关键词: 轮廓跟踪，水平集，卡尔曼滤波，外接矩形框

Abstract: Aimed at the characteristics of low deviation between targets and background and weak boundaries, a novel approach of fast convergence and strong ability to capture boundaries was presented based on level set method. We used index function with quicker convergence as indicator function and normalized Gauss distribution function to improve tra- ditional Dirac function. In the tracking process of the moving target, this article obtained minimum circumscribed rectan- gular frames of human moving target in every video frame with the kalman filter method, and used the method of the level set curve evolution to obtain moving human target outline eventually. Experiments on targets under the visible light and infrared moving video sectuences show this method can greatly improve the tracking speed compared with tra- ditional methods and has better results in target-tracking and contour extraction for infrared images with weak bounda- ries and huge convex-and-concave features.

Key words: Contour extraction, Level set, Kalman filter, Circumscribed rectangular frames