Human Following Based on Visual Perception in the Context of Warehouse Logistics

Abstract

Warehousing and logistics robots, which have benefited from the development of 5G, the internet, artificial intelligence, and robot technology, are commonly used to assist warehouse personnel in picking up or delivering heavy goods at dispersed locations along dynamic routes. However, traditional programs that can only accept instructions or be preset by the system lack flexibility and existing human auto-following techniques either have difficulty accurately identifying specific targets or rely on a combination of lasers and cameras that are cumbersome and not effective at obstacle avoidance. This paper presents an algorithm that combines DeepSort and a width-based tracking module to track targets and uses artificial potential field local path planning to avoid obstacles. The algorithm is evaluated in a self-designed flat bounded test field and simulated in ROS, and is found to achieve state-of-the-art results in following and successfully reaching the end-point without hitting obstacles.