Q. What should I watch out for when taking pictures?
A. The most important factor, from our experience, is the separation between the two shots (or the length of the baseline). The baseline must be of a reasonable size. We recommend 1 to 2 m (3 to 6 ft) for indoor environment and 2 m (6 ft) and above for outdoor environment. A general rule of thumb is that the further away the object, the large the baseline is needed to ensure high metrology accuracy. Hence, if the metrology accuracy is not satisfactory, always try to retake pictures using a larger baseline.
Q. What is a good reference?
A. A good reference must possess the following properties:
* It cannot be too small. It should be roughly the same size as or bigger than the objects that you want to measure (e.g., using your ipod as the reference for measuring the size of your back yard is a bad idea),
* You must be able to obtain the true 3D length of the reference accurately,
* You must be able to locate the references in the two images accurately, or the references should have sharp, easily discernable end points, and
* When multiple choices are possible, use a reference that is physically close to the object that you want to measure.
Q. Does it make a difference if the metrology algorithm can infer the camera motion automatically?
A. Our experience has been that metrology accuracy is generally better if the metrology algorithm can infer the camera motion automatically.
Q. What is the best way to take pictures to ensure that the metrology algorithm can infer the movement of the camera automatically?
A. The algorithm has the best chance of success in inferring the movement of the camera if (1) the scene is sufficiently textured, and (2) the viewing directions of the two shots are roughly parallel.
Q. How come I got different results when I take the same measurements on the same pairs of images multiple times?
A. If variation in measurements is small, the most likely cause is that you may not have located the corner points exactly the same from one run to the other. If variation in measurements is large, the most likely cause is that you may not have located the reference in the images accurately or the reference used is too small comparing to the size of the objects that you want to measure. In this case, it is necessary to choose a larger (longer) reference, obtain its true dimension accurately, and measure its locations carefully in images.
Q. How come I got different results measuring the size of a repetitive pattern (e.g. picket fence, windows) in the same image?
A. Small variation is inevitable. If variation is large, the most likely cause is that the camera’s lens has significant aberration. The telltale sign of lens aberration is that long vertical lines (tall book shells, doors, buildings, etc.) appear curved. Distortion is usually most pronounced at the edges of the image. So if you obtain inconsistent measurements of a repetitive pattern, you should trust those from the center of the image more.
Q. Do you provide a solution to correct for camera lens distortion?
A. Yes, it will be available in our next release.
Q. Sometimes when I see the message “Please specify at least 6 correspondences for further processing”, I notice that measurement results can change quite a bit with each additional correspondence added.
A. That is true. When you see the warning message, the algorithm will refine all measurement results with each added correspondence. Drastic changes in existing measurements do occur if measured objects are far away or the depth variation is large (this can happen for outdoor scenes). To ensure high metrology accuracy, you should specify significantly more correspondences (a dozen or more or until the measurements stabilize). Another possibility is to retake pictures with a larger baseline.
Q. What happens if I got bad results?
A. There are a number of things that you can try, including
* Use a larger sideway movement (this is the most important point)
* Use a higher camera resolution
* Use a larger focal length (or smaller field of view)
* Do not pan the camera too much (i.e., the viewing directions of the two shots should be roughly parallel)
* Make sure that the objects you want to measure are sufficiently large in the photographs
* Turn the flash on to obtain high quality images
* Select a larger reference and measure its length carefully
* Exercise care in marking the feature locations (using a high zoom)
Combinations of all above
For more information, contact us at firstname.lastname@example.org