The Understanding of Traffic Police Intention Based on Visual Awareness

摘要

This paper presents a system that can be used to automatically recognize the intention of traffic police based on visual awareness, which is important for driver assistance systems and autonomous vehicles. Traffic police play an important role in traffic scenes because the presence of traffic police often means there are traffic jams, accident-prone areas, or traffic failures. In this system, key points of the human body used to express the spatial pose of traffic police are extracted by OpenPose, and these key points are used to generate a spatio-temporal map by motion representation. Then, the graph convolutional network and modified Transformer are respectively used to obtain the spatial features and temporal features from the spatio-temporal map. Finally, the above features are used to infer the intention of traffic police in continuous frame images. Experimental results demonstrate that the proposed method had a higher accuracy than other state-of-the-art recognition algorithms in understanding traffic police intention.