Abstract

Hand tracking is one of most natural and unrestricted forms of interaction used in Augmented Reality (AR), thus extensive research has been conducted in this field. In this thesis, two marker-based hand tracking algorithms are selected and implemented. The first method is the combination of tracking a fiducial marker that is recognized by machine learning models and colored markers using template matching and Kalman filter. In the second method, a cloth glove with a distinct color pattern is used to estimate both the hand pose and depth from camera, this is achieved by labelling the captured image and finding a nearest neighbor in the database. The algorithms are then evaluated under five criteria: Lighting condition, rapid movement, accuracy of hand pose estimation, noisy background, and lastly, occlusions. Results show that the first method is relatively robust under different lighting conditions and temporary occlusions, while the advantages of the second method lie in rapid movements and accurate hand pose estimation. As a result, potential combinations of elements from both methods might address challenges identified in this thesis.

Results/Implementation/Project Description

Conclusion

[ PDF (optional) ] 

[ Slides Kickoff/Final (optional)]