Case Study Details

Road safety is a critical concern, with highways often seeing a disproportionate number of incidents. This project addresses this by developing an intelligent surveillance system that uses AI to analyze real-time video feeds from highway cameras. The system is designed to detect a wide range of anomalies and hazardous situations, such as vehicles moving in the wrong direction, over-speeding, sudden stops, pedestrians on the highway, cattle crossings, and fallen objects. Our research involved an evolutionary approach, starting from foundational concepts like video classification and progressing to sophisticated Generative Adversarial Network (GAN) based models for anomaly detection. These models are trained to understand normal traffic patterns and identify deviations. A key focus has been the development of computationally efficient models, like CVAD-GAN, suitable for deployment on edge computing platforms, ensuring real-time processing (e.g., achieving ~32 FPS on specific edge hardware) without requiring high-end GPUs for inference. We created custom datasets capturing diverse anomalies to train and validate these models. The system also incorporates object detection and tracking capabilities to precisely identify traffic violations. Challenges addressed include robust performance in varying conditions, effective vehicle tracking across multiple camera views, and defining the precise area of interest for surveillance. This project aims to provide actionable alerts to authorities, contributing to faster incident response and overall enhancement of highway safety. Video demonstrations of the system in action on benchmark and custom datasets are available.