Performance Considerations

Both pose tracking and pose detection rely extensively on the GPU. On systems with a single GPU they will interfere with each other. This can reduce the frame rate and consequently the quality of the optical flow. This is not an issue for pose detection, but will reduce the robustness of tracking.

Care must be taken to correctly assign the GPU(s) to the modules in parameters.yaml:

simtrack/tracker/device_id: 0
simtrack/detector/device_id: 1

The above example is for a dual-GPU-system. Try swapping the values to see which configuration results in the smoothest display. This depends on which GPU is driving the screen (and should typically be assigned to the tracker).

In a dual-GPU-system tracking and detection run continously, which results in higher precision and fast recovery from tracking failures.

In a single-GPU-system parameters.yaml should be set as follows:

simtrack/tracker/device_id: 0
simtrack/detector/device_id: 0

Now the detector will be dynamically controlled to reduce interference. As soon as all the objects mentioned in the configuration file are tracked reliably, the detector will shutdown. It will re-activate when an object is lost. For this reason it is important to not include objects in the configuration file that are not in the scene!