Traffic Police Gesture Recognition for Autonomous Driving
Chaojun Li, Shengqi Yang · 2018
In order to solve the problem of poor traffic police gesture recognition rate in complex background, proposes an algorithm for extracting traffic police gesture area and key frame based on image pyramid based L-K optical flow method. The algorithm firstly uses the optical flow method to track the moving objects in the traffic police gesture video to obtain the tracking points of the moving objects. Then the motion area of the traffic police gesture is extracted according to the feature that the traffic police gesture is different from the motion direction of the tracking point of the moving object in the background. On the basis of extracting the traffic police gesture area, according to the periodicity of the traffic police gesture tracking point motion, the key frame in the corresponding traffic police gesture video sequence is found. Finally, the traffic police gesture key frame obtained by the algorithm is input into the pre-trained deep learning model for recognition, and the recognition rate can reach 95%.