Object Detection Using Local Contour and Global Structure Features

In this paper, we propose an approach of object part contour features combination associated with global spatial structure for object detection. The object part contour is described using object contour direction histogram method to model thevariant characters in different shape. Since the edge points are related to the shape information closely, the local shape can often be characterized rather well by the distribution of the edge directions. The Histogram of oriented Gradient (HoG) is used to represent object part. The spatial relations between object parts can be described by global spatial structure. A group ofobject partclassifiers are trained in order to detect the object parts. In detection process, we apply a generalized Hough voting scheme based on global spatial structure to generate object locations and scales. We evaluate the proposed approach on ETHZ object test set. By comparing our proposed method with the conventional method and our previous works, the experimental results show that the proposed approach is efficient and robust in object detection.