Traffic Statistics Based on Lightweight Multi-objective Tracking Algorithm

Abstract

In the field of intelligent transportation, traffic flow statistics has been an important research branch. With the continuous development of deep learning technology, there have been many algorithms applied in the field of intelligent transportation, but the existing traffic flow statistics methods have poor accuracy, are greatly affected by environmental lighting, etc., and the large amount of computing leads to high requirements for hardware equipment and other disadvantages. In this paper, we propose a lightweight multi-target tracking algorithm based on the improved YOLOv5 and DeepSORT. A new structure of self-attentive mechanism and convolutional network integration is added to the backbone network of YOLOv5, which effectively improves the accuracy of the algorithm without enhancing the original computation. In DeepSORT tracking, a light-weight network ShufflenetV2 is used instead of the original heavy identification network to reduce the amount of network computation and make the algorithm less configurable for mobile devices. The experimental results show that the proposed algorithm is highly accurate, feasible and can calculate the traffic flow in real time.