Fall Detection under Privacy Protection Using Multi-layer Compressed Sensing

As the phenomenon of empty nest elderly becomes increasingly obvious, how to protect the health of the elderly living alone has attracted researchers' attention. Among all the researches, fall as the most common accident of the elderly, its corresponding vision-based detection system design is the most important. However, directly using image-clear traditional cameras to monitor the daily lives of the elderly at home will bring the risk of privacy disclosure. Therefore, this paper proposes a fall detection system under privacy protection. Firstly, multi-layer compressed sensing (CS) model is introduced to process the video frames, so that the video can reach visual shielding effect. Then, for the compressed video, we improve the local binary pattern on three orthogonal planes (LBP-TOP) feature to represent the object behavior effectively. Finally, the fall detection problem is transformed into a behavioral binary classification problem. The experimental results on two public datasets show that the specificity, sensitivity and accuracy of the algorithm proposed in this paper have maintained at a good level.