Hybrid Model for Traffic Violation Detection and E-Challan Generation-An Experimental Approach

Traffic violations are an ongoing problem that has significant impacts on public safety in India resulting in a significant number of road traffic accidents and fatalities. We present an automated approach for the detection of traffic violations and the issuance of echallans in an Indian context, considering the applicable road and traffic conditions in India. The detection of objects is done in real-time using YOLOv11-Nano, while tracking multiple objects simultaneously is performed with DeepSORT achieving mAP@0.5 of 92.3%, MOTA of 94.7%, and processing times of 8.2 ms/frame on an NVIDIA Tesla T4 GPU. The automated pipeline classifies helmet infraction, triple riding and running red lights considering 12 classes of objects with 94.7% precision and 89.0% recall. A multi-stage OCR pipeline utilizing EasyOCR detects text with 82.6% overall accuracy and 91.7% accuracy when extracting Indian number plates. The entire pipeline consists of detection, tracking, OCR, searching the database, and issuing an e-challan, to which the average response time was 3.4 seconds and 100% uptime was achieved during an 8-hour continuous validation run. This approach outperforms existing frameworks and provides the ability to economically and realistically sustain scalable multicamera solution on a single GPU framework. Future suggestions include detection of speed violations, deploying an edge-based framework, an adaptive learning approach, and integration with existing smart city infrastructure.