Black Cloud Generation Method for Privacy Protection in Fall Detection
Runqing Zhang, Haoran Yan, Dunbo Cai, Ling Qian · 2022
Fall detection is an important application in computer vision. However, the image data for fall detection contain unprotected privacy. For example, the silhouette of a person who falls may divulge information about whether the person is disabled. In addition, privacy, such as the age and height of the fallen person, will also be apparent in the image. To tackle these issues, this paper proposes a novel approach, called Black Cloud for Falling Detection (BCFD), to protect the privacy information of the fallen person. We introduce a new image segmentation method to estimate the contour of the human body image, and we generate 'black clouds' to accurately cover the complete human images. The detector was conducted using the contour information to obtain the fallen person. We also propose a method to construct a privacy-protected fall detection dataset and provide a sample dataset. The proposed method can protect the privacy information precisely compared with the state-of-the-art methods. We have validated our method on the proposed PPFD-sample dataset, observing 90.2% precision and 81.8% recall. Experimental results demonstrate that our approach can protect image privacy effectively in the fallen person detection task.