Multisensor Collaborative Particle Filters for Simultaneous Robot Localization and People-Tracking

A conditional particle filter algorithm with distributed multisensor collaboration is proposed for joint estimation of the position of people and the pose of the coexisting robot.In global vision system,particle filter-based target tracking in the image plane is performed by each view whilst synchronized principle axes are integrated across views using ground plane homography.The visual observation is further asynchronously incorporated with the laser data using sequential particle filtering,in which a smoothed likelihood field model with people hypotheses is proposed to improve the robustness against the positional error and adaptive sampling based on Kullback-Leibler divergence is employed to reduce the number of particles to represent the joint distribution.Experimental results illustrate the favorable performance of the map-based simultaneous robot localization and people-tracking with multisensor collaborations,in situations with sensor noise and global uncertainty over the human-robot position.