I understand that in the first pipeline, it takes *644.3* ms o receive the
data stream from the camera and in the second pipeline it takes *473.2* ms.
Is this interpretation of the data correct? If so, why does it take so long
to acquire the flow?
This is not what the latency tracer measure. It mesure the latency added by the pipeline while processing the frames. In your case, it says you have about 1ms latency, which is near to nothing. Basically it's the average processing delay. This is considered zero-latency.
The ts= value is the time the trace was recorded. Reported latency can be changed at run-time by elements, so we have to add this information here in order to differentiate the records. What it says is that it took 473 and 644ms for the pipeline to reach a point we're latency could be calculated. This is likely close to when the pipeline started streaming, but this is just correlation, and not the intended information.
Measure the glass to glass latency is not possible from a tracer. You need special streams and equipment to achieve this.