利用多重假設法基於極座標網格之室內行動機器人強健人類追蹤演算法

Abstract

Mobile robots manipulating in human environments with the information about where the target pedestrian is and in which direction the target pedestrian is heading typically can improve their interaction behaviors. By providing the capability of tracking the target pedestrian in an appropriate manner, the robot can assist people in various ways under different environments. In this thesis, a complete robot perception system which consists of robot localization, mapping, moving point detection, pedestrian detection, and target tracking for robust target pedestrian tracking in an unknown indoor dynamic environment is presented. To acquire robot position, a modified scan matching algorithm, called multi-scans ICP (MICP), is used by correcting raw odometry information. To map the environment, a novel polar grid based mapping approach is proposed to accomplish mapping task without suffering sensor information digitization problem, and the proposed polar grid mapping platform is able to handle the laser based sensor limitations which might critically diminish the mapping and moving objects detection results. To detect moving points under grid based system, a modified inverse observation model is proposed to overcome several frequently happened detection limitations which are not considered in the original framework. To detect pedestrian from moving points, three pedestrian extraction techniques are proposed to filter out less possible clusters. To track targets, the multiple hypothesis tracking (MHT) algorithm is chosen to deal with the data association problem for robust pedestrian tracking purpose. The key contribution of the thesis is the combination of polar grid based system and MHT which enhances the reliability and robustness of the pedestrian tracking especially when measurements are noisy. Three laser range finder tests are performed to demonstrate the laser limitations. Besides, three pedestrian tracking experiments are designed to evaluate target tracking performance using different algorithm frameworks. The results compare and evaluate the effectiveness of each step in the proposed algorithm framework. In addition, two real-life pedestrian tracking scenarios are performed in order to show that the proposed system can be applied in real situation. The proposed robot perception system has been proved that it is still capable of tracking the target robustly when environmental and detecting noises are present.