Camera calibration
This tool provides Camera Calibration from a single image based on vanishing points. The details of the technique can be found in the paper [1]. Note that this tool does not include the human tracking algorithm introduced in the paper. It assumes that the following information is available.- Parallel lines in three orthogonal directions OR the equation of the Horizon line OR more directly, the image coordinates of three vanishing points.
- The head and feet position of a reference object (e.g. a person) with known height.
Three steps to calibrate a camera from a single image.
- Step 1: Open the image for calibration and run this tool. A dialog will show up as follows.
An ideal image for calibration using this technique is the one with some parallel lines in three orthogonal directions in the scene, such as the image shown here.
- Click the "Pick" button and pick one line on the image at a time. Lines in the same group should be parallel and lines from different groups should be orthogonal to each other in the real world. If you know the image coordinates of two end points of a line, you can input this information directly into the edit box. Press CTRL+Enter to add one more line. Lines in x, y and z direction are marked in red, green and blue, respectively. See the example here.
- If parallel lines in three orthogonal directions are not available but you know where the Horizon line (see the paper for details) is, you can still use this tool. Either provide the line equation or pick the line on the image. The Horizon line is marked in magenta.
- If you know the image coordinates of three vanishing points (for example, from other program), you can provide this information directly to VIPBase.
You need to provide the head and feet position of a reference object (e.g. a person). You can either input this information directly in the edit boxes or pick the line on the image and it will be marked in yellow. See the above example. You also need to specify the height (in millimeters, centimeters, inches or meters) of the object.
Click the "Step 1" button to finalize the first step.
- Click the "Pick" button and pick one line on the image at a time. Lines in the same group should be parallel and lines from different groups should be orthogonal to each other in the real world. If you know the image coordinates of two end points of a line, you can input this information directly into the edit box. Press CTRL+Enter to add one more line. Lines in x, y and z direction are marked in red, green and blue, respectively. See the example here.
- Step 2: Click the "Step 2" button to compute the vanishing points and principal point. The result will be shown as follows.
- Step 3: Click the "Step 3" button to compute other camera parameters. The result will be shown at the bottom of the dialog.
You can save and loading all the settings in a text file.
You can visually evaluate the calibration result by generating images from virtual cameras. If you have points with known 3D coordinates, you can use the calibration result to re-project those points onto 2D to find out how close they are from their imaged points. See details here.
Reference
[1] Fengjun Lv, Tao Zhao, Ramakant Nevatia, "Camera Calibration from Video of a Walking Human", IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI), vol. 28, no. 9, pp. 1513-1518, Sept., 2006.
 |
VIPBase © 2006 Fengjun Lv |
Last update: 08/01/2006 (check html) |