Hand Tracking Based on the Kinect and Kalman Filter
Liwei Yang, Meiling Wang, Tao Li · 2020
Gesture Recognition (GR) is one of the hot issues in the field of computer vision. Gestures are generally divided into static and dynamic gestures. The recognition of static gestures mainly lies in the extraction and matching of hand shape features, whereas the key feature of dynamic gestures is hand trajectory. Therefore, it is necessary to track the hand to obtain the temporal and spatial information. In this paper, we use the Microsoft Kinect, with the function of recognizing and positioning human joints, to detect the hand position from the captured body image sequence. However, in practice, hand detection results of the Kinect are sometimes biased, leading to a significant impact on the feature description of gesture trajectory. Therefore, the Kalman filter is introduced to correct the hand position returned from the Kinect. The experimental results show that the hand trajectory curves obtained by the proposed tracking algorithm are more smooth and stable.