Head Pose Tracking Method Based on Face Detection And Stereo Visual SLAM
Suibin Huang, Hua Xiao, Peng Han, Jian Qiu, Peng Li, Dongmei Liu, Kaiqing Luo · 2021
For purpose of solving the fine estimation and real time problems of existing head pose tracking methods, we propose a method based on face detection and stereo visual SLAM algorithm. Firstly, the stereo image is pre-processed by Gaussian filter and the face region is localized using adaptive Haar+Adaboost algorithm. Secondly, we use the improved ORB algorithm to extract the feature points in the image grids and match them through stereo vision technology. The invalid 3D points are excluded according to the depth threshold. Thirdly, we obtain 3D-2D matching relations through projecting mappoints from last frame to current frame. Finally, we obtain the optimized head pose through the Bundle Adjustment and the coordinate system transformation relation. The experiments show that the average error of the method in this paper is about 0.5°, indicating that the method has high precision.