5 Replies Latest reply on Jul 3, 2018 10:25 AM by MartyG

    Depth cloud precision - Intel D415

    Nicolas.B

      Hi,

       

      I'm currently using the D415 camera to get the colored point cloud of some tomatoes. In this point cloud we can see several holes on the stems part of the tomatoes (see image below).

       

      I was wondering if I could make the point cloud look better by adjusting the parameters. The one i'm currently using is "High Accuracy". Also, I've seen this topicImprove depth sensing for IR absorbent material and I didn't really knew if it was related.

       

      Thanks,

       

      Nicolas

        • 1. Re: Depth cloud precision - Intel D415
          MartyG

          The image does not seem to be attached to your message. 

           

          It could be that light is reflecting off the top of the tomato's shiny surface and causing the holes in the stem.  Or the stem might be too thin for the camera to be able to pick up its details well.

           

          Apart from the methods described in the link you provided, such as baby powder, noise can be reduced using post-processing filters.  You can easily test whether such filters will improve the image by using the post-processing section of the RealSense Viewer settings.

           

          Another means of reducing noise is to reduce the value of the Laser Power setting.

          1 of 1 people found this helpful
          • 2. Re: Depth cloud precision - Intel D415
            Nicolas.B

            Sorry I forgot to add the image, here it is :Capture du 2018-07-03 17-37-22.png

             

             

            I'm using ROS to get the depth cloud of the tomatoes but I tried using the post-processing filters in the RealSense viewer and it didn't really enhance the point cloud. I used to get the point cloud with a Kinect 2 but I didn't had such holes then and I was wondering why...

            • 3. Re: Depth cloud precision - Intel D415
              MartyG

              Looking at the image, the above theories about reflection / thinness affecting the stems does not seem to be the case.  Given the level of light disruption in the room and the apparent office-like environment, I wonder if the room has fluorescent lighting like ceiling strip-lights?  These can be more disruptive than bulb-based lights because the gas in them flickers at a frequency hard to see with the human eye.

              • 4. Re: Depth cloud precision - Intel D415
                Nicolas.B

                The image are taken in an office like environment with the sun being the only lighting on the camera. We actually tried it in a greenhouse with fully natural light, here is a sample:

                tomato.pngtomato2.png

                 

                Thanks for your time!

                • 5. Re: Depth cloud precision - Intel D415
                  MartyG

                  The black spots on those tomatoes in the pictures are likely because of light reflecting off the surface of the tomatoes.  Given the totally black areas around the tomatoes, I wonder if there are panes of reflective greenhouse glass behind them that are preventing that area of the image from being seen by the depth scan.

                   

                  .You can also see the effect of strong light sources in your office scene, where the reflections on the white monitor screen is making a lot of the screen appear to be black.

                   

                  In this situation, I would recommend that you turn off the auto-exposure option in the RealSense Viewer and adjust the exposure and gain settings manually with the sliders.

                   

                  Edit: RealSense stream programming expert JB455 recently offered a good explanation for the large black offset that can appear around people and objects, like the silhouette of your body in the earlier pictures:

                   

                  " The reason for the black region is ... an effect known as 'occlusion' - the depth and color cameras are in different physical positions so can't see the all the same parts of the scene.  Hold something close to your face and open one eye at a time to understand this!"