Live Helmet Detection
Mrs. R. Vijaya Lakshmi1, K.Hindu2 , G.Hariharan3
1 Asst. Professor , Mahatma Gandhi Institute of Technology
2,3UG Student, Mahatma Gandhi Institute of Technology
Abstract: In recent years, computer technology has found a significant application in the realm of real time surveillance, particularly in the automatic recognition of motorcyclists wearing helmets. This endeavor has been greatly facilitated by the rise of deep learning methods, which excel in tasks such as object detection and classification. However, despite their efficacy, these methods encounter several challenges that limit their accuracy in identifying motorcycle helmets. Issues such as limited resolution, poor lighting conditions, adverse weather, and occlusion present formidable obstacles. To address these challenges, a novel approach leveraging the Faster R CNN model has been proposed. This method, unlike traditional approaches, adopts a two-step training process. Initially, the Region Proposal Network (RPN) is trained using the input image. Subsequently, the weights obtained from RPN are utilized to train the Faster R- CNN model. This methodology aims to enhance helmet detection accuracy in live surveillance footage. Experimental results have showcased promising outcomes, with a remarkable 95% accuracy rate achieved in identifying motorcycle helmets within live surveillance streams. These findings underscore the potential of deep learning methodologies in the domain of
automatic helmet detection for motorcyclists in real-time surveillance scenarios. Moreover, they demonstrate the effectiveness of the proposed strategy in mitigating the challenges faced by existing models. By harnessing the power of deep learning and refining the training process, this innovative approach has demonstrated its capability to overcome the inherent limitations of current methods. It not only enhances the safety measures for motorcyclists but also underscores the broader applicability of deep learning techniques in real-world scenarios. In conclusion, the successful implementation of the Faster R-CNN model in automatic helmet recognition signifies a significant advancement in surveillance technology. It not only showcases the adaptability of deep learning algorithms but also underscores their potential in addressing real-world challenges. Moving forward, continued research and development in this field hold the promise of further enhancing safety measures and optimizing surveillance systems for diverse applications.
