DeepTrack: Lightweight Deep Learning for Vehicle Path Prediction in Highways

Abstract

Vehicle trajectory prediction is an essential task for enabling many intelligent transportation systems. While there have been some promising advances in the field, there is a need for new agile algorithms with smaller model sizes and lower computational requirements. This article presents DeepTrack, a novel deep learning algorithm customized for real-time vehicle trajectory prediction in highways. In contrast to previous methods, the vehicle dynamics are encoded using Agile Temporal Convolutional Networks (ATCNs) to provide more robust time prediction with less computation. ATCN also uses depthwise convolution, which reduces the complexity of models compared to existing approaches in terms of model size and operations. Overall, our experimental results demonstrate that DeepTrack achieves comparable accuracy to state-of-the-art trajectory prediction models but with smaller model sizes and lower computational complexity, making it more suitable for real-world deployment.Predicting multiple possible trajectories for an active subject in the scene [4, 5] is a common practice. These trajectories are ranked based on the probability distribution of the prediction model, which may not be helpful in a real-time scenario. Hence, a deep learning algorithm for vehicle trajectory forecasting in a real-time safety-critical application must provide a single trajectory with high precision. It is imperative to consider the interactions of the surrounding