Multimodal Data Trajectory Prediction: A Review

Abstract

Trajectory prediction refers to predicting the future movement of an object, person, or vehicle based on past motion trajectory information and other relevant environmental information. Traditional trajectory prediction solutions are suitable for simple driving scenarios and only applicable for short-term predictions. With the improvement of computing power and data processing speed, people can analyze and utilize large-scale datasets faster, use more data and more complex algorithms to build models, and therefore better predict future trends and behaviors. In addition, different solutions focus on how to efficiently extract features from different types of information and how to more accurately predict the future trajectory of the target object. Our goal is to analyze the impact of different information on prediction, and to classify and compare deep learning-based trajectory prediction methods for vehicles or surrounding pedestrians based on the same data. According to the type and quantity of inputs, they are divided into five categories, and the use and prediction effect of different information solutions are elaborated in detail.