Practical Problems

Sky: The test images we initially tested on did not have the sky in them, but the pictures we took from the drone during mission testing did. We accounted for this problem by adding another mask to remove the sky from the drone images.

Camera stability: Since we did not use a gimbal we had concerns about the camera shaking and saving a blurry image. We adjusted for this by running each picture taken through a BRISQUE metric in python (library in the Software section) before saving it.

Colors & Corners: Our image processing algorithm had problems with distinguishing corners and distinguishing between darker colors. To address this, we added a series of masks

Finding Correct Corners: We struggled to find the correct corners to use in our image processing algorithm. We solved this by polling for more than the anticipated number of corners and then mathematically determining which ones were the most likely to be correct.

Point to Heading: The point to heading command will not rotate in the negative direction, which caused issues in initial tests where the drone would spin around to reach a target heading instead of taking the shorter path. We did not end up resolving this issue but ended up not needing to as we only used one waypoint during our final demo run.