Real time 3D Hand Gesture Recognition by Weighted Depth Difference Motion History Image in Networked HCI
Haoyang Luo, Haikuan Wang, Wenju Zhou, Kangli Liu, Jinqi Fu · 2021
Hand gesture recognition applied in human-computer interaction (HCI) is a useful but challenging task because of the low inter-class and high intra-class variability. It is difficult to improve real-time performance while ensuring accuracy. Since simple but robust features are crucial for the performance of recognition, we proposed the weighted depth difference motion history image (WDD-MHI) to extract spatial-temporal information. Subsequently, a weighted random forest classifier with second training is employed for action classification. The developed method is shown to be computationally efficient allowing it to run in real-time. We designed a multimedia video-wall control system based on online gesture recognition, the recognition results indicate competitive performance of our method over the other methods.