Best pattern/method for low resolution marker detection
Hi All,
Long time reader, first time poster.
I am trying to design and detect a marker that can be identified from as little as 20x20 pixels out of a 20MP photo (for aerial surveying purposes). Orientation is not fixed, and I need to design such that there are >15 unique colour combinations to determine unique targets.
My thinking so far is a circular black rim, then a smaller, concentric white rim, then four quadrants of colour (R,G,B,Black) inside that white, inner circle. Black will be in the same spot and act as an anchor. This combination works at about 30x30 px but my houghcircles python script struggles, detecting pretty much everything else except these markers at 20x20.
Does anyone have some experience in low DPI detection? I see lots on checkerboards, edges, etc, but the minimum feature size must be significant to not have cameras just smear out the information.
Any advice would be appreciated...
Cheers!
The 20x20 pixels is a problem already. We used a 2x2 checkerboard pattern for aerial imagery from a UAV and had to stop at 50x50 pixels not to ruin the detection phase and start detecting everything else. How about upscaling your image first?
Relying on color: You can probably differentiate 15 markers based on color, but significantly more than 15 will be challenging, I suspect.
Aruco markers: 20x20 would be pushing it for reliably detecting a 3x3 marker in real-world images. A 3x3 marker takes 3 marker bits, +2 black border bits, and then enough white background adjacent to the border bits. So really more like 7x7, which is only 3 pixels per bit. You would need a well focused (and not motion blurred), properly exposed / high contrast image for that to work reliably. I detect 3x3 markers with a 28x28 pixel area (under favorable conditions), but 20x20 would give me some heartburn, particularly in an outdoor imaging context.