Research On Image Correction & Design On Embedded Object Tracker Based On Omni-direction Vision System

There are an increasing number of applications in Omnivision navigation systems for fisheye lens due to their ultra-wide-angle properties and cost-effectiveness. However, while fisheye lens provide very largewide-angle views, the images produce severe distortion as the hemispherical scene being projected onto a flat surface. This paper presents a novel methodology to correct image distorted by fisheye lens on embedded system. An embedded objects tracker based on Omni-directional vision system is provided to implement recognition and tracking for beacons landmarks. The main content in this paper is as follows:First, a projection model is proposed for estimating intrinsic/extrinsic parameters. The intrinsic matrix, which maps the camera coordinates to the image coordinates, is parameterized by principal points and skewness. Principal points and skewness are easily estimated using Ellipse fitting. The extrinsic matrix, which maps the world coordinates to the camera coordinates, is estimated by least squares fitting. The wide field-of-view (FOV) is the only prior parameter used in our technique.Second, the distorted images in the fisheye space can be corrected and transformed into the perspective space using the parameters we estimated. A lookup table is used to correct the distorted images in real-time.Third, a fisheye lens distortion correction system on a custom board that includes a DM642 and a Cyclone II EP2C20 is developed. DM642 is designed to be main chip that processes image data. And EP2C20 is used to control the clock signal and data bus. Solutions to the problem that JTAG interface was illegally designed are proposed.Last, a novel tracking algorithm, the analysis of gray level corner detection being used to locate beacon, is presented.