A Dynamic Height Analysis on Vision Based Fall Detection System
Yuxi Chen, Xiangbo Kong, Lehan Chen, Lin Meng, Hiroyuki Tomiyama · 2019
The senior citizens need support or help after a fall accident. But they may not be able to summon help due to the injuries or impairment of the fall accidient, and the cost of human based health care monitoring system is high. The artificial intelligence and image processing give a good solution of these problems, as it can monitor the room round-the-clock and detect the fall accident with low cost. Although a dataset is important for machine learning or deep learning based fall detection systems, few works construct a dataset with enough images. Moreover, the dataset in related works is taken by a sensor or camera at a low height; this is not very suitable for a fall detection system because the camera or sensor may be blocked by the furniture. In this study, images are taken by a depth camera, which is fixed at 1.7 meters, 1.9 meters and 2.1 meters height, as the height of roof should be no less than 2.1 meters according to the laws in many countries. This paper gives a brief dynamic camera height analysis and study how the camera height affects the detection accuracy.