1 Reply Latest reply on Jul 17, 2018 4:50 AM by MartyG

    Depth map very noisy with D435

    Simone_C

      Dear Realsense Community,

       

      We are using  the d435 to perform skeleton tracking and gesture detection with external libraries.

      For our application we  would need a good quality depth frame in the range of 1-4 meters

      From our test the depth frames retrived from the sensor are really noisy and grainy.

       

      As I can see from Intel Realsense d4xx documentation, the RMS error should go with the square of the distance so i would expect:

      - RMS of 5mm at 1 meter;

      - RMS of 15mm at 2 meters;

      - RMS of 35mm at 3 meters.

       

       

      As you can appreciate from screenshots attached below, at 1 meter distance the human figure is quite good,

      but from 2 meters away it becomes very noisy and rough. The effect is more clear looking at the floor and walls,

      the gradient of the color is not regular as i should expect.

      The 3D reconstruction of the surface of the human is very distort, wavy and clearly not compatible with the RMS error listed above.

       

      Person at 1 meter distance:

      D435_depth_1m_Depth.pngD435_3D_1meter.PNG

      Person at 2 meter distance:

      D435_depth_2m_Depth_Depth.pngD435_3D_2meter.PNG

      Person at 3 meter distance:

      D435_depth_3m_Depth_Depth_Depth.png

       

      Do you think that the performances of our sensor are in line with d435 specs? or is there a clear problem of calibration?

      It would be possible to improve the quality of depth frames to obtain an RMS near to the expected value also at large distances (from 2 to 4 meters away)?

       

      Compared to other depth sensor (kinect or asus xtion), which have similare specs, the quality of depth map is definitly lower.

      In these conditions is quite impossible for us to apply our skeleton tracking and gesture libraries as the quality of depth map from 2 meter away becomes too low.

       

      I'm planning to do more precise test to directly evaluate the RMS of our sensor at different distances.

       

      Thank you in advance.