Vehicle Tracking Algorithm Based on Deep Learning in Roadside Perspective

Traffic intelligence has become an important part of the development various countries and automobile industry. Roadside perception is intelligent transportation system, which mainly realizes effective road environment information by using sensors installed on roadside. Vehicles are main targets in most traffic scenes, so tracking a large number vehicles subject field roadside perception. Considering characteristics vehicle-like rigid from view, vehicle algorithm based deep learning was proposed. Firstly, we optimized DLA-34 network designed block-N module, then channel attention spatial modules were added front to improve overall feature extraction ability computing efficiency network. Next, joint loss function intra-class inter-class discrimination algorithm, can better discriminate objects similar appearance color vehicles, alleviate IDs problem robustness real-time performance algorithm. Finally, experimental results showed that method had good effect for task perspective could meet practical application demands complex scenes.