This study proposes a target person tracking method using image recognition. We developed a human tracking robot equipped with a small camera microcontroller, three thermal array sensors, and four omniwheels. A lightweight object detection model, FOMO, is used for image recognition to reduce computational load and robot size. It is also possible to estimate the distance between the person and the robot from the change in the coordinates of the center of gravity of the person to be tracked, which is obtained from a low-angle viewpoint unique to small robots. Then, we are conducting experiments to compare the tracking performance between the tracking algorithm using the developed FOMO and the tracking algorithm using the thermal array sensor. We have also developed a person tracking algorithm that combines a thermal array sensor and FOMO, and have shown that it can track a target person even in a crowded environment.
Itoya et al. (Wed,) studied this question.