1.How did you derive the end-to-end latency data? Was it based on the clock display transmitted via the camera feed, or via SEI timestamps? I consider other measurement methods to be inaccurate.
2. How did you compare the latency measurements? Were these latency measurements conducted within the same network environment at same time ? This program involves no encoding operations; therefore, nvv4l2h264enc will provide no benefit.
3. With a latency of 50ms for rtspsrc, network fluctuations can lead to issues such as screen corruption and jitter.
Refer to this FAQ to determine the latency introduced by gstreamer/ds element; however, I believe that 800ms is currently a rather not bad result.