High-resolution stereo video rectification through a cost-efficient real-time GPU implementation using intrinsic and extrinsic camera parameters

Abstract

This paper describes a method to correct the distortions in the images captured by a stereo video camera rig. We developed a code which can be implemented both on CPUs and GPUs which are used respectively to process still images or videos. The processing scheme is centered on the pre-calculation of a deformation texture, based on the physical properties of the camera setup (intrinsic parameters). The pre-calculated texture contains deformation parameters for each pixel. The GPU implementation of our method is cost-efficient in terms of computer resources and can therefore be used to rectify high-resolution image pairs in real-time. Moreover, due to the pre-computation step, any camera model of arbitrary complexity can be used without impacting the execution frame rate.