Recently sequential and stochastic human tracking method becomes popular in order to apply to complex situations that people are close to each other or are occluded by others. This paper proposes a human detection method for such tracking method, dealing with an appearance of human in observation area. We regard this human detection as an initial setting for stochastic tracking system. Firstly we detect human candidates and set initial distribution. Secondly we develop observation model using range and color information from sensors and calculate a posterior probability distribution. Thirdly we calculate expected value of human position and delete false positives, and then combine with tracking process. At last we apply the proposed method to the data acquired at the ticket gate of the railway station and show that the detection is succeeded under complex situations.
抄録全体を表示