We have been using the following components together, yet the FPS remains low (range 2 to 6 fps)
- raspberry pi 4
We are thinking of switching to AAEON UP squared board instead of the RPI 4 to increase FPS.
Would that be a good decision even if we don't have multiple USB3 ports?
What FPS should we expect to get with this component combination if the models we are using are "human-pose-estimation-0001.xml".
It would depend on your application bottleneck.
If the bottleneck is on the parsing of image data from the D415, changing this board should help.
If the NCS2 is the bottleneck you could use the "multi' plugin for the inference engine.
Getting the AAEON UP squared board with the AI Core X (equivalent to NCS2 as both have single
Movidius Myriad X ) can achieve up to 105 fps (80 typical)
The "human-pose-estimation-0001" model relies on tuned MobileNet v1 as its feature extractor.
MobileNet v1 performance on NCS2 is expected to be approximately 119 FPS.
The following page contains the performance benchmarks of selected topologies against Intel hardware.
Not having multiple USB3.0 ports is fine as long as you have at least 1 for the RealSense D415 camera ( RealSense camera requires USB 3.0 for the depth and color data streaming).