Hybrid of deep recurrent network and long short term memory for rear-end collision detection in fog based internet of vehicles

The development and use of intelligent transportation systems as an emerging trend in the application of computational intelligence within the concept of internet of vehicles (IoV) is attracting attention in the academia and industries. The computational intelligence algorithms such as the deep learning is playing a significant role in the IoV. The detection of rear end collision in the IoV is a critical aspect of the IoV because of safety. Previous research applied genetically optimized artificial neural network (ANN) (GA-ANN) for the detection of rear end collision in the IoV. However, GA-ANN has limitations such as stretching of the entire network leading to reduced flexibility, deteriorating performance as data increases to very large size, lack of generalization power resulting from over-fitting of the training data and calling for additional effort for regularization functions to monitor the model's complexity. In this research, as an alternative to GA-ANN for improving the accuracy of rear end collision detection in IoV, we propose a hybrid of Deep Recurrent Neural Network (DRNN) and Long Short Term Memory (LSTM) (DRNN-LSTM) for rear end collision detection in the IoV. The scenario for the IoV is created and simulated to generate the dataset. The propose DRNN-LSTM is applied to detect the rear end collision in IoV. The performance of the proposed DRNN-LSTM is compared with the constituent algorithms of the DRNN-LSTM, ANN and GA-ANN. The results show that the proposed DRNN-LSTM detects rear end collision in IoV better than the GA-ANN, LSTM, DRNN and ANN. Thus, our propose approach has potential to improve safety, and effectiveness of the overall vehicle mobility in the IoV environment. Keywords: Collision Detection; Internet of Vehicles; Long Short Term Memory; Deep Recurrent Neural Network; Deep Learning Algorithm.