Enhanced 3D Action Recognition Based on Deep Neural Network
Sung‐Joo Park, Dongchil Kim · 2022 Thirteenth International Conference on Ubiquitous and Future Networks (ICUFN) · 2022
In the video surveillance system operating in the real environment, the recognition of detailed behavior of objects has important meaning in terms of understanding whether security events have occurred. In particular, the perception of behavior in a poor environment such as low light and overlapped objects is recognized as an important technical factor that must be overcome in the existing 2D image-based surveillance system. Action recognition of objects using 3D depth map provides a way to solve these issues. In this paper, we propose the enhanced 3D action recognition method based on convolution neural network (CNN) for the video surveillance system. And we evaluated the action recognition performance using the real environment DB, and the recognition result for 6 detailed behaviors was confirmed to be an average of 68.56%.