Helmet Violation Detection using Hybrid Model YOLO and LSTM

Traffic safety remains a critical global concern, with helmet non-compliance among motorcyclists significantly contributing to accident severity. Existing detection systems often struggle with accuracy under varying environmental conditions and high traffic volumes. This study introduces an innovative traffic violation detection system utilizing a hybrid YOLO-LSTM model designed to enhance helmet use enforcement. The approach in the study leverages the strengths of YOLO for real-time object detection and LSTM for sequence prediction, enabling robust performance across diverse scenarios. Employed are over 1,100 images captured under varying lighting and weather conditions to train and validate the designed model. The proposed system achieved a mean average precision (mAP) of 0.777 and demonstrated superior accuracy and recall compared to traditional YOLO-based models. This research uniquely integrates temporal sequence analysis with spatial detection, offering a novel solution that significantly improves detection rates in challenging environments. The study demonstrates that hybrid deep learning models can significantly enhance automated traffic enforcement and contribute to safer roads.