Research of visual tracking based on prior knowledge

Abstract

Despite the high maneuverability of Human's finger in videos, there are some principle in this kind of motion. A prior knowledge base is built on historical observation information .To solve the problem that Kalman filter responses not timely enough toward moving fingers in gesture videos, an adaptive acceleration extremum according to prior knowledge in current statistical (CS) model is introduced. On the other hand, taking advantage of interactive multi model (IMM) algorithm, the mixed models are used to make up for the inaccuracy of knowledge base when the motion pattern is unusual. Furthermore, motion termination forecast from prior knowledge base alters the model transition probability, boosting the speed of response in IMM. Simulations and practical engineering proves that the algorithm proposed by this article track efficiently in low quality videos whether the finger's trajectory is straight or tortuous.