Hand Detection and Tracking in an Active Vision Environment

Abstract

The combination of computer vision systems and humans operations has inter-related more closely, this relationship driving computer vision research community to think and come up with more and more interac- tive applications with human .This research proposing set of novel broad range of solutions for the visual based hand tracking. Visual based hand detection can be cheap compare to other sensor based hand tracking and we are using a economical camera to detect hand in real time to provide the information regarding hand posture .Expensive systems may be little bit more accurate compare to vision based systems but it's a compromised between cost , usability and accuracy. Vision based tracking systems are de nitely cheaper and easy to use than those highly equipped unusable systems. We presenting two comprehensive reliable approaches to visual based hand tracking where hand posture can be estimated in real time . Visual based hand tracking considered as two major approaches marker based hand tracking and Skin color based hand tracking. Marker based hand tracking we have provide a solution with color ltering in RGB color space and using contour nding technique that nd the markers in a bi- nary image then we applied some morphological operation to achieve the best accurate result . Another procedure is to track the markers in a video scene is feature matching that detect marker in 3D environment . Both of our approaches are been proposed in our research is capable of track the hand in real time with excellent reliability. Camshift algorithm used as motion detection algorithm and image skeletonization been used to capture the hand Skelton , in terms of geometry analysis hough lines technique used to obtain hand parameters. Condensation algorithm was another procedure where we have presented in this research as a motion based hand tacking approach. Probability distribution of the hand pixels is identi ed and to obtained the hand parameters we proposing a math- ematical model that model the hand motion in 2D Cartesian space. We carried out performance evaluation for two approaches with evaluating procedure address that which approach is reliable under which conditions and carried out a accuracy evaluation for marker based tracking .