Camera-Specific Simulation Method of Fish-Eye Image

Abstract

Simulation tools are widely used in intelligent vehicle systems and robot auto-control systems. They considerably contribute in the cost reduction of the validation and testing stages of the new intelligent functionalities. As one of the common parts of these systems, visual sensors need to be simulated physically, in order to enable the implementation of the computer vision-based perception techniques in the virtual environment. Fish-eye cameras are often used in this techniques to provide an omnidirectional field of view, and thus need to be simulated. The existing methods cannot simulate the output image of specific real fish-eye cameras. In this paper, we present a post-processing fisheye camera simulation method. According to the stereographic projection model, a mapping is established between the output image and a cube-map rendered with the graphic engine. In order to simulate the output of a specific camera, the distortion of the real camera is measured and added to the simulated image. Experimental result shows that our simulation method can give output close enough to the image captured by a specific fish-eye camera. The practicality of our method is validated in an actual application.