RGB-D people tracking by detection for a mobile robot

In this work, we propose a fast and robust multi-people long-term tracking algorithm for mobile platforms equipped with RGB-D sensors. The approach we followed is based on the clustering of the scene by using 3D information in conjunction with a reliable HOG classifier to identify people among these clusters. For each detected person, we instantiate a Kalman filter to maintain and predict his location, and a classifier trained on-line to recover the track even after full occlusions. We also perform some tests on a challenging real-world indoor environment whose results have been evaluated with the CLEAR MOT metrics. Our algorithm proved to correctly track 96% of people with very limited ID switches and few false positives, with an average frame rate of more than 25 fps. Moreover, its applicability to robot-people following tasks have been tested and discussed