Target Tracking Based On Fusion Of Un-congeneric Information In Clutter

In the systems of human-computer interaction, smart classroom, video conferencing and specific environmental monitoring, etc., how to utilize natural spatial and temporal correlation to achieve effective fusion of vision information and auditory information and to achieve tracking, recognition and decision. This is the hot topic of dissimilar information fusion, which is usually referred to as Multimedia Information Fusion.Basic theories and algorithms about information fusion, visual tracking, sound source location and filter performance have been discussed in this thesis. And on this basis, two kind of new algorithms about people tracking based on vision and auditory information have been proposed. One is utilized multi-level Kalman filter as fusion tool to establish the model of Audio-visual information fusion in characteristic level, and another is utilized improved ISPF(importance sampling particle filter)as fusion tool to establish the model of Audio-visual information fusion in decision-making level.In the model of Audio-visual information fusion in characteristic level, firstly, separate treatment is necessary for vision and auditory information collected. Image speed of target is accessed by machine vision, and then the actual velocity of target is accessed by the technology based on the delay of sound source position. The ratio of two types of speed is considered as a variable. Keep updating this variable and give feedback to vision tracking system to get optimal estimation of the target.In the model of Audio-visual information fusion in decision-making level, Use the algorithm simplified by ISPF,Changes the problem of designing the importance of function into the problem of selecting the importance of particle, template matching algorithm based on color information is used as bottom tracking, and the stable results of audio information is considered as center particle with special weight. The result of vision tracking can be restricted and guided through this center particle.Finally, the feasibility of the model of Audio-visual fusion is validated through experiment, the tracking algorithm based on Audio-visual information fusion is robust for the clutter interference of light and background changes, person-aliasing, etc. Compared this two fusion strategy and analyze the simulation results, it can be determined that the model of Audio-visual information fusion in decision-making level has more superiority.