Lower Limb Action Recognition with Motion Data of a Human Joint

Abstract

Human action recognition technology has been an international research topic for the past 20 years owing to its range of potential applications. To realize fast and accurate recognition of actions of human lower limbs, we propose an action recognition method for human lower limbs based on the dynamic Bayesian network (DBN) model. First, the hip joint was chosen as the recognition object, and then, its y coordinates were chosen to achieve motion information. Second, the coordinate, velocity, and acceleration information of the hip joint in the y direction were filtered based on wavelet transform and weak tracking Kalman filter. Third, the filtered coordinate and velocity were used to calculate human action characteristics based on wavelet transform, and K-means was introduced to mark the action characteristics quickly. Finally, an action recognition method based on the DBN model was introduced to realize the judgment of the actions of human lower limbs. The proposed action recognition method for human lower limbs only needs the motion information of a human joint and has a fast calculation speed. Experimental results proved the high recognition rate and good application prospect of the method.