Real-time and Robust Grasping Detection (bibtex)
by Chen, Chih-Fan, Spicer, Ryan, Yahata, Rhys, Bolas, Mark and Suma, Evan
Abstract:
Depth-based gesture cameras provide a promising and novel way to interface with computers. Nevertheless, this type of interaction remains challenging due to the complexity of finger interactions and the under large viewpoint variations. Existing middleware such as Intel Perceptual Computing SDK (PCSDK) or SoftKinetic IISU can provide abundant hand tracking and gesture information. However, the data is too noisy (Fig. 1, left) for consistent and reliable use in our application. In this work, we present a filtering approach that combines several features from PCSDK to achieve more stable hand openness and supports grasping interactions in virtual environments. Support vector machine (SVM), a machine learning method, is used to achieve better accuracy in a single frame, and Markov Random Field (MRF), a probability theory, is used to stabilize and smooth the sequential output. Our experimental results verify the effectiveness and the robustness of our method.
Reference:
Real-time and Robust Grasping Detection (Chen, Chih-Fan, Spicer, Ryan, Yahata, Rhys, Bolas, Mark and Suma, Evan), In Proceedings of the 2nd ACM symposium on Spatial user interaction, ACM, 2014.
Bibtex Entry:
@inproceedings{chen_real-time_2014,
	address = {Honolulu, HI},
	title = {Real-time and {Robust} {Grasping} {Detection}},
	url = {http://ict.usc.edu/pubs/Real-Time%20and%20Robust%20Grasping%20Detection.pdf},
	abstract = {Depth-based gesture cameras provide a promising and novel way to interface with computers. Nevertheless, this type of interaction remains challenging due to the complexity of finger interactions and the under large viewpoint variations. Existing middleware such as Intel Perceptual Computing SDK (PCSDK) or SoftKinetic IISU can provide abundant hand tracking and gesture information. However, the data is too noisy (Fig. 1, left) for consistent and reliable use in our application. In this work, we present a filtering approach that combines several features from PCSDK to achieve more stable hand openness and supports grasping interactions in virtual environments. Support vector machine (SVM), a machine learning method, is used to achieve better accuracy in a single frame, and Markov Random Field (MRF), a probability theory, is used to stabilize and smooth the sequential output. Our experimental results verify the effectiveness and the robustness of our method.},
	booktitle = {Proceedings of the 2nd {ACM} symposium on {Spatial} user interaction},
	publisher = {ACM},
	author = {Chen, Chih-Fan and Spicer, Ryan and Yahata, Rhys and Bolas, Mark and Suma, Evan},
	month = oct,
	year = {2014},
	keywords = {MxR},
	pages = {159--159}
}
Powered by bibtexbrowser