[Intel RealSense D435] Stereo vision algorithm

• MartyX Grover

  • January 18, 2021 12:53

  Hi Johan E Ekman  Pages 16 and 17 of the current edition of the data sheet document for the RealSense 400 Series stereo cameras provide basic details of how the 400 Series cameras generate depth from the left and right sensors.

  https://dev.intelrealsense.com/docs/intel-realsense-d400-series-product-family-datasheet 

  0

  Comment actions Permalink
• Wl2776

  • February 15, 2021 13:01

  Hi Johan!
  Could you give a link, where Intel refers to SGBM?

  Thank you!

   

  0

  Comment actions Permalink
• MartyX Grover

  • February 15, 2021 13:20
  • Edited

  Hi Wl2776  I wonder if the information resource being referred to is Intel's Depth from Stereo guide, as it discusses the SSD (Sum of Squared Differences) block-matching algorithm, and use of stereobm and stereosgbm with OpenCV for disparity calculation.

  https://github.com/IntelRealSense/librealsense/blob/master/doc/depth-from-stereo.md

  1

  Comment actions Permalink
• Wl2776

  • February 15, 2021 14:13

  Thank you, Marty!

  However, details about depth calculation algorithm, implemented inside D435, are still hidden. 

  I've got the same question, as the topic starter, and it is still unanswered.

  Maybe, you've got links to patents or published papers, describing this algorithm?

  0

  Comment actions Permalink
• MartyX Grover

  • February 15, 2021 14:51
  • Edited

  Algorithms used by RealSense cameras are typically not shared publicly due to being Intel's private intellectual property and are not open-source.  

  The RealSense SDK does offer the software_device interface for emulating RealSense devices.

  https://github.com/IntelRealSense/librealsense/tree/master/examples/software-device

  1

  Comment actions Permalink
• Wl2776

  • February 15, 2021 15:08

  Does it mean, that I can

  * take two web-cameras

  * grab video streams from them

  * pass frames to that device

  and receive depth map?

   

  0

  Comment actions Permalink
• MartyX Grover

  • February 15, 2021 15:42
  • Edited

  There is not much documentation for software_device outside of the link above.   My understanding of it is that instead of attaching a physical device such as a webcam, you provide the scripting with the stream intrinsics values of the camera that you are emulating.

   

  

   

  0

  Comment actions Permalink
• Wl2776

  • February 15, 2021 16:15
  • Edited

  I've downloaded SDK and launched the example... Only depth post-processing can be tested. Examples assume that depth is coming from outside.

  Up to now I've found this paper: https://arxiv.org/abs/1705.05548

  It outlines some details on older devices.

  0

  Comment actions Permalink

Please sign in to leave a comment.