Robot detection with multi-target tracking

We propose a vision-based method for detecting and tracking a mobile robot in dynamic, complex and unstructured environments, such as an office. When there are many moving objects (e.g. robot and persons) in the environment, and they interact with each other, it is not easy to estimate the correct correspondence between detected moving objects and individual targets. We introduce GPF (generic particle filter) to discard and multiply possible hypotheses about which moving object is the robot. Also, we utilize MCMC-PF (Markov chain Monte Carlo-based particle filter) to track multiple targets efficiently and robustly by predicting interactions between targets. As a result, we have achieved robust detection and efficient tracking of targets.