A Motion Detection Model Based On Biological Vision

The brain mechanism of biological vision is much complex and the motion perception is one of the important part of it. Based on the established motion perception model of biological vision, it is the goal of this paper to build a dynamics neural model which can be used to detect the motion direction and motion figure in the real scene of the moving target.The visual cortex of human brain related to motion perception as we know so far involves area V1, V2, V3, MT and MST. Especially the path of V1-V2 is for static figure processing and the path of MT-MST is for motion processing. At present, motion boundary contour system and Formotion BCS are two of the established motion perception model based on biological vision, and the motion perception effects are good to use them in the structured scene. But the perception effects to use MBCS in the real scene are worse, and Formotion BCS laid a strong emphasis on the analysis of vision psychology phenomenon and could not to be used to detect the motion figure and motion direction in the real scene directly. Based on analyzing those motion perception model in detail, a dynamics neural model which is used to detect the motion figure and motion direction in real scene has been built thought combined their advantage in this paper.It is the idea of this model that feature signals in each preferred direction of the motion target, such as corners, line terminators or line intersections are extracted at first, and then, via the boundary regrouping and segmentation function of the motion bipole cells that synthesize static boundary signals and feature signals, the final motion boundary is output.The model includes the following steps: static boundary processing, transient network processing, space filter processing, competition network processing, inhibition network processing and boundary regrouping and segmentation (motion bipole cells). Feature signals of the motion target can be gained through the combination of transient network, space filter, competition network and inhibition networks, which constitute the processing stages to confirm feature-tracking signals, and the static boundary preprocessing provides the static boundary information for the transient network and the motion bipole cells.The dynamic image sequence in the structured and real scene have been simulated successfully by using the model in this paper, and the factors that influenced the perception effects have been analyzed in detail also. Simulation results indicated that this model could be used to detect the motion figure and motion direction of the moving target in real scene as well as in the structured scene.