5 Replies Latest reply on Feb 4, 2018 2:18 PM by MartyG

    Long range blob tracking in 2D?

    BassGroovin

      I have read the specs for all the RealSense cameras, and I'm trying to understand what may be an edge case that is not addressed by the standard specs.

       

      I would like to know if any cameras in the RealSense product line can execute 2D identification and tracking of blobs from 100 meters away if the physical objects are large enough?  Please note I am not asking about depth tracking here, this is simple 2D X/Y tracking.   In my mind, this comes down to two questions:

       

      • Does the camera have the ability to be set to essentially infinite focal length, so the blobs are at least somewhat defined at their edges (enough so they can be first identified)?  Some focal fuzziness is fine, as long as the 2D tracking engine can identify and track the blobs.  No human is going to be looking at the video stream, only the 2D tracking engine will be using it.

       

      • What is the minimum object size at 100 meters necessary for the blob tracker to ident and track them?  For instance, at 100 meters, would a physical object that is 2 meters by 4 meters be identifiable to the tracking engine?  Another way to characterize this question is: What minimum percentage of the X field of view, and the Y field of view, must be covered by a blob in order for it to be identified and tracked?

       

      Thanks!

      BassGroovin

        • 1. Re: Long range blob tracking in 2D?
          Intel Corporation
          This message was posted on behalf of Intel Corporation

          Hello BassGroovin, 

          Thanks for posting in the RealSense community. 

          Intel has only tested and validated our cameras within 10 meters. Intel does guarantee that under ideal conditions your camera will work properly within this 10 meters. Going over the 10 meter range is out of the scope of our testing environment and therefore we cannot comment if what you are planning to do will work or not. 

          Feel free to create a solution for your need and test it, however your solution is not validated by Intel. 

          Thanks for your understanding. 

          Best Regards, 
          Juan N. 

          • 2. Re: Long range blob tracking in 2D?
            BassGroovin

            Juan, thank you for your reply.  Yes, I understand this is outside the scope of validation by intel, and I accept that any results I may or may not achieve are at my own risk.

             

            With that having been established, I would like to ask the following question, which does fall within the validated specifications as published:

             

            • For each camera in the RealSense product line, at maximum validated distance as shown in the specifications for each camera, how large must a blob be in the video stream, expressed as a percentage of X field of view and a percentage of Y field of view, in order for the 2D tracking capability to correctly identify and track the blob?

             

            Thank you!

             

            Best Regards,

            BassGroovin

            • 3. Re: Long range blob tracking in 2D?
              MartyG

              You may be able to achieve blob tracking with the latest D415 and D435 cameras if you combine its RealSense SDK 2.0 software with a blob tracking module in the OpenCV software.  Google 'opencv blob tracking' for more details.  Here's an example:

               

              c++ - Blob tracking algorithm - Stack Overflow

               

              Older RealSense models could do blob tracking but my experience is that the surface being tracked had to be right up against the camera lens for it to be detected.

               

              The focal length of the D415's left and right imagers is 1.88 mm and the focal length of the D435's left and right imagers is 1.93 mm.  The color sensor's focal length is 1.93 mm.  In the RealSense SDK 2.0 software, the focal length is described by the variables fx and fy, as a multiple of pixel width and height.

               

              In theory, if you were using RGB mode only then the camera could see up to 100 m in a straight line without obstruction (on the 400 Series cameras, the depth sensing range maximum is around 10 m).  The important question is whether the detail of objects at that far range would be clear enough to do useful processing with them, or whether they would just be a distant blur.

              • 4. Re: Long range blob tracking in 2D?
                BassGroovin

                @MartyG, thanks for the reaponse. Yes, I had been looking at OpenCV for a while already.  My goal with the Aero kit (and RealSense) was to get a dev  platform with everything integrated at the image processing level already so I could focus on my app-level experiments out of the gate. However that is clearly not going to be the case.

                 

                So, I know that a simple GoPro with fisheye correction, and any one of the simple blob ID/track packages will likely do what I want. I I just need to do the dsp-level integrations before I can get to my app stuff. So be it.

                 

                Thanks again!

                • 5. Re: Long range blob tracking in 2D?
                  MartyG

                  You're very welcome.  I'm glad I could provide useful input.  Best of luck with your project!