Final tests & comments - OSLL/duckietown-cslam GitHub Wiki

The final version of the marker detector from the new_marker_detector branch shows the following statistics of the number of processed 1296x976 frames per second:

The final version from new_marker_detector_only_cpp processes a 1296x976 frame with 1 marker in about 30ms, with 3 — in about 61ms. However, for this, detection is called on the callback argument itself (in order not to make copies — one takes about 13 ms), which makes it impossible to directly implement image_queue, and hence the simultaneous operation by several ImageProcessors.

Possible next steps include tasks:

• Move detection node to c++

• Send images from camera_node via publish(const boost::shared_ptr<M> &) (in this case the argument in the callback is allowed to be weak copied)

• Organize receiving frames directly from the detection container (the authors of ArUco also have a library for working with the RPI camera)

• Adjust the ArUco configuration to the specifics of the problem being solved