Human body volume / mass estimation using orthographic projections

I am trying to estimate the human body volume and thus mass using the two images: one a front view and another a side view. the human body divided into multiple slices, slices can be assumed to be oval. The two orthographic views we can get the three dimensions of the slices hence volume of each slice.

I need help/suggestions for:

1) The best possible edge detection technique for getting the human body dimension

2) Calculate the dimensions in best possible way so that correlation errors are minimized. By correlation I mean I have to select number of slices and match them in the two images.

3) Best possible way to correlate two images since two images are either captured at different time and/or by different instruments

4) converting the pixel info into inches

Can the method I am using be called stereovision ? I would like to get some references links etc where this method is used for same or similar applications.

Answer 1

[Walter Roberson]

Why the repeated call for "best possible way" ? The "best possible" way to do any one of those steps might take a few years to implement (after several decades of research to identify it as being the best possible), and might only improve the fit by (say) 0.2% over a lesser method.

Your steps make no attempt to identify the content of each slice, and no attempt to identify the density of each portion of the volume, so you are only going to be using some kind of statistical mean of human body density when you estimate the mass. Don't you think that might introduce far worse errors than not using the "best possible" method at some stage ??

For example, if you were analyzing the content of the images, you might find that (say) 1/3 of a midsection slice volume is occupied by fat due to a "beer belly". Fat is heavy, yes, but only roughly half as dense as bone, so failing to make a "fat vs bone" mass correction could be equivalent of getting the volume wrong by 15% or more.

There is a saying: "Measure with a micrometer. Mark with chalk. Cut with a chain saw." Your insistence on the "best possible" methods is like measuring with a micrometer, but your treatment of all volumes as being of equal mass significance is like cutting with a chain saw.

[Suhas Deshpande]

Sorry if the phrase "best possible way" bothers you. The questions I have here may have multiple solutions and the fact is that no matter what method you try there will be errors in the final output that is the human body volume estimation. And matter o fact is that the error will be a part of my research. The application of my research does not ask or accuracy but something else. To talk about Fat vs bone, on a vertical reference the bones are about evenly distributed while fat is not. And average body mass of human body will be almost equal in each slice of the body (there will be 20-30 slices) since 50-75% of human body is water.

Final word there may be and will be errors but the "best possible way" may reduce the error. And again i am not expecting everything from one person. Multiple person having their own expertise may comment or suggest me on this.

Thanks

Answer 2

[Image Analyst]

This has been looked at for a long time in the medical community. For example http://circ.ahajournals.org/content/60/3/547.abstract. I saw another paper yesterday in my files (it was like 15 years old) where they got the volume of a heart from two orthogonal projections - and the final volume didn't look just like you simply did an extrusion in one dimension and then an extrusion in the other direction and found the intersection - it gave more of a rounded shape. You might research in the medical area : http://iris.usc.edu/Vision-Notes/bibliography/contentsmedical.html#Medical%20Applications,%20CAT,%20MRI,%20Ultrasound,%20Heart%20Models,%20Brain%20Models