Applying Matching Coefficients in the 2D Stick-Model to Classify Human Motion: An Experimental Study in Applied Mathematics

Abstract

Abstract Recently, different techniques have been employed toward motion estimation. Some of these approaches include image based, model based, and silhouette based estimations. Despite their promising nature and outcomes, it remains notable that the techniques rely on motion data quality before producing optimal classification with precision and accuracy. Also, most of the existing algorithms have been complex relative to motion estimation, making interpretation challenging. Therefore, this study strived to respond to these dilemmas by modeling simple human motions through which various patterns of activity behavior could be recognized and aid in classification analyses. Three body components were used to develop the framework. These components included the lower body (LB), the upper body (UB), and the backbone (BB). Indeed, it was through these parts that a simple 2D human stick figure was formed. It is also notable that upon completion, the motion estimation mathematical model was compared to the performance of real motion phases to determine its efficiency in classification. The classifiers to which the model’s performance was compared included Rules and Tress, Misc, Meta, Function, Lazy, and Bayes classifiers. From the results, it was established that the 2D stick-model matching estimation was feasible and could be used to play a crucial rule in analyzing human motion classification.