In recent times, computer technology has been increasingly applied to recognize motorcycle helmets in real-time surveillance footage. Deep learning techniques, particularly for object detection and classification, have gained significant popularity for this purpose. However, several challenges, such as low resolution, poor lighting, unfavorable weather conditions, and occlusion, continue to impact the accuracy of existing models in detecting motorcycle helmets. To overcome these limitations, a novel approach utilizing the Faster R-CNN model has been proposed. Here, the input image is used to train the Region Proposal Network (RPN), then the Faster RCNN model is trained with the RPN weights. Live surveillance footage has been used to identify motorcycle helmets using this method, and the results have been promising, with a 95% accuracy rate. As a result, real-time helmet detection for motorcyclists is made possible by employing deep learning approaches, and the suggested strategy is an effective way to overcome the shortcomings of existing models.