matlabbe

From the recent OpenCV versions, I am not sure if they still use a nonfree SURF and SIFT versions. xfeatures2d contain both free and nonfree detectors. It was easier before...

Did you try http://wiki.ros.org/find_object_2d#A3D_position_of_the_objects ?

Using the homography and size of the object, you can derive the corners like in this example: https://github.com/introlab/find-object/blob/fd9270a5617f31087bc61c3fae919961a2240b78/src/ros/print_objects_detected_node.cpp#L60-L75

Yeah, they should be in pixels. When launching the node `print_objects_detected`, here is an example of output of the code above, the corners are in pixels: ``` $ rosrun find_object_2d...

When doing the homography, we play in subpixel accuracy, thus float. You can round the values if you need intergers.

It is not supported on Mac. Actually 3D localization of objects is only supported with ROS on Ubuntu: http://wiki.ros.org/find_object_2d#A3D_position_of_the_objects

- **RootSIFT** - Link: http://www.pyimagesearch.com/2015/04/13/implementing-rootsift-in-python-and-opencv/

- **MODS** - Link: http://cmp.felk.cvut.cz/wbs/index.html

- **Color Feature Detector** - https://github.com/eokeeffe/quasi_invariant-features

It is like the wrong depth is used for TF. The algo is pretty simple: https://github.com/introlab/find-object/blob/f7c38f91d0b076f365592974ff75972aa668b7f1/src/ros/FindObjectROS.cpp#L87-L92 it finds the depth in the middle of the object, then apply camera calibration...