Experimental Evaluation of Behavior Perception and Prediction Methods in Automated Port Logistics Operations

With the rapid development of autonomous driving technologies, intelligent ports, particularly autonomous logistics, have become the focus of industry attention. Ensuring safe and efficient operations require port management systems to perceive and predict the behaviors of people and vehicles. In the filed of behavior perception, research efforts have primarily focused on the detection and tracking of vehicles, pedestrians, and obstacles under various sensor configurations. Common approaches include vision-based, LiDAR-based, and multi-sensor fusion methods. In terms of behavior prediction, existing approaches can be broadly categorized into four paradigms: model-driven, data-driven, environment-assisted, and anomaly prediction methods. Model-driven approaches rely on physical and motion models, while data-driven approaches utilize deep learning techniques. Environment-assisted approaches integrate prior knowledge such as maps, while anomaly prediction focus on identifying unexpected behaviors to improve safety. Our study shows that Probabilistic 3D significantly outperforms traditional methods in both detection and tracking accuracy for unmanned port logistics scenarios. By integrating 2D and 3D data, our method improves the detection of small obstacles such as pedestrians, bicycles, and motorcycles, addressing key limitations of pure 3D-based approaches. This enhancement makes autonomous logistics systems more reliable and robust in complex environments. Future work will focus on refining multi-sensor fusion techniques and developing adaptive learning models to further enhance system adaptability in smart port operations.