are SR300 measures totally wrong?

[FLeac]

Hello, I using the SR300 as a 3D Scanner. Unfortunatelly i found that all the items i get from it have totally wrong measures.

The file attached is a fast (and not so accurate) example, but still a valid one. http://www.filedropper.com/book http://www.filedropper.com/book

Apart from the fact that the surface is totally inaccurate (probably because the book was glossy), if you use any measure tools for 3d objects, you will see that the scanned book is around 30x24.

In reality the original book is exactly 27.6x22.9 cm.

Did anyone else find that the measure are wrong? Do you have a solution? I am missing something? The book was scanned with the provided 3D Scan example in the SDK.

[MartyG]

Intel's recommendation for 3D scanning with RealSense is that you position the object so that it fills as much of the camera's view as possible. Please try doing this if you have not done so already and see if the measurements are more accurate.

[FLeac]

Yep, I already do that with all my normal scans.

Another note: if I use https://software.intel.com/sites/landingpage/realsense/camera-sdk/v1.1/documentation/html/index.html?setarea_pxcscan.html SetArea function the scanned area is smaller than the one I set.

[MartyG]

SetArea sets the size of the area that will be scanned. If you give the Area variables (Shape and Resolution) a value of '0' then they are set to the default values.

With the SR300, you can use an instruction called SetIVCAMFilterOption that enables you to alter the camera's depth scanning parameters to provide high accuracy at close / very close range scan.

https://software.intel.com/sites/landingpage/realsense/camera-sdk/v1.1/documentation/html/index.html?setivcamfilteroption_device_pxccapture.html Intel® RealSense™ SDK 2016 R2 Documentation

[FLeac]

Thank you for the reply. I wanted to explain that if i set (for example) 30 cm width, it scans less then 30. Like if his measures are wrong.

[MartyG]

I wonder if the scanning area width should maybe be set a bit larger than the width of the object, such as 32 cm. Otherwise it risks the edges of the object being cut off when the scan is made unless the camera is aligned with the object to absolute precision.

I used to face this problem when setting up the cover image of books on the Lulu online book publishing platform - Lulu's online publishing software would cut the edges off the cover image when analyzing it to make it fit to their book printing measurements, so you always left some empty space at the edges of the image so that text or important parts of the cover image would not be cut off in the published cover.

[FLeac]

Yup, It could probably be a good practice. But unfortunately the main problems still persist (with or without SetArea) Seems like the SR300 is useless for anything that needs accuracy.

N.B.: I tried multiple SR300 cameras, all have the same problem.

[MartyG]

If you are using the SR300 to create 3D model files, and what you are effectively getting is a squashed version of the original object, then a workaround - though not ideal - would be to load the created 3D model file into 3D model software such as SketchUp, Blender or Maya and then manually re-scale the object to the desired proportions by inputting the width and height values that you want to resize the model to.

On reflection, thinking about Intel's advice about the cameras - that the SR300 is suited to small objects and the R200 is suited to larger objects and full human bodies - a book like you are scanning probably falls into the 'larger object' category that the R200 is more suited to.

[FLeac]

Unfortunately manually rescale the object is not an option for my application. Moreover I tried to scan objects like 5x5 cm, but no improvements.

[MartyG]

I don't know the precise measurement scale for when a 'small' object becomes a 'large' one. Are you scanning on a flat surface such as a table?

[FLeac]

I tried both full and object scan modes. Both have the same problem

[MartyG]

The number of possible things to try are sadly running out. If you are having the same issue with multiple SR300's, it is not likely to be a calibration or lens distortion issue.

The company 3D Systems, whose Sense system utilizes RealSense, had this advice: "Before scanning a small object or one without a lot of distinct features (such as a standard coffee mug), add some items around the object. This will help the scanner maintain tracking by giving it more to lock on to."

[idata]

Hi FrancesoLeacche,

Do you still need assistance with this thread? We'll be waiting for your response.-Sergio A

[FLeac]

I still need help, totally.

[idata]

Let us investigate on your case. We'll post our findings here soon.

Thank you for your patience.-Sergio

[idata]

Hi FrancesoLeacche,

We're currently working on your case. Would you please provide details on how you measured the book after it was scanned? Could you also send us the OBJ file resulting from the scan? We also had some problems opening the link from your original post, could you try to resend that information? Also, how far was the camera from the object?We need more information to reproduce this problem, the more detailed information you can give us the better.Regards,-Sergio A

[idata]

Hi FrancesoLeacche,

We're following up on this case to see if you still require assistance. If you do, please provide the information we requested in our previous reply. We'll be waiting for your response.Regards,-Sergio A

[FLeac]

I tried different approaches but still can't find a way to receive acceptable measures

I use Maya 2017 measure tool to measure the scans (it's not perfect but has an acceptable error threshold).

Here is another model I just scanned. it's a 4cmx4cmx4cm Cube that I modeled, 3d printed and measured before the scan to be sure about the measures.

Unfortunately the resulting mesh is (around) 39mm x 45mm x 52mm taken from (around) 30 - 35 cm.

[idata]

Thank you for your reply. We'll review this information and contact you back soon. Thank you for your patience.

-Sergio A

[idata]

Hi FrancesoLeacche,

Thank you for your patience. We've reviewed the information you provided and tried to reproduce your environment to test the accuracy of the SR300. The Box measurement tests are a difficult use case for the SR300 and unfortunately Intel has not developed a solution for it. We observed that the uniformity of the object makes edge detection difficult. We did have slight improvements in performance in the measurements under certain conditions. These conditions include doing the measurements at different distances, backgrounds, and lighting conditions.Regards,-Sergio A

[FLeac]

Hello Sergio,

Thank you for looking into my problem.

I would add that the cube was just a demonstration. We have this problem with other models.

We use the Realsense to scan anatomical parts, like arms and legs, but the measures are totally wrong and we cannot use them for medical purposes where we need an error < 1mm.

If you could give me a direct contact (like an email) i could send you some demonstration video and explain it better.

Kind regards,

Francesco Leacche