problem during peak hours. In this paper, the author proposes a solution to identify traffic jams with the method of using the view of many cameras at intersections in the city to recognize the number and volume of traffic moving at the time and from that predicts traffic jams or not. The method proposes a combination of deep learning and traditional feature extraction methods such as SIFT (Scale-Invariant Feature Transform- SIFT). In which ScaledYOLOv4 (You Only Look Once) is used to detect vehicles such as motorbikes and cars. Then the tracking method calculates the moving speed of the vehicles in the crowd using feature matching such as SURF, SIFT is applied. The results show that the proposed method gives a good result in detecting traffic jams.