Road Rage Detection System Using Deep Learning and Computer Vision

Road rage detection in India's chaotic traffic conditions demands robust, real-time solutions. This paper proposes a 3D CNN-based system that leverages transfer learning from a violence detection model (98 % accuracy on Kaggle datasets) to a road rage-specific model (94 % accuracy on a custom Indian dataset). By incorporating temporal smoothing with a 5-frame sliding window, the system reduces false positives by 15 % compared to existing methods. Experiments demonstrate its efficacy under varying illumination and traffic densities, validated through precision (0.99 for non-violence) and recall (0.99 for violence). The system's deployment in Bangalore highlights its potential for traffic monitoring, law enforcement, and insurance claim verification. Future work includes edge-device optimization and multimodal data integration.