An object detection method based on topological structure

Due to the influence of both background interference and time-varying position, it is difficult to detect the moving object by computer vision in the complex background environment. This paper novelly builds a topological structure of typical features to detect object. The presented method can effectively detect the object which is scaling, rotating or in affine transformation, because of the using of geometric properties of the topological structure. Firstly, the feature points are extracted and clustered, then the supposed region which includes the object will be obtained from the image. Secondly, the HOG-SVM classifier is used to detect typical local features in the supposed region. Lastly, the typical local features are regarded as nodes to build a topological structure based on the object´s contour, then the constraints of the geometric properties are used to figure out the object. The experimental results indicate that the proposed method can detect the object when it is in different postures or a part of the object was sheltered.