Comparative Performance Evaluation of Real-Time Webcam and HOG-Based Facial Recognition Systems for Children's Screen Time Management

Abstract

This research examines two distinct facial recognition methods: a real-time webcam-based system utilizing a custom dataset and an approach employing Histogram of Oriented Gradients (HOG) with the Olivetti dataset. The study focuses on evaluating the accuracy and efficiency of these methods in facial recognition tasks. The webcam-based system demonstrated strong proficiency in live environments, accurately identifying individual children and dynamically updating their screen time, showcasing real-time adaptability and personalized monitoring. Its performance is illustrated through screenshots displaying the interface, where individualized screen time percentages are overlaid on the children'’s faces. In contrast, the HOG method, applied with the Olivetti dataset, achieved a peak accuracy of 95.00% using optimal HOG parameters, reflecting its robust recognition capabilities. This study contributes to the field of screen time management by exploring how facial recognition technologies can support healthier digital habits among children. It highlights critical security and privacy considerations, particularly the responsible handling of facial data in environments accessible to minors. The study concludes that while both methods are effective, the real-time system offers superior adaptability for dynamic applications requiring continuous user monitoring. These findings advance the field of computer vision, offering insights into optimizing facial recognition technologies for practical and ethical applications.