Estimation of pedestrian pose and velocity considering arm swing using point-cloud data

For autonomous vehicles, it is necessary to estimate the position and velocity of nearby transportation participants. This paper introduces a point-cloud-based tracking algorithm considering the dynamic variation of the pedestrian shape. The pedestrian shape is approximated into a circle whose radius is calculated from the point-cloud data by the RANSAC, which removes the fluctuated point cloud data corresponding to the arm swing of a pedestrian. Then, the Kalman filter estimates the position and velocity based on the estimated circle. The experimental results show that the proposed method successfully suppress the variation of the arm swing.