Understanding Video Caliper Accuracy
??J. Kovacs, May 01, 2002
The MicroImage VMU-300 and VMU-400 are capable of measuring to an accuracy of better than 0.2% under optimal conditions. This accuracy, however, is highly dependent on a number of factors. Understanding these factors will help you to minimize their impact on your readings.
The minimum line resolution of the VMU-700 operating with an NTSC RS-170 compatible signal is 1:1280 in the horizontal dimension and 1:480 in the vertical. Using a PAL video source on a VMU-700P offers a horizontal line resolution of 1:1536, compared to a vertical of 1:576. As a result, a horizontal measurement can be more than twice as accurate as one taken vertically. This lower vertical resolution is due to inherent limitations of the NTSC and PAL video standards. An inaccuracy of + or – 1 pixel, due to this minimum resolution, is unavoidable. Other sources of error, such as cumulative rounding errors and control non-linearities, have been minimized by the use of extra precision digits for all internal calculations, and by the use of rotary digital encoders rather than analog potentiometers for user controls.
The VMU-700 allows the user to set a reference distance, either vertical or horizontal, and assign it a numeric dimension. A larger reference distance allows for greater accuracy. If a horizontal reference distance of 10 pixels (out of the 1280 available) is chosen, the best possible accuracy is only 10%, since a single pixel change is 10% of the total reference. If a reference distance of 1000 pixels is chosen, a single pixel is only 0.1% of the 1000, giving a far more accurate reference. Generally, measurements made over a larger portion of the screen can have a higher degree of accuracy than those over only a few pixels. Assuming ideal optics, keeping both the reference distance and the actual measurements to their largest possible sizes on screen offers the highest degree of accuracy.
All camera lenses create some distortion and non-linearity of the displayed image. This is due to a series of factors, such as the difference in distance between lens and subject at the center versus at the edge of the display. These distortions consist of a combination of spherical distortion, pincushioning, and barrel distortion, and usually become most pronounced near the edges of the screen, particularly with wider-angle lenses. Fortunately, lens distortion will have a similar effect on the measurement of both the reference object and the subject itself. Because of this, optical errors can be partially negated by making both the reference and actual measurements at roughly the same part of the display, and by selecting a reference object similar in size to the actual object to be measured. A WORST case scenario would be to set a reference dimension using a small object near the center of the display, and then to measure a large object located along one edge.
The camera itself may be a source of linearity errors. The regular spacing of pixels on a CCD or CMOS image sensor vary very slightly across its surface, and older tube camera sensors are notorious for severe nonlinearity. Because of this, the use of tube cameras for measurement purposes is not recommended.
As the distance between subject and camera changes, so does the apparent size of the displayed image. If two objects, identical in all but thickness, are alternately placed against the same background, the thicker object will appear larger, since its viewed surface is closer to the camera. To maintain a reasonable degree of accuracy, both the reference object and the subject to be measured must be at the same distance from the camera. A lens with a wide angle field of view will be more susceptible to this effect, while a concentric lens (in which the field of view may be represented by nearly parallel lines) is relatively immune.
Any time the camera is refocused, it changes the apparent size of the displayed object, and requires that the VMU be recalibrated for an accurate measurement. There is a narrow band of distance from the camera within which objects will appear in focus, and moving either closer to or farther from the camera will cause them to blur. The size of this band is known as depth of field. The aperture size of the lens has an inverse effect on the depth of field. As the lens aperture is reduced (higher f stop number), the amount of light will decrease, but the image will be sharper over a wider depth of field. A shallow depth of field is more difficult to work with, because of the blurring of objects out of the plane of focus, but serves as a better indication of what is within range of the current calibration.
After setting the crosslines on the VMU to a known reference dimension, the user can enter that dimension using the SET button and the numeric keypad. The on-screen display will then provide a numeric readout of the distance between lines. Through the menu, the user can choose the number of digits which the VMU will display. The ability to display dimensions using less than the maximum number of digits available allows the user to reduce the impact of the numeric readout on the active display area. Obviously, choosing too low of a number of digits of resolution will limit accuracy. On the other hand, displaying too many digits reduces readability, and may provide a false sense of accuracy.
For each pixel that the distance between lines changes, the display will change by an increment equal to the reference dimension divided by its size in terms of pixels. For example, a 50 pixel wide reference calibrated to a dimension of 2.000 would change by 2.000 divided by 50, or 0.040 per pixel. While the resolution of the numeric readout would appear to be 0.001, or 0.05% accuracy, a typical reading in this case would only be to within 0.040, or 2%. The last displayed digit would not be meaningful, and the next would only be partially significant.
Reference dimensions may be chosen within the range of 0.0001 to 9999, although this choice of dimension may impose other limitations on the accuracy of the readout. The VMU is limited to displaying no more than 5 significant digits per dimension. If, for instance, the user set a large reference distance of 500 pixels and entered a dimension of 0.001, the best accuracy obtainable would be 10%, or 50 pixels, since any change less than that would be under the 0.0001 minimum which the VMU could display. Choosing a reference number closer to 1 (between 0.1 and 100) should prevent this from occurring. For example, rather than enter a reference measurement as 0.003mm, using 3.00nm would keep the numbers well above the VMU’s lower boundary.