With the rapid advancements in artificial intelligence and deep learning, the importance of gaze tracking technology in various applications has increasingly grown, especially in human-computer interaction and user experience research. Traditional gaze tracking methods primarily rely on tracking the user's eyes; however, these methods may not provide accurate and stable results under certain conditions. Therefore, this study aims to design and implement a novel gaze estimation technique that integrates hand and facial features to provide more accurate and stable gaze estimation results. Moreover, this technique is ultimately deployed on an embedded platform to realize a system that allows users to select desired advertisement content without physical contact. In this work, the YOLOv7 model is applied to detect and locate the hand and facial features. After difference and angle calculations with the detected features, the horizontal and vertical SVM-based classifiers will be well trained. By jointly training with the horizontal and vertical features furtherly, the single SVM classifier is applied to predict the user's gaze and interest area on an upright advertising display. By experiments on four-grid advertisements on display, the recognition accuracy with the proposed design is up to 94%.