Robust target detection, tracking and following for an indoor mobile robot

Abstract

Target detection, tracking and following is viewed as one of the most essential topics in mobile robotics. In this paper, we develop a robust target detection, tracking and following system for an indoor mobile robot. In the system, target detection relies on real-time histogram of oriented gradient (HOG) feature extraction and effective support vector machine (SVM) classification. Particularly, a novel and fast target tracker is designed to track detected target. The designed tracker improves traditional spatio-temporal context (STC) model by extracting middle-level feature rather than raw pixels, which is capable of tracking target better than traditional STC tracker under the circumstance of illumination changes and partial occlusion. Furthermore, proportional, integral, derivative (PID) motion control strategy is also employed to conduct robot to follow target safely and robustly. Overall system is evaluated on Pioneer3-DX mobile robot equipped with Microsoft Kinect. Excellent experimental results illustrate the effectiveness of developed system.