Aspect Ratio and Calibration

Note: If you are only interested in what to do about the aspect ratio problem, you can skip this introductory material.

Since the early days of television, the American television standard (NTSC) has specified an image that is about 525 lines high and 64 microseconds wide. Note the difference in units! This is because TV images are defined to be scanned rather than static. Part of the image is blanked out, to give the scanning circuits time to prepare for scanning the next line. Of the remaining image, part is expected to lie beyond the edges of the picture screen, invisible to the viewer. What is left, the part called the "clean aperture," is the only part that is supposed to be seen by the viewer. This is the image that forms a picture with an aspect ratio (i.e., ratio of width to height) of 4:3. In other words, if a TV camera is properly aimed at a physical object that is 4 meters wide and 3 meters high in the object plane, and the camera lens is zoomed so the the image of that object is 480 scan lines high, then the image of that object would precisely fill the clean aperture.

To the computer industry, a computer monitor is thought to consist of square picture elements (pixels), even if it is actually a scanned cathode-ray tube. The height and width of an image on a computer monitor is therefore specified in terms of the number of pixels of height and width. The "pixel aspect ratio" of the image is the ratio of width in pixels to the height in pixels. If a photograph is digitized and displayed on a computer monitor, then it will appear undistorted as long as its pixel aspect ratio equals its original aspect ratio. With this in mind, companies who design video capture cards make them so the 52 microsecond line width of the clean aperture in a TV signal is digitized into 640 pixels, giving a pixel aspect ratio of 640:480, or 4:3. A video file captured with such a card can be opened in VideoPoint, calibrated with a known length in any direction and analyzed correctly.

Unfortunately, the television industry sees things differently. The Society of Motion Picture and Television Engineers (SMPTE) published a standard that defined a digitzed television image to have a width of 720 pixels and a and height of 486 pixels (the most important part of this standard for our purposes is "Center, Aspect Ratio and Blanking of Video Images, RP 187." SMPTE Journal, August 1995, page 570.). The width of 720 was chosen to be compatible with both the American NTSC standard and the European PAL standard. The height of 486 was chosen to be slightly larger than the clean aperture so there would be room for digital edge artifacts to go unnoticed by the viewer. For the same reason, the width of 720 was defined to include slightly more than the clean aperture.

The DV consortium defined a standard that is used in all consumer equipment carrying the DV logo. The DV standard must differ somewhat from the SMPTE one, since it calls for a height of 480 pixels rather than 486. However, the DV standard was not published in any journal that that I could find on library shelves, so I do not have a reference to the defined size of the clean aperture. Nevertheless, by analogy with the SMPTE standard and in view of the measurements described below, it is clear that 720x480 pixels is NOT the size of the clean aperture. According to an interesting article by Chris Pirazzi, you should convert to 654x480 pixels.

I used two different camcorders (a Canon Optura and a Sony VX-700) to photograph a scene with several vertical and several horizontal meter sticks. After transferring the DV output of the camcorders to a computer, I opened each file in VideoPoint, calibrated using one meter stick and then measured the length of a perpendicular meter stick. I did this for several sets of meter sticks and used the average ratio of these measurements to see what factor was needed to resize the horizontal dimension to yeild equal-length meter sticks on the final image. For both camcorders, the result was that a width of about 658 pixels worked. I resized each file from 720x480 to 658x480 and found that calibration in either dimension led to correct measurements in both dimensions.

However, recently some RIT students calibrated several newer Sony DV camcorders. They found that resizing the files from 720x480 to 640x480 yeilded correct calibrations in both dimensions. Apparently Sony has changed the way it manufactures camcorders.

 Whatto do before making measurements with DV files If you are only making vertical measurements, use a vertical calibration object. If vertical measurements are more important than horizontal ones (this could be the case for projectile motion if you need to measure g), use a vertical calibration object. If you need to make accurate measurements in both dimensions, check the calibration for your camera! To do this: Make a short video of two metersticks at right angles to each other. Set up the shot carefully so the metersticks are in a plane perpendicular to the camera's optical axis, and the camera is aimed at the centers of the metersticks. If you are using VideoPoint Capture or LoggerPro, capture and save the movie with the default settings. Otherwise, capture the video in its native DV format, which has a pixel aspect ratio of 720x480. When you save the file, recompress it in another codec (H.264 or Sorenson 3, for example) and simultaneously resize it to 640x480. Open the file in LoggerPro or VideoPoint. Use the horizontal meterstick to calibrate, and then measure the length of the vertical meterstick. If the result is between 0.995m and 1.005m, your videos are being properly captured. If the result is different, try resizing the video to 658x480 and check the calibration again. If the vertical measurement is still not close to 1 meter, you will need to find another size by trial and error that does work for your camera.