In this paper we propose an autonomous docking and human-robot collaboration system for an automated guided vehicle (AGV). The AGV can not only navigate and dock autonomously, but also collaborate with the human by recognizing human in the environment. A human motion detection system is developed for the proposed human-robot collaboration design. A deep learning network is adopted to detect and recognize humans in the environment. By knowing of human motion, the AGV adjusts the automatic docking behavior in a collaborative manner. Practical experimental results demonstrate that human workers can co-exist with an AGV in an unstructured environment for autonomous docking tasks.