A Survey on Vehicle Trajectory Prediction Procedures for Intelligent Driving

Aimed at vehicle trajectory prediction procedures, this survey provides a comprehensive review for intelligent driving from both theoretical and practical perspectives. Vehicle trajectory prediction procedures are explained in terms of the perception layer, core technology of trajectory prediction, decision-making layer, and scenario application. In the perception layer, various sensors, visual-based perception devices, and multimodal fusion perception devices are enumerated. Additionally, the visual-based multimodal perception and pure visual perception techniques employed in the top five intelligent vehicles in China are introduced. Regarding the core technology of trajectory prediction, the methods are categorized into short-term domain and long-term domains, in which the former includes physics-based and machine learning algorithms, whereas the latter involves deep-learning and driving intention-related algorithms. Identically, the core technologies adopted in the top five intelligent vehicles are summarized. As for the decision-making layer, three main categories are summarized theoretically and practically, decision-making and planning for cooperation, super-computing and closed-loop, and real-time and optimization. As for the scenario application, open scenarios and closed scenarios are discussed in theory and practice. Finally, the research outlook on vehicle trajectory prediction is presented from data collection, trajectory prediction methods, generalization and transferability, and real-world application. The results provide some potential insights for researchers and practitioners in the vehicle trajectory prediction field, and guides future advancements in this field.