Here’s the first video that shows how computer vision is going to be used to control the drone in future versions of the library:
Basically, what the drone controller does is:
- Apply some filters to the drone image
- Detect contours in the filtered image (using an HSV color model)
- Match the contours against the sign shape
- Match the image within the contour against an ideal sign model
If the image passes all these tests, I paint the white frame shown in the video.
The next steps will be to detect the direction in which to fly (which is not as trivial as it may sound since the drone’s normal does not always point in the same direction as the normal of the ground). When this is done, I can adjust the pitch and roll values so that the drone follows the sign. One problem that may arise is that the bottom camera has a limited field of view (63°), so fast sign movements will easily move the sign out of the FOV.
I will continue to work on this project when I return from my holidays.