Image Processing Based Human Following Cart Using 360° Camera

摘要

This study involves a robot that aims to follow a human target after detection, recognition, and tracking. The system has used 360 ° cameras to provide solution to the limitation of the tracking zone of a conventional camera. The image captured is in fisheye format and was then converted into a panoramic image to be processed by You Only Look Once (yOLO) Detection Algorithm to detect the human target and Channel and Spatial Reliability Tracking (CSRT) algorithm to track the target. The proponents included a 10-second target recognition methodology after detection so the system does not need to find the target from time to time.