Neural Network Framework for Collision Avoidance in Connected Vehicle Environments

Most current automotive vehicles feature a sensor-based safety system that warns the driver if there is a likelihood of a crash based on the current system parameters and output of the sensor detection system. However, despite the advancement in vehicle safety technologies, the number of fatal car crashes has been steadily increasing in the last decade. Recent research has introduced machine learning methods in different vehicle motion and prediction systems. In most of the proposed methodologies, the model prediction accuracy is not sufficient for human safety. This paper proposes a novel vehicle collision avoidance system based on vehicle trajectory prediction using machine learning (ML) techniques. The proposed method calculates the time-to-collision (TTC) of the host vehicle to a lead vehicle in a connected vehicles (CV) environment. The developed convolutional neural network (CNN) model was evaluated using a dataset of natural-istic driving scenarios. Results show that the proposed model can predict vehicle trajectories with higher accuracy, outperforming compared methods. The proposed collision avoidance system shows promising results in accurately calculating TTC and warning involved vehicles of potential collisions.