Proceedings of The 7th International Conference on Knowledge and Innovation in Engineering, Science and Technology
Year: 2020
DOI:
Fusion of Infrared Image and Visible Image for Fall Detection Base on Discriminant Feature
Zong-Huei, Chen, Mei-Yung, Chen
ABSTRACT:
Fall detection is a very important application in surveillance systems. Therefore, it is very important to build a complete fall detection system. However, this is not suitable for environments with weak light sources since the most of the detection lenses are in a sufficient light source. Thus, in this paper, we introduce a fall-detection system based on the fusion image. It can not only solve the problem that the human cannot be detected under a dim background, but also locate the human more accurately. We use an open source deep convolutional neural network (CNN)-based approach named OpenPose to extract the discriminant features which let us to build the human centerline. Since OpenPose will detect and focus on the human activity, the other heat sources cannot be detected. In other words, OpenPose can help us remove the things that we do not interested. Finally, we define the human falls by detecting the variation of centerline. The experimental result shows that the proposed system can be applied in both bright and dark environments. Besides, the detection of the fall has 99% accuracy, higher than the other methods which use the visible image and the infrared image. The accuracy can guarantee the performance.
Keywords: fall detection, Image fusion, OpenPose, discriminant feature.