Authors: Cristina Manresa et al.
Year: 2000
- I came across this paper when I searched for previous work on using convexity defects for hand tracking. They didn't use convexity defects for tracking fingertips though. Instead, they use it for gesture recognition. As convexity defects can characterize the state of the hand, I'm also thinking of using the convexity defects as one of the features for gesture recognition.
- Hand tracking
- Finding correspondence between between the detected blobs and the tracked hands is an important and necessary step.
- What they did is a bit similar to Kalman filter, but less rigorous and formalized.
- The gesture recognition method they describe is very application specific, and involves a lot of threshold values.
- I think what PrimeSense does in their NITE library for gesture recognition is also similar to this.
- Basically you define some gestures you want to recognize and observe their characteristics.
- Extract and compute features that describe those characteristics and compare with threshold values.
- This is more accurate probably we are actually incorporating more human intelligence in it.
- Although it may not scale to other gestures, in this restricted domain, it can perform well.
No comments :
Post a Comment