Design and Implementation of Anti-Peeping Screen Algorithm Based on Soft-NMS YOLOv7

The occurrence of privacy breaches through screen peeping has highlighted the increasing importance of anti-peeping screen algorithms. In the traditional YOLOv7 algorithm, it may not accurately detect faces when individuals are mutually obscured. In order to improve the success rate of face detection, we propose a new anti-peeping screen algorithm, Soft-NMS-YOLOv7. Compared to the traditional NMS-YOLOv7, Soft-NMS-YOLOv7 prevents face leakage by specially handling the scores of adjacent detection boxes. Therefore, Soft-NMS-YOLOv7 exhibits better detection performance and robustness, increasing the accuracy of face detection in anti-peeping screen algorithms from 90.05% to 94.71%. Additionally, we have developed a physical model of Soft-NMS-YOLOv7 based on Raspberry Pi 4B.