In today's digital age, photos uploaded daily by social media users often contain significant amounts of private information about daily life. While this private information can help businesses provide better services, it also faces the risk of being leaked. Particularly with the advancement of deep learning technologies for object detection tasks, user privacy can be easily extracted. To address this, our research team proposes a method to prevent deep neural network (DNN) detectors from identifying private objects, especially human figures. This method leverages the inherent vulnerability of deep learning models to adversarial samples to achieve privacy protection in images. The experiments on the PASCAL VOC dataset demonstrate that the proposed algorithm reduces the recall rate of human detection from 81.1% to 18.0%, while barely affecting the visual quality of the images. The results show that this method effectively prevents DNN detectors from exposing privacy while preserving the visual quality of images to the greatest extent possible.