Traffic Police Pose Estimation Based on Multi-branch Network

To precisely recognizing traffic police gesture, a two-branch multi-stage convolutional neural network (CNN) and an BP neural network (BPNN) classifier has been adopted to achieve the task of traffic police pose estimation in this paper. We simultaneously detect the human joints and associate body parts. When the detection work is done, a BPNN cascades after CNN and serve as a classifier. The experiment results show that traffic police-men's gestures are well recognized by this method.