Face detection and tracking in video using MTCNN

Abstract

This study delves into the groundbreaking integration of artificial intelligence (AI) into crime prevention and perpetrator identification through the lens of facial recognition technology. By harnessing the intricate features of the human face, facial recognition emerges as a promising tool in these critical domains. The research aims to tackle the inherent challenges in achieving precise and reliable facial detection by developing and refining a robust machine learning model. The foundation of this investigation lies in the utilization of the Chokepoint dataset, a comprehensive repository comprising 28,470 indoor images. Through meticulous analysis and model refinement, the study achieves an impressive accuracy rate of 96.64% in facial recognition. Moreover, the research extends its scope to outdoor scenarios by leveraging the Bengal dataset, specifically designed for multi-face detection. With the integration of a supplementary 1000- image dataset, the developed model demonstrates a remarkable confidence score of 98.19%. Central to the success of this endeavor is the adoption of the Multi-Task Convolutional Neural Network (MTCNN), a versatile framework celebrated for its proficiency in simultaneously detecting faces, facial landmarks, and estimating facial attributes. By harnessing the capabilities of MTCNN, the research not only enhances detection accuracy but also improves operational efficiency. Beyond the technical advancements, the study also emphasizes algorithmic enhancements and detection mechanism optimizations, aimed at pushing the boundaries of facial recognition systems. These innovations are poised to elevate the efficacy of security measures and enhance user interface interactions, thereby bolstering the overall utility and applicability of facial recognition technology in diverse real-world scenarios.