Camera calibration is one of the most important and fundamental problems in computer vision. The estimated parameters by the previous methods, however, are not accurate and stable enough due to the errors of control points localization arising from the lens distortion and perspective distortion. A halation and an occlusion due to illumination and an obstacle such as a human hand also result in large errors in control point localization step. The previous method can not deal with the errors of control points localization from these factors.
In our software package, we use circular control points (circle grid and ring grid pattern) which can be localized more precise than square, and also adopt an iterative refinement approach of control points. Moreover, we estimate the uncertainty of control point localization and optimize the camera parameters using the weighted bundle adjustment iteratively. Our method can obtain camera parameters precisely and stably by using ring grid pattern, iterative refinement of control points, and iterative weighted bundle adjustment. We have conducted an extensive set of experiments with real and synthetic images for the square, circle and ring grid pattern. The pixel re-projection errors obtained by our software package are about 55.7% lower than those of the OpenCV Camera Calibration Toolbox using square grid pattern and 27.3% lower than those of the OpenCV Camera Calibration Tool- box using circle grid pattern.

Flow Chart

Accuracy of Control Point Localization (Step 5)
Datasets
Experimental Results
Visual Hull Reconstruction Result: ant
Ground Truth (video)	OpenCV (video)
Ring (video)	Circle (video)
Visual Hull Reconstruction Result: seaweed
Ground Truth (video)	OpenCV (video)
Ring (video)	Circle (video)

Download

Matlab source code for Precise Camera Calibration

Sample dataset for Precise Camera Calibration Part 1 (Real data)

Sample dataset for Precise Camera Calibration Part 2 (Synthetic data)