Robust and customized methods for real-time hand gesture recognition under object-occlusion
This addresses a domain-specific problem for applications requiring robust hand tracking in occluded environments, but it is incremental as it builds on existing methods.
The paper tackles the problem of real-time hand gesture recognition under object occlusion by proposing a system that models hand keypoint angles, uses GANs for data generation, and applies time series forecasting and sliding window matching. It achieves state-of-the-art results on 11 typical gestures.
Dynamic hand tracking and gesture recognition is a hard task since there are many joints on the fingers and each joint owns many degrees of freedom. Besides, object occlusion is also a thorny issue in finger tracking and posture recognition. Therefore, we propose a robust and customized system for realtime hand tracking and gesture recognition under occlusion environment. First, we model the angles between hand keypoints and encode their relative coordinate vectors, then we introduce GAN to generate raw discrete sequence dataset. Secondly we propose a time series forecasting method in the prediction of defined hand keypoint location. Finally, we define a sliding window matching method to complete gesture recognition. We analyze 11 kinds of typical gestures and show how to perform gesture recognition with the proposed method. Our work can reach state of the art results and contribute to build a framework to implement customized gesture recognition task.