Publication

Abstract : The advent of deep learning combined with computer vision has brought forth unparalleled advancements in facial detection and landmark identification. One pivotal player in this transformation has been the YOLO (You Only Look Once) series, setting new milestones in object detection methodologies. Our research is centered on harnessing the YOLOv8 model to optimize face detection processes. We incorporate the OpenCV library for image processing, enhancing detection fidelity through adjustable parameters like confidence and intersection over union (IoU) thresholds. A standout feature of our methodology is its innate ability to adjust to diverse image proportions. This is achieved by innovatively resizing and padding input images, which not only maintains consistency in detection but also augments accuracy. The proposed technique not only demarcates the face but also pinpoints facial landmarks, thus offering a comprehensive spatial map for each detected face. Preliminary results on benchmark datasets underscore the model's dual advantage of speed and precision. Our approach promises not only improved face detection but also paves the way for its amalgamation into expansive facial recognition and analytic platforms.