Current techniques for detection of hand pose have high computational cost, and have accuracy and speed limitations which make them less than satisfactory for use in some high precision and time-sensitive applications. This disclosure describes the use of a combination of machine learning based pose estimation and optical flow processing techniques for hand gesture recognition, e.g., in AR and VR applications. Machine learning (ML) based pose estimation is utilized to identify hand pose and identify areas of interest that are processed further using optical flow processing. The use of optical flow processing enables higher precision in the detection of small movements as well as good low light performance. The use of ML based pose estimation reduces the solution space by reducing the size of an image that is subjected to optical flow processing. Such reduction of the solution space reduces the computational load and time required for optical flow processing.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Anonymous, "Hand Input Gesture Recognition Using ML-based Pose Estimation and Optical Flow Processing", Technical Disclosure Commons, (August 04, 2020)