Abstract
Gaze estimation is a fundamental component in human–computer interaction, assistive technologies, and behavioral analysis. Conventional gaze-tracking systems rely on infrared sensors or head-mounted hardware, which are expensive and difficult to scale for real-world deployment. Moreover, software-based approaches often require complex calibration, reducing usability. The proposed work presents an efficient machine learning–based gaze-tracking system that operates using standard webcams, thereby eliminating the need for additional equipment. The proposed system employs MediaPipe for facial landmark localization and a Support Vector Regressor (SVR) to estimate gaze direction from extracted facial features. Frame segmentation analysis incorporating pitch, yaw, and roll parameters further enhances prediction accuracy. Experimental evaluations demonstrate high real-time precision, supporting applications in behavioral research, heatmap generation, and user engagement assessment. The proposed framework offers a cost-effective, hardware-independent, and easily deployable alternative to traditional gaze-tracking systems.