View my account

calculating similarity using the random dot method

Comments

6 comments

  • MartyG

    Hello Yuichiro Sano  This website provides support for Intel RealSense depth sensing cameras.  Do you want to calculate similarity using a RealSense camera, please?

    0
    Comment actions Permalink
  • Yuichiro Sano

    Doesn't the RealSense camera measure depth using the stereo method and random dot pattern (projecting random light patterns) technique? I would like to understand the principle of how the random dot pattern technique helps in calculating the similarity for the stereo method.

    0
    Comment actions Permalink
  • MartyG

    RealSense cameras project a semi random dot pattern from their infrared projector hardware component.  The pattern is cast onto surfaces in the real-world scene and the camera can use these dots to aid the analysis of the surfaces for depth information.

     

    The dot pattern is not used for similarity calculation though.  Instead, the dots projected onto surfaces provide a texture for the camera to analyze.  Surfaces that have some kind of pattern or visible texture are easier for the camera to read than plain surfaces with little texture detail or no detail, such as tables, walls and doors.

     

    RealSense cameras do not require the dot pattern projection to generate a depth image.  They can also alternatively use the lighting in the real-world scene (artificial or natural) for depth analysis.  However, when the dot pattern is not being used then the depth image typically has less detail on it, with more gaps and holes.

     

    A RealSense guide at the link below provides information about how the camera uses the dot pattern.

    https://dev.intelrealsense.com/docs/projectors#5-the-dot-pattern

    0
    Comment actions Permalink
  • Yuichiro Sano

    I seemed to have a slight misunderstanding. Let me ask an additional question. Can 'texture' be read as pixels in an image? Do the pixels in the image captured by the camera sensor change when there are dots present versus when there are no dots?

    0
    Comment actions Permalink
  • Yuichiro Sano

    I'm looking at this page: https://www.intelrealsense.com/stereo-depth-vision-basics/ I think that when a dot pattern is projected, the calculation of SSD becomes more stable. In other words, the minimum SSD becomes more prominent. Am I mistaken?

    0
    Comment actions Permalink
  • MartyG

    On the depth image, the surface texture will be represented by colored pixels whose color represents the depth value of that pixel.  A common way of colorizing the pixels is between the blue and red spectrums, with depth values closest to the camera being blue and then the color shifting from blue to yellow to orange to red as the distance from the camera increases.  In a blue to red spectrum, (near to far) red represents the detail in the scene that is furthest from the camera.

     

    The pixels would typically be affected by whether the pattern was present or not.  For example, if the pattern was absent and the level of lighting in the scene was not good then the depth values might be more inaccurate, causing them to be rendered as a different color.  Or if that area of the scene was not readable by the camera, it could appear on the image as an empty plain black area, meaning that there is no depth information in that part of the image.

     

    The more projected dots that are in the scene, the better for depth analysis.  The number of dots could be increased by using an external projector or by placing two or more RealSense 400 Series cameras close to one another so that their pattern projections overlap and create a denser field of dots.

    0
    Comment actions Permalink

Please sign in to leave a comment.