atleaa / wp3_calibrator Goto Github PK

info that should be found using a config file:

• topic names
• sensor name
• intrinsics??

Using SD resolution, targets are often not recognized at far distance. When it is recognized, the corner estimation may not be very good. So by averaging the corner detections, the result can be used for cropping depth images even if the marker was not identified on the specific image.

possible solution:

Mat result(CV_64FC3, listImages[0].size());
for(int i = 0; i < listImages.size(); i++) {
result += listImages[i];
}
result /= listImages.size();
result.convertTo(result, CV_8UC3);

When Arucos are found. Crop the depth cloud around them using a mask that captures the targets and white edge only. (Remove rectangular cropping)

• Generate point clouds from aruco marker images
• Create point clouds similar to reflective patches
• combine point clouds at known poses in a world reference map.
• Calculate average aruco pose of world map.
• Align combined sensor cloud with world reference using averaged aruco pose.
• Refine alignment using ICP between combined sensor cloud and world reference map.

There may be something wrong with the multithreading.

This should be done by using a subscriber to the camera_info topic

When the resolution is low or the marker is far away, perspective is lost and ambiguous poses are detected. This can be solved by comparing corner points of estimated pose with corners in the depth image.

afer all calibrations are saved in a file. A separate ROS node should only load and publish these.

Calculate average pose from a set of transformations

If file exist, load. If not calibrate?..

keeping all images requires too much memory.