A new algorithm for human detection in a static image is proposed in this paper. Human is seen as a model constructed by a torso and four limbs. The torso is fitted by a quadrangle, and each limb is fitted by one or two quadrangles. Detecting human is to search a combination of quadrangles constrained by some geometrical and topological relations. The first step is to detect and fit rectangles in the image, and then try to search some reasonable combinations of rectangles that should satisfy the geometrical and topological relations. If the probability of the combination is more than a fixed threshold, the vertex positions of the rectangles are adjusted to get a compact model of human in that image. At last, the probability of the compact model is recalculated. Experimental results show that this algorithm can be applied in the content-based image retrieval and the analysis of human motion based on video.