Robust dynamic hand gestures recognition for human machine interactionusing multimodal features and manifold learning
Abstract
Interactions via human gestures have been widely used in many areas thanks to recentadvances of intelligent computing and smart devices. However, deploying existing hand gesturerecognition methods in practical applications require further investigations due to the diversity ofhand shape, appearance, movement, phase, sensor, etc. The ultimate goal of this project is topropose new hand gestures recognition algorithms that are able to easily handle variations of usersperforming gestures, viewpoints and different visual modalities of data (e.g. vision and depth). Theproposed algorithm allows their deployment in reality becoming more feasible and then havestrong impacts in many areas ranging from machine interaction, sign language, and medicalrehabilitation to virtual reality. To tackle this, we pay much focus on a powerful representationtechnique which is manifold. Manifold is a topological space which represents geometricalstructure of data with a lower number of dimensions than vector spaces. It seems to be the mostnatural way to discover the best multiple factor variations and interactions of the high dimensionaldata (e.g. video stream in our case) or even multi-modalities data. The research project outputswill be demonstrated through applications of controlling home appliances using dynamic handgestures in smart-homes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 09, 2018
- Source ID
- FA23861714056
Entities
People
- Thanh-Hai Tran
Organizations
- Air Force Office of Scientific Research
- Hanoi University of Science and Technology
- United States Air Force