Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 28, 2023 - July 01, 2023
This study proposes a system to predict a trajectory through detecting pedestrian postures for a mobile robot. The posture information, which is the position coordinates and quaternions of 32 joints, is obtained as time series data using a near-infrared camera when pedestrians are walking straight or turning left and right. They are used as inputs for machine learning. Since the length of the input data greatly affects the prediction speed and accuracy, this paper discusses the effect of the length of input data on the performance of trajectory prediction. The objective is then to construct a system for detecting instantaneous changes in the posture of pedestrians.
