Measuring an object from an image using a known object size

Question

So what I need to do is measure the length of the leg from an image taken by a regular user. This image will contain a leg with a black toe, a coin (or other known dimensional object) and white paper (for example, A4), where the other two objects will be located. What do i already have? -I have already worked with opencv, but just with simple projects; -I have already started reading some articles on camera calibration ( "Learn OpenCv" ), but still do not know if I should go so far.

Now I need a certain orientation, because I still do not understand if I am solving this problem correctly. I have some questions: will I really need to calibrate the camera to get two or three foot measurements? How can I find the points of interest to me to measure the line, each picture represents a different picture, or are there methods to follow?

Ps: sorry for my English, I really need to improve it: - /

Answer 1

Firstly, some things for obtaining images:

• Can you count on a black sock and a white background? Colors do not matter, as is the high contrast between the sock and the background.
• Can you standardize the viewing angle? Looking directly at the foot will reduce perspective distortion.
• Can you standardize stage lighting? This will facilitate the processing described below.
• Finally, you’ll get a more accurate estimate if you zoom in (or position the camera closer) so that the foot fills more of the image frame.

analysis. (Please note that this discussion will focus on your question about determining the axes of the foot. Identification and analysis of coins will use a similar process, but there will be some differences.)

• The next task is to highlight the area of ​​interest (ROI). If your camera looks down on its foot, then the ROI may be limited to a white rectangle. My answer to this post is a good start for determining a square / rectangle: What is the easiest * correct * method for detecting rectangles in an image?
• If the leg is completely in the white rectangle, you can copy the image to the rectangle found in step # 1. This will limit the image analysis to the area inside the white paper.
• "Binarize" image using the threshold function: http://opencv.willowgarage.com/documentation/cpp/miscellaneous_image_transformations.html#cv-threshold . If you have chosen the threshold parameters well, you can reduce the image to a black area (sock pixels) and white areas (pixel without sock).
• Now the fun begins: you can try to match the contours, but if this was my problem, I would use the bounding fields for a quick solution or moments for a more interesting (and possibly reliable) solution.
• Use cvFindContours to find the outlines of the black (sock) area: http://opencv.willowgarage.com/documentation/structural_analysis_and_shape_descriptors.html#findcontours
• Use cvApproxPoly to convert the outline to the polygonal shape http://opencv.willowgarage.com/documentation/structural_analysis_and_shape_descriptors.html#approxpoly
• For a simple solution, use cvMinRect2 to find an arbitrarily oriented bounding box for the shape of the toe. The short axis of the box should correspond to the line in largura.jpg, and the long axis of the box should correspond to the line in comprimento.jpg. http://opencv.willowgarage.com/documentation/structural_analysis_and_shape_descriptors.html#minarearect2
• If you want more (possible) accuracy, you can try cvMoments to calculate the moments of the figure. http://opencv.willowgarage.com/documentation/structural_analysis_and_shape_descriptors.html#moments
• Use cvGetSpatialMoment to determine the axes of the foot. More information about the spatial moment can be found here: http://en.wikipedia.org/wiki/Image_moments#Examples_2 and here http://opencv.willowgarage.com/documentation/structural_analysis_and_shape_descriptors.html#getspatialmoment
• With known axes, you can rotate the image so that the long axis is aligned with the axis (i.e. vertically). Then you can just count the pixels horizontally and vertically to get the line lengths. Note that there are several assumptions in this moment-oriented process. This is a fun solution, but it may not provide more accurate accuracy, especially because the accuracy of your size measurements depends largely on the camera positioning problems discussed above.

Finally, I provided links to the older C interface. You can take a look at the new C ++ interface (I just didn’t manage to translate my code to 2.4)

Answer 2

Antonio Criminisi probably wrote the last word on this subject many years ago. See His "Single View Metrology" and his Ph.D. dissertation, if you have the time.

Answer 3

You do not need to calibrate the camera if your image has an object of a known size. Well ... at least if your camera is not too distorted, and if you do not expect high-quality measurements.

A simple approach would be to detect a white (perspective distorted) rectangle, map corners to an undistorted rectangle (using, for example, cv :: warpPerspective ()), and use the known size of this rectangle to determine the size of another object in the picture. But this only works for objects in the same plane as paper, preferably not too far from it.

Answer 4

I'm not sure if you need to build it yourself, but if you just need to do it, rather than encode it. You can use KLONK Image Measurement for this. There is a free and paid version.