Human Target Distance Estimation System Using Mono-camera On Human-Following Mobile Robot

Abstract

A reliable and accurate target distance estimation system is very much needed in the human-following mobile robot system. Because the distance estimation system is used as input to control the mobile robot so that it can follow the human and stably maintain the distance. In other studies, additional proximity sensors have been used, such as LiDAR, ultrasonic, and laser range sensors. But the use of additional proximity sensors requires more power and for components that are more accurate usually also require more power, thus saving power consumption is an important aspect of human-following mobile robots because of the limited battery capacity. In this paper, a human-following mobile robot system is proposed that uses a mono-camera as the primary sensor without additional sensors for the human target distance estimation to save power consumption. By using Google’s human detection framework, Mediapipe, which can capture key points of the human body, this distance estimation system can be more efficient. Because it does not require additional energy by using the same output from the Mediapipe detection results, unlike using additional proximity sensors. The test results show that the use of the proposed method does not increase the power used during the detection and distance estimation process, and the results of the estimated distance have a small error with an average error of 1.3% with the human target being at a distance of 60cm to 180cm.