Action Recognition System with the Microsoft KinectV2 Using a Hidden Markov Model

This paper proposes the independent human action recognition for more practical use, using the depth image and the Joint Points from the Microsoft KinectV2. The threshold value not only for the binarization of the depth image but also for the others except for the circularity was chosen automatically. Moreover, the value of the background is not even value but some depth value. In my previous research, the background image was simple so that the foreground area could be easily extracted as the binary image from the depth image by the threshold value chosen automatically. The whole human silhouette is difficult to be get with the uneven background. This problem can be overcome by subtraction between the two depth information in the current image and the background image.