Article: Learning hand latent features for unsupervised 3D hand pose estimation

Published on: 2019-06-18 | Updated on: 2019-09-02

Journal of Autonomous Intelligence

Summary:Method LDTM can accurately estimate a hand pose based on the prior knowledge of the hand representation.

Hand pose estimation from depth is the first step for several human-computerinteraction applications. It has been widely applied to human-machine interaction(HMI) since it provides the possibility for future multi-touchless interfaces. Anaccurate hand pose estimation provides a natural way of interaction between humanand virtual space that achieves greater user experience. Different from theconventional human-machine interactions which are limited to 2D plane display, andwhich are only suited where users sit behind the computing devices, hand poseestimation offers 3D user interaction without direct contact with the computing device.This provides a possibility for the new interface leading towards seamlesshuman-computer interaction.

However, hand pose estimation is still a difficult task owing to some challengesthat a human hand possesses. The hand is very dextrous, has many degrees offreedom. Similarly, fingers have high self-similarities and severe self-occlusion.The input depth image is accompanied by the large amount of noise which willprobablymislead the poseestimator and distort the output results.

Jamal Firmat Banzi, Isack Bulugu and Zhongfu Ye conducted an research on the learning hand latent features for unsupervised 3D hand pose estimation.The survey result published on the journal of autonomous intelligence.

In this paper, the team  present a novel approach to model hand topology based on LDTM which captures hand latent features and observable features to construct hand joint representation. Then they integrate LDTM with the deep PCM using a data-independent method to encode the hand representations and map with the decoded hand depth map. Finally, the multi-layered convolutional neural network based on deep PCM was utilized to regress a 3D pose space based on the joints location.

As a result, their system can accurately estimate a hand pose based on the prior knowledge of the hand representation. This confers robust and reliable hand pose estimation system that can achieve greater user experience.

For more information, please visit:

https://en.front-sci.com/index.php/JAI/article/view/36