Kinematic Constraint Method for Human Gesture Recognition Based on Weighted Dynamic Time Warping

This paper presents a constraint method based on the kinematics to improve the accuracy of gesture recognition using a weighted dynamic time warping (DTW) algorithm. The traditional approaches of gesture recognition using 2D images are some limitations to detect the certain actions of human due to the lack of the full motion data. As the development of 3D depth sensors, it is possible for the gesture recognition technology to use the 3D motion data of human. In particular, the weighted DTW method is commonly used for the gesture recognition using 3D motion data since it can consider the sequential changes of the motion in a timeline. The weighted DTW, however, is an accuracy problem that is frequently identifying the false- positive and the true-negative about the complex gestures based on the timely stream of motions only. In this paper, we proposed the kinematic constraints generating the certain gestures so that we could improve the accuracy of gesture recognition using the weighted DTW algorithm. Finally, we carried out the experimental test with four types of gestures to evaluate the performance of the proposed kinematic constraints. The experimental results showed that the proposed method can enhance the precision of gesture detection compared to the standard weighted DTW algorithm even though the motion data captured by the 3D sensors include a lot of noises such as the occlusive movements and the ambiguous 3D measurements.