Simultaneous person identification and activity recognition with indoor human localization from waist-level contours captured by 2D-LIDAR

Abstract

In this paper, we propose a method for multiple people tracking, identification and action recognition in a room using 2D-LDIAR at waist height. The method estimates the positions of multiple people by particle filter and, enables simultaneous person identification among 4 people and action recognition among 4 actions(standing, sitting, walking and door opening/closing). As an experiment, two people walked in a room. The RMSE of position estimation without occlusion was 0.229 m, and with occlusion was 0.307 m. These results indicate that our method have sufficient capability for human position estimation. While the accuracies of person identification and action recognition were over 90% in single person case, the accuracies in two people case showed 67% for person identification and 50% for action recognition. There is a room for improvement in multiple people case.