Improving Traffic Surveillance: Deep Learning Approach for Road Anomaly Detection in Videos

In recent years, Closed-circuit Television (CCTV) cameras have been playing a vital role in the surveillance of both public and private areas. The primary objective of surveillance is to monitor human behavior and road conditions. In the real-world situation, detecting, and recognizing abnormal activities poses significant challenges due to the densely crowded environment and the complex nature of transportation systems. These factors make it difficult to automatically identify various anomalies that occur while traveling, leading to emergencies, and endangering human life and property. This study introduces an automatic detection framework for recognizing road anomalies such as accidents, fighting, car fires, and armed snatching (gunpoint) in road surveillance videos. After reviewing the literature, the review directs that convolutional neural networks (CNNs) are a specialized deep learning approach well suited for image and video analysis. The proposed methodology combines the pre-trained CNN models with Data Augmentation (DA) techniques to fine-tune hyperparameters such as learning rate and momentum that enhance the model learning accuracy and performance for recognizing road anomalies. Furthermore, it introduced a rolling prediction algorithm to solve the flickering problem during testing and created a new road anomaly dataset (RAD) as a benchmark consisting of road surveillance videos and images. Our proposed model combined with the InceptionV3 pre-trained model achieved a best accuracy is 98.81% for detection and classification as compared to other deep learning models.