Advanced Gait Event Recognition and Pose Estimation Model through Deep Learning

Analyzing activities of humans, events, Gait, human-to-human interaction, and sports analysis are some prominent subjects, and key areas of interest in machine learning, artificial intelligence, and deep learning fields. Nevertheless, beginning with intelligent computing in a somewhat less complex context for enhanced scene perception remains a topic that still waits for scholars’ exploration. In this system, we have made a model to allow machines to understand the behaviors in the external environment by proposing a novel methodology to discern sports and events. This research aims to incorporate AI functions in information technologies to innovative future trajectories of intelligent media. Our proposed system recognizes complex human activities from two well-known Kinetics and YouTube datasets. After preprocessing and human detection, we have used numerous feature algorithms, such as DOF and 3D cartesian plane for joint points while distance transform and optical flow for full body points. These feature point values are then fed to joint feature descriptors and PSO for enhancement before passing to the CNN for recognition. However, the obtained data prove that the suggested approach is effective in realistic scenarios with a high level of uncertainty with an accuracy of 91.2% and 88.2%, on Kinetics and YouTube datasets respectively.