A novel uncalibrated human hand tracking based on transition model

In this paper we deal with the tracking problem of human hand. The algorithm consists of two steps: At first the move part is detected and recognized with a advanced DOF model; then the curve tracking analysis is applied to get the key features with more details. The camera based on computer vision has the potential to provide much more natural, non-contact tracking. Although move part such as arms has a more faster speed, it can also been located exactly. With the semantic breakpoints extracted, some active part in a complex background can be location at any time in a test period.