Study On Self-calibration Method For Non-central Catadioptric Cameras

Non-central catadioptric camera system has been more and more popular due to its unique merits, which includes great flexibility of designing, low cost of building, easy and simple for imaging and excellent real-time characteristic. It is able to capture 360 degree field of view, making up for the limited field of view of the traditional camera and the difficulty in precisely calibrating the single viewpoint catadioptric camera system. With the rapid development of large-scale CCD imaging technology and image processing technology, non-central catadioptric system has broad application prospects in many areas such as mobile robot navigation, tele-conference, video surveillance, virtual reality and external space exploration. Although calibration is a preliminary step in most applications. few calibration methods have been proposed for non-central catadioptric camera systems, moreover the existing methods are complicated to realize. Therefore, it's urgent to find a novel calibration method which is convenient to carry out and has high accuracy for non-central catadioptric camera.This paper presents a novel self-calibration method for non-central catadioptric camera system firstly. Without the aid from any special calibration patterns in the scene, the proposed method is able to estimate the rotation and translation parameters between the mirror and the camera using only the camera system itself. First it uses the ellipse corresponding to the mirror boundary in the image to compute the possible mirror postures. With four possible pose solution candidates thereafter, we propose a novel selection method to find the correct one by using the image of lens boundary. Experimental results show the effectiveness of our method.To further improve the calibration accuracy and execution speed of our algorithm, we then proposed an improved ellipse detection and fitting algorithm based on RANSAC and an improved mirror posture selection method. This improves the robustness and accuracy of ellipse fitting, and meanwhile we can obtain the calibration results much faster and more accurately. The algorithm is simple to carry out, has high precision and good anti-noise robustness. Finally, we carried out experiments on image transformation and 3D reconstruction with our calibration results. The experimental results shows that our proposed algorithm accurate and effective, and therefore is suitable for non-central catadioptric camera system's calibration.