Measuring glass to glass latency

Application Development
1

HI all,

I am currently working on a project where I need to estimate the glass-to-glass latency of my system. I have a simple camera that I use to frame a precise timer, and I display the live captured frame on the same computer. I then take several screenshots to measure the time difference.

image
image1323×965 142 KB

My GStreamer pipeline reports a latency of 45 milliseconds, with 30 milliseconds from the camera appsrc and 15 milliseconds of processing-deadline. This is in line with the data captured by the tracers.

pipeline_full.dot
pipeline_full.dot3264×492 96.6 KB

Figure_1
Figure_11280×960 87.5 KB

However, the measured latency is still approximately 120 milliseconds. I am curious if there is any additional buffering occurring somewhere in the system, as the vsync is disabled.

I would greatly appreciate any insights or suggestions you might have regarding this issue

Fabien

2

You allow for 100 buffers on capture queue, and probably that the delay is matching your timestamp?

3

Even with the max-buffers at 1 the delay stays the same

4

One last thing you can check, set sync=0 on glimagesink, that will workaround any timestamp issues, if the delay stay the same, then it’s an external one, which is the code the feeds appsrc, and the path from rendered GL canvas to your display.

1 Like
5

Thanks I’ll continue to investigate. I’ll let you know if I find anything.

Rewriting the appsrc from python to rust doesn’t change much. There is a simple loop that pass the frames from the camera to the appsrc.push_buffer().