Lots of work to refactor c++ track_targets into threaded model, with separate threads for incoming and outgoing vision frames. Better handling of vision_landing python code for latency, including use of a monotonic clock to better synchronise with track_targets, and take correlation timestamp as close to the sensor shutter as possible.
Note: also update to aruco3 and optimise compilation. All this work is to make vision_landing work on all platforms including slower boards like raspberry.
Lots of work to refactor c++ track_targets into threaded model, with separate threads for incoming and outgoing vision frames. Better handling of vision_landing python code for latency, including use of a monotonic clock to better synchronise with track_targets, and take correlation timestamp as close to the sensor shutter as possible.