Getting Depth Position Value from Depth Image RealSense SR300

[idata]

Hi,

Right now I am working on a project measuring the depth accuracy of RealSense SR300.Thus I need to obtain the world position of depth for every pixel in the depth image.

I use "projection->QueryVertices(depthIm, &vertices[0])" in the SDK to get the depth value, however, when I compared the data with the one from "JointData.positionWorld.z", the values are different and sometimes they differed a lot. Do you guys meet the same problems before?

PS: I am assuming the data from "JointData.positionWorld.z" are more accurate and I am comparing "vertices[pos].z" with that value where "pos" is "(JointData.positionImage.y - 1)*depth_info.width + (JointData.positionImage.x - 1)".

Best,

Eric

[idata]

Hi Eric,

Thank you for your interest in the Intel RealSense Technology. Let us try to help.I'm not sure if I'm correct about this, but it is possible that you're comparing values with different units. JointData.positionWorld, which is the joint position in 3D world coordinates, is expressed in meters. While the JointData.positionImage (the joint position in 2D coordinates) is expressed in pixels. This might be the reason why you're getting such different values.Regards,Pablo M.

[idata]

Hi Pablo,

Thanks for your reply.

I am comparing the depth values in the same unit. The first one is "JointData.positionWorld.z", which is in meters, and the other one is "vertices[pos].z", which is in mm.

I only use JointData.positionImage to get the joint pixel position, in order to find which vertice (or "pos" above) contains the same data.

The link below shows the vertice's depth unit.

https://software.intel.com/sites/landingpage/realsense/camera-sdk/v2016r3/documentation/html/queryvertices_pxcprojection.html?zoom_highlightsub=vertices QueryVertices

Best,

Eric

[jb455]

Your problem may be that the result of QueryVertices is aligned to the depth image, while presumably the JointData is relative to the colour image. Check out https://software.intel.com/sites/landingpage/realsense/camera-sdk/v1.1/documentation/html/index.html?mapcolortodepth_pxcprojection.html MapColourToDepth for a quick way to convert a small number of points between the colour and depth images.

[idata]

Hi all,

I still cannot figure out what the problem is.

Right now I obtain the depth information directly from the depth image, using "PIXEL_FORMAT_DEPTH". The result seems satisfactory.

Best,

Eric

[idata]

Now I face another problem. The depth value for joint is good in accuracy, however when I tried measuring the distance from a planar plane, the depth value is not accurate at all. Do you guys have any suggestions?

[idata · Accepted solution]

I reduced the data capturing range and the result is good. Problem solved.

[FLeac]

Hi Eric,

is your tool open source? I would be happy to play a little with it.

[idata]

Sorry, I'm sure I cannot give you the code until the project is finished.

The program is also just a console application since I am not good at GUI.

[PV1]

Hi Eric,

I am facing fairly similar issue. When the range is beyond 1000mm, the depth values are not consistent and most of them are invalid. Did you experience the same ?

What is the data capturing range that solved your problem?

Thanks in advance !

Best,

Prathyusha

[idata]

Hello pratzz,

Thank you for your interest on Intel® RealSense™ Technology. For future references, your ticket number is 03154411. Please keep this number handy in case you are asked for it. Beyond 1000mm you are getting close to the maximum effective range of the camera, which is 1200mm, so degradation of depth accuracy at this range is expected. If long range is important, set the IVCAMMotionRangeTradeOff to 100 using https://software.intel.com/sites/landingpage/realsense/camera-sdk/v1.1/documentation/html/setivcammotionrangetradeoff_device_pxccapture.html I hope you find this information useful.Best regards, Josh B.Intel Customer Support

[PV1]

Thanks Josh. It helped.

Best,

Prathyusha