In automated container terminals， automated intelligent vehicles （AIVs） replace traditional internal container trucks， transporting cargo between the quay and yard areas. A significant challenge faced by AIVs in this process is the proactive identification of various obstacles within the port environment. This paper proposes an obstacle detection approach for port AIV operational scenarios based on deep learning， submitting the identified information to the intelligent electronic container system for subsequent dispatch planning. In laboratory scenarios， the deep learning approach achieves a 95% accuracy rate in recognition. In actual port application scenarios， it attains an identification rate of over 90%， and has been put into practical use.