TrajForeNet: A Trajectory Forecasting Neural Network for Vehicle Movement Analysis and Prediction

Traffic accidents and fatalities impose significant social and economic costs globally. Conventional enforcement methods, such as speed cameras, reactively punish violations rather than prevent them. TrajForeNet, a neural network meant for better traffic management and safety by forecasting vehicle trajectories, is presented in this paper. Using predictive analytics and computer vision techniques, such as object detection and tracking, this research aims to leverage vehicle movement data to anticipate risks and improve traffic management, ultimately contributing to safer and more efficient road networks. The system uses modified YOLOv8 model for vehicle and pedestrian on traffic camera videos collected in Hubli-Dharwad cities. TrajForeNet uses the speeds and direction of recognized vehicles and pedestrians between frames to estimate future positions and directions of road users by means of a 3 layer Long Short-Term Memory (LSTM) network. Measured by average displacement error (ADE) and final displacement error (FDE), the performance of the model shows competitive outcomes in relation to state-of-the-art models with and ADE of 0.095 and FDE of 0.22. TrajForeNet can significantly improve road safety by providing exact movement estimates, probable future uses in traffic infraction identification and crash prediction.