Video based human activities recognition using deep learning.

Human activities recognition from motion capture data is a challenging problem in the computer vision due to the fact that, in various human activities, different body components have distinctive characteristics in terms of the moving pattern. In this paper, a learning method of detecting an activity from different angles based on various sources of information is proposed. with high accuracy. The bottom up approach is used in OpenPose which is the tool used in this paper's experiments. The proposed method achieves promising results on Berkeley Multimodal Human Action Database (MHAD) datasets at 98% accuracy. [ABSTRACT FROM AUTHOR]