Kinect v2 Calibration - Depth mapping to color space RRS feed

  • Question

  • Hi everyone.

    When I am using the MapDepthFrameToColorSpace, I wonder if the default extrinsic parameter given by the SDK is accurate enough? Because in my application I want to track an object as accurate as much, I am thinking if extra calibration is needed for best performance? Thank you!

    Tuesday, December 30, 2014 7:05 AM

All replies

  • What have you tried? This is going to be very dependent on the environment you are running in and your requirements for accuracy.

    Carmine Sirignano - MSFT

    Saturday, January 3, 2015 12:41 AM
  • Hi Carmine Sirignano, I have read document on opencv & ros package for calibration on RGB-D sensor, but haven't tested the difference between "MapDepthFrameToColorSpace" & mapping by my own. 

    In my application, I am going to track a ping pong ball and predict its trajectory, so I believe method that provides higher accuracy is better for me. 

    Besides, what you mean by "dependent on the environment you are running" ?

    Saturday, January 3, 2015 10:10 AM
  • How fast is the ball travelling? Kinect is only going to give you 30fps. Typically accuracy measurements are going to be on static points.

    The environment factors you need to concern about are sources of IR noise, material reflective properties and amount of light(although this is only for color). If you have any IR source(the sun) that is going to add noise which affects the IR intensity values the sensor is picking up. Reflective properties of the material, if it is overly reflective or absorbent of IR light you will get varying results.

    Carmine Sirignano - MSFT

    Monday, January 5, 2015 6:45 PM