Shape is a kind of important feature used to detect or recognize objects in computer vision.Edge is the straightest way to describe the shape of an object,so the description of shape based on the edge is effective.At present,some disadvantages still exist in many methods based on shape feature,such as having no local property or being easily affected by rotation and the scale change.The approach described in this paper is based on matching simple shape features of scene and model with a technique called HYPER（HYpotheses Predicted and Evaluated Recursively）.The description of shape feature is obtained by polygonal approximation,which is local and compact.This approach is to provide strong robustness to translation,rotation and scale changes,simultaneously evaluating the important parameters about movement including the location of object.In addition,the approach is robust to partial occlusions due to shadows,touching and overlapping object in some complicated environments.The experimental results demonstrate its effectiveness and merits.