Handling camera frame drop

Hi Guys,

I wish to access camera frames and do some processing on the frames. My problem is that for the specified time and frame rate given I am unable to obtain the estimated number of frames (when there is no processing done the number of frames dropped/lost is much lesser compared to when processing is enabled). A glimpse of the problem is shared on another thread which was specific to camera object identification code. However based on the discussion on that thread I believe this issue is prevalent when I use other sample codes as well. I felt it is appropriate to create a separate thread for the same reason.

Earlier post can be referred here :

https://devtalk.nvidia.com/default/topic/1022153/11_camera_object_identification-frame-count-issue/#5203701

In this context I have the following queries :

  1. How to detect frame drops ?

  2. Is there a way to estimate the exact number of frames (and time) lost between two frames which are successfully processed ( may be using any metadata which is supplied along with the frame) ?

  3. Is there a mechanism to prevent loss of frames ? Can a queue implementation solve this problem to some extent ? If yes, can someone refer me to some implementation which I can refer to ?

Thanks.

Hi lamegeorge,

You can check timestamps as suggested in
https://devtalk.nvidia.com/default/topic/1022153/jetson-tx1/11_camera_object_identification-frame-count-issue/post/5203652/#5203652

For 30fps you will see 33ms interval between frames. If you see 66ms, it obviously drops one frame.

You can compare the timestamps with default board(ov5693). If you can see constant 33ms in ov5693 but frame loss in your sensor, it looks to be an issue in sensor driver.