Vehicle Maneuver Prediction using Deep Learning Networks

Abstract

Vehicle maneuver prediction plays an important role in ADAS (Advanced Driver Assistance Systems) and autonomous vehicles. It predicts the future behaviors of surrounding vehicles based on the current and past driving states of vehicles. Accurately predicting a vehicle's future trajectory and maneuver intentions is essential for safe and efficient navigation in traffic. Compared to conventional physics-based models, deep learning approaches are getting more popular due to their better performances in complicated real-world scenarios. This dissertation studies the temporal and spatial dependencies of vehicle maneuvers in a driving trip and investigate an innovative deep learning system to predict maneuvers of surrounding vehicles. Our method utilizes a combination of sensor data such as GPS, speed, acceleration, and videos to predict the future maneuver of a vehicle. The system contains LSTM (Long Short-Term Memory) or Transformer networks to learn information from past driving states, and graph neural networks to exploit the spatial relations between surrounding vehicles. We evaluate the proposed method on a large-scale real-world dataset and compare its performance with several state-of-the-art approaches. Our results show that our method significantly outperforms existing methods in terms of accuracy and robustness. In addition to the prediction performance, we also analyze the interpretability of the proposed method and demonstrate how it can be used to identify critical factors affecting maneuver prediction. This research provides a significant contribution to the field of vehicle maneuver prediction and lays the foundation for the development of advanced ADAS and autonomous driving systems. Our method has the potential to improve the safety and efficiency of road transportation and can be used to support the deployment of autonomous vehicles in complex driving scenarios.