Technology increasingly governs the globe, and machines are becoming increasingly sophisticated. To make living better, scientists are aiming to offer machines having the ability to understand the environment and smart skills development tools and techniques. Multiple techniques have been utilized to investigate event classification in videos and sequential images such as shape, height, or orientation of body segments along with their contextual influences. In this research article, a robust approach has been proposed for the human body 3D reconstruction and gait analysis. Initially, the crowd-based datasets are considered as the input, pre-processing steps are performed to reduce noise and computational costs. Then, human detection from given frame data, human 2D model, and human 3D reconstructions are implemented followed by features extraction using degree of freedom, local and global coordinate system, and motion direction flow (MDF) features. Finally, the classification stage with a data mining approach and classification through a genetic algorithm has been proposed. Two publicly available datasets are chosen for this research, namely, the mpii-video pose and pose track dataset. We attain the landmark recognition accuracy for the mpii-video pose dataset of 80.38% and gait analysis of 83.68%. For the pose track dataset, we attained a landmark recognition mean accuracy of 80.07% and a gait analysis mean accuracy of 81.81%. The proposed method represents a substantial enhancement associated with current state-of-the-art systems.