Modelling a deep network using CNN and RNN for accident classification

Abstract

There is a rapid evolution in highway investigation due to the increased usage of vehicles. Recently, various investigators have huge attention to the Intelligent Transportation System (ITS). Predicting a particular image plays a crucial role in digital image processing. However, this process is more challenging due to its changing viewpoints, illumination, colour, shape, etc. This work concentrates on modelling a novel deep learning (DL) approach. Initially, a deep-Convolutional Neural Network (D–CNN)–based sample-level classifier is proposed iteratively to measure the sampling uncertainty. Here, the uncertainty evaluation method is integrated with the sample learning process for training the robust sample classifier. The proposed D-CNN model has the competency to identify the abnormal activities and feature representation simultaneously. Subsequently, Deep Recurrent Neural Network (D-RNN) is proposed to examine the feature representation from the bag of samples and generates the final classification output by considering the local and global sampling aggregately. The simulation is performed using MATLAB 2020a and the proposed D-CNN and D-RNN model provides better trade-off with the classification accuracy compared to other approaches. The model accuracy is 92.9%, sensitivity is 76.40%, and specificity is 100% which is substantially higher than other approaches.

Keywords

• Classification
• Deep CNN
• Deep RNN
• Uncertainty measure
• A bag of samples
• Global sampling