Motion Estimation based on Particle Filter with Multiple Predict Model

This paper presents a method of motion estimation from the time series image data based on particle filter with multiple prediction models. The particle filter is known to be effective for tracking people, however, the accuracy of the estimation strongly depends on the prediction model. If the motion that does not fit well with the prediction model occurs, the accuracy of estimation is degraded. In this paper, we propose a particle filter with multiple prediciton models wherein the prediction model is selected according to the likelihood value of each prediction model. Furthermore, the proposed method is applied to the carrying task with an omnidirectional camera, and the effectiveness is verified.