Huang Chao, Wang Weilin, and  Wu Xian. Human Action Recognition Method Based on Dynamic Bayesian Network. Dynamic Systems and Applications 29 (2020) No. 5, 1988 – 1996

https://doi.org/10.46719/dsa202029532

ABSTRACT.
The recognizing human activities from video series events or still, images can be challenging, such as background noise, partial disturbance, scale changes, overview, lighting, and appearance issues. It is required in multiple functional authentication systems, including various applications, video surveillance systems, human-computer communication, and robots that characterize human behavior. The existing system does not provide action recognition accurately and involves many activities and concerns in more complex situations. During the computer learning phase, the video function uses a Hierarchical Action Semantic Dictionary (HASD), manually extracted using the first in-depth neural network, using hierarchical clustering. Modern technologies need to know about human functions. The proposed technique with the Dynamic Bayesian Network (DBN) is used. DBN used to guide different types of learning domain knowledge processes in the form of first-order probabilistic logic (FOPLs). FOPLs models are pre-qualified into two kinds, and they are used to lead system and parameter control. Our proposed learning algorithm is subject to the Constrained Structural EM (CSEM), which pre-training integrates the sample system. The common problem has been limited by the lack of adequate training data for our policy’s successful operational recognition. As a result, it becomes an effective detection, monitoring, and recognition system for real-time human movement. The Accuracy of the action recognition in video is simulated along with the various videos.

KEYWORDS: Human action recognition, maximizing likelihood learning by FOPL, Dynamic Bayesian Network