Trying to detect potholes in the road surface, is Kinect right for me? RRS feed

  • General discussion

  • Hi. I am completely new to Kinect besides playing it once. For awhile now I've had this idea about having a mounted camera or sensor to detect potholes in the road and then using that info to choose the roads with the least potholes and such. Google led me to Matlabs and image processing but I am afraid that it would just not work for potholes.

    I ran across a paper by a team of Uganda researchers who used Kinect with some success to detect potholes. Like I said I am very new to this and have access to Kinect hardware but I was hoping to make sure Kinect was a good choose for this project before I went crazy with it. Would anyone be willing to give me their advice or answer a few questions I've had?

    • Would Kinect be able to detect the difference between surfaces which were only 1 inches apart? What about a half an inch?
    • How fast could the car move and the Kinect still receive accurate data?
    • Should this be done in real time processing or after? I would like to do it real time.
    • What would be the best software for me to use? Basically I would want to set up a trigger that said if the surface of the road has a blob which is X amounts deeper then the rest of the surface tell this (which is separate but connected to the computer) camera to take a picture.
    • Would the fact that the vehicle is moving and bouncing around due to the same very potholes I would want to read causing the Kinect to also mess this whole thing up? I would think it would be ok because I am talking about reading single frames of depth images opposed to full on video.

    Thank you so very much for any help. I need it.

    Wednesday, March 19, 2014 1:09 AM

All replies

  • The sensor will only operate at 30fps so you will be limited with the speed at which you will capture. If you travel too fast, you will miss spots. Others can provide more detail on the speeds you can travel. Kinect Fusion is an example of how you can integrate Kinect data and reconstruct 3D volumes, but that will only focus on local volumes.

    Depending on the hardware you have available, it might be possible to do real-time, but really would depend on your algorithms. The Kinect itself has an accelerometer, so you can pull that data in which you are capturing the data.

    Carmine Sirignano - MSFT

    Thursday, March 20, 2014 6:33 PM