A ROBOT POSE ESTIMATION APPROACH BASED ON OBJECT TRACKING IN MONITORING SCENES

Wenbo Yuan, Zhiqiang Cao, Yujia Zhang, and Min Tan

Keywords

Object tracking, improved particle filter, multimode RANSACalgorithm, pose estimation, trajectory fitting

Abstract

In this paper, a robot pose estimation method based on object tracking is proposed. Robot detection is firstly implemented based on the histogram of gradient (HOG) descriptor with a dynamic scale to improve the real-time performance to some extent. An improved particle filter with temporal and spatial dynamic function and optimized appearance function is then employed to ensure the continuity of object tracking precision. Using the multi-mode RANSAC algorithm, the historical trajectory is divided into some sub-trajectories, each of which is corresponding to a motion mode. On this basis, an attitude estimation model is designed, and the attitude angle of robot in monitoring scenes is solved by trajectory fitting. High-precision tracking and attitude estimation are obtained in the experiments, and the results show that the proposed method is robust to noise and outliers.

Important Links:



Go Back