Unwarping of images taken by misaligned omnicameras without camera calibration by curved quadrilateral morphing using quadratic pattern classifiers

Abstract

A method for solving the problem of unwarping a distorted omni-image taken by a lateral-direction misaligned omni-camera with its optical axis incoincident with its mirror axis is proposed. The method does not conduct camera calibration and is based on a new concept of two-stage image mapping from the real-world space to the distorted image space. The first stage is conducted in the camera manufacturing process and includes the generation of a pano-mapping function for mapping the real-world space to an undistorted image taken by an omni-camera with its optical and mirror axes being coincident. The second stage is conducted in an in-field environment when the omni-camera becomes lateral-direction misaligned and includes the generation of a distortion-mapping function that maps undistorted image pixels to distorted ones and the generation of a misalignment adjustment table that combines the pano-mapping and distortion-mapping functions to map the real-world space to the distorted image space. The distortion mapping function is generated by a new technique of curved quadrilateral morphing using quadratic pattern classifiers. The misalignment adjustment table is last used to unwarp distorted images conveniently by table lookup. Experimental results using simulated and real-image data show the feasibility of the proposed method. (C) 2009 Society of Photo-Optical Instrumentation Engineers. [DOI: 10.1117/1.3204238]