Enhanced Siamese Network for Tracking Two-Wheeler Riders Without Helmets Using Traffic Surveillance Video

Motorcycles are an integral part of transportation in many parts of the world, offering a flexible and efficient means of travel. However, with the convenience and freedom they provide, motorcycle riders face a heightened risk on the road. Among the many safety measures available, wearing a helmet stands out as a critical, life-saving practice. Nevertheless, enforcing helmet use remains a challenge. This research delves into the realm of traffic surveillance and deep learning to provide an innovative solution for monitoring and tracking motorcycle riders without helmets, ultimately contributing to improved road safety. Our study introduces a cutting-edge method for tracking motorbike riders without helmets in traffic surveillance videos. We employ an “Enhanced Siamese Network” (ESN) that assesses the similarity between input data pairs to identify and track non-compliant riders. To boost ESN's performance, we use a “triplet loss function” instead of the conventional logistic loss. This enables the network to distinguish helmetless riders from those with helmets more effectively. We further enhance the network by fine-tuning its hyper-parameters with an “Artificial Flora Optimization (AFO)” algorithm. The Proposed method's effectiveness is contrasted with alternative tracking methods such as ESN (Triplet Loss), Siamese network, SORT as to the average overlap Rate corresponds to Robustness(Failure rate) and Accuracy. The outcomes of the trial demonstrated that our suggested strategy achieved an outstanding maximum accuracy of 0.972, a high overlap rate of 0.867, and a minimal failure rate of 0.120. These results highlight the effectiveness of our approach in tracking helmetless riders, marking a significant contribution to road safety and helmet use compliance.