A New Corner Detection Method for Omnidirectional Images

Recently, many robotic applications adopted omnidirectional vision sensors. The achievement of those applications is usually based on computer vision algorithms by means of visual indices extraction from the acquired images, such as edge, region, and interest points detection. The use of classical image processing algorithms for omnidirectional images proved its limitations. Some new methods which take into account the special geometrical properties of omnidirectional images have been developed and used with success. Along these lines, this paper proposes a new corner detector more adequate for this class of images. The method is based on a virtual spherical electrostatic model for edge detection and the standard Harris corner detector. Experimental results and quantitative assessments affirm the performance of the proposed approach against several image degradations. Confrontation to other classical methods is provided.