Multi-Input Faster Region-Based Convolutional Neural Network(ConvNet) and Facial Landmark Feature-Based Crowd Recognition: A Deep Learning Approach

In light of global health crises, such as the COVID-19 pandemic, the importance of touchless facial authentication systems has become evident. Traditional face recognition-based models face difficulties in dynamic environments where individuals are in motion and their facial features may change over time. This research proposes a method based on the Faster RCNN architecture, which achieves an accuracy of 98%, surpassing conventional models like LBPH (Local Binary Pattern Histogram), Eigen face, and Haar Cascade. By incorporating a Region Proposal Network (RPN) and utilizing inputs from various angles, with a demonstrated average accuracy of 94% in practical examinations. In nine standard cases, the system achieves an accuracy rate of 96%, showcasing its reliability in diverse scenarios. Furthermore, the model effectively addresses challenges posed by facial obstacles such as masks and sunglasses, exhibiting a commendable detection accuracy of 89%. By storing processed RPN for future recognition, the proposed model not only enhances accuracy but also improves detection speed.