DEVELOPMENT OF A DEEP LEARNING MODEL FOR THE PREVENTION AND CONTROL OF ROAD ACCIDENT

Subject Area: Control System Engineering

This paper describes the development of a convolutional neural network (CNN) based vehicle accident prevention and control system which is developed to improve vehicle safety by providing real time predictions and control decisions. The model has been developed with several convolutional layers followed by ReLU activation and drop out to enhance the accuracy while avoiding overfitting. The CNN model was trained on the Honda Deep Drive (HDD) dataset and other locally collected tricycle data for the training of the model, 21,300 samples. The system has been developed and implemented in MATLAB’s Simulink environment and performed effectively, with a detection rate of 98.1% and an ROC score of 0.98. It was found that the proposed system is better than the current state of the art models in terms of accuracy and reliability. These results have shown that the proposed CNN based system is efficient in accident detection and prevention when compared with the conventional methods. Due to its robustness and low computational complexity, it can be easily adopted for use in mobile and hardware environments for real time applications. Thus, this research is significant in the improvement of intelligent vehicle safety systems and provides a basis for future work on deep learning for autonomous and semi-autonomous vehicles.