Edge Extraction And3D Reconstruction Of Catadioptric Image

With the increasingly demand for visual system with large visual field, the traditionalimaging system is unable to meet the demand of the people. On the other hand,catadioptric omnidirectional imaging system with its advantages of large field is gettingmore and more attention. But, to obtain3D information of large field through catadioptricomnidirectional imaging system, there is still some inadequate, such as complexity of thesystem and the large amount of computation, so further research is needed.In this paper we introduces a sparse3D reconstruction method based on a singlecatadioptric image and its edge feature points. It contains two parts: the first part is theintroduction of the hardware system, which is used to gather the single catadioptric image;the other part is the algorithm research of sparse3D reconstruction based on the singlecatadioptric image and its edge feature points.The hardware system is composed by two hyperboloid catadioptric mirrors which arecoaxial up and down and a camera which is used to gather images. The advantages of thisstructure is that there is less occlusion in the system. Besides, the uniformity of quality ofthe two sets of image information is ensured. What is more, because of the singleviewpoint constraint of the system, corresponding points locates the same line which startsthe center point of the image, so the matching of the corresponding points is simplified.The main steps of3D reconstruction algorithm in this paper are: image unwrapping,edge extraction, stereo matching and3D calculation. The method of edge extraction is animproved algorithm based on the traditional Canny edge extraction algorithm, by addingtwo diagonal gradient information, the resulting edge image is more complete andcoherent. The method of stereo matching is based on the edge pixels and theirneighborhood pixels, in order to choose the neighborhood pixels, a neighborhoodmatching matrix with24neighborhood pixels is designed, through this method, a dense3D point cloud reconstruction obtained.In this paper, by studying the catadioptric system and the algorithm of3Dreconstruction, a kind of sparse3D reconstruction method based on the edge features of a single catadioptric omnidirectional image is proposed. By doing experiments, a goodresult is obtained, this is very helpful for obstacle recognition of the mobile robot,real-time navigation and path planning.