• DepthAI-v2Hardware
  • Different of AutoExposure algorithm in OAK-1 AF when running on PC and Android

  • Edited

I have developed a C++ module to control OAK-1 AF, this module is then used by other modules to perform object detection using the images recovered from the camera. This modules can be used in PC, and also have a JNI and java implementation to be used by an Android device. So, the same camera parameters are set either when the camera is used in a PC or in an Android device. The issue is that I am having a difference in the behavior of the autoexposure for the same scene, where several times it seems as if the algorithm gets locked when the camera is connected to the android device. You can see in the image the difference between the two images obtained with the same camera and using the same c++ module to set parameters and camera control.

I use depthai 2.17.4 since is the only version that provides a cleaner image when using the GRAY8 format (the reason is explained here displaced pixels for depthai version higher than 2.17.4).

I have measured the energy consumption when connected to PC and Android and the results are similar:

  • For PC: 2.05 W (0.42 A @ 4.89 V)
  • For Android: 2.037 W (0.41 A @ 4.97 V)

I have read that a lack of energy might impact the performance of OAK-1 AF, but it might not be the case according to this data. Could you please provide some help or recommendations to find the reason and solution for such behaviour?

Thank you in advance

    Hi alexv
    So it's only the AE algo that causes a different behaviour? The manual exposure functions the same - meaning same setting prodeuces the same image on both OSes?

    Thanks,
    Jaka

      jakaskerl I have not tested the manual exposure, the scenario where the system works always has variations with respect to illumination conditions. This behavior of locked exposure when in autoexposure is not always present, which makes difficult to identify the cause.

        alexv
        You can try switching between AE and ME to potentially unblock the AE mechanism... Though I believe I have seen this before.
        Your depthai version is quite old and may suffer from some issues that have been fixed by now. Is the pixel displacement still an issue even in 2.29?

        Thanks,
        Jaka

          20 days later

          jakaskerl
          Hello Jaka,
          The issue with the pixel displacement still exists in 2.29, and also, the exposure blocking. What I did implement was the switch between manual and automatic exposure, it seems solve the blocking, but still there is an extra added time due to this switch while the user waits for the image to get the best exposure condition once it is reset.