Update 3

This week we accomplished many things. We can now separate all the objects in an image into separate matrices, edge detects, and takes the dct (to capture straight lines of the tank). Once we tell matlab how large the tank is (it's larger than a small truck, but most likely smaller than any large rocks or mountains nearby) then matlab can spit back at us which object was the tank and what its coordinates are by telling us information about those separate matrices.

This method works right now for 7 of 8 pictures (we've just taken more pictures, so we can test our method on these soon). The one picture our method doesn't work on is for the tank being partially hidden underneath a tree (the tree really obstructed the view, and made it hard even for the naked eye to tell (from the binary images we created) that it was a tank. This method all works on images that we first dilated, then eroded, and then converted to a binary image using some reasonable threshold. We used the same threshold for all of our images, although it might be a problem if we had pictures of a tank on a really white or really black background.

We've also got a pretty good edge detector working, and once we have the edges we can then calculate the amount of space in between the edges so that if we know the tank is X pixels wide or tall (which is a reasonable thing to have information about - you'll notice most things are reasonable in our report!) then we can check if the edges we have fit in some range around that width or length, and if so, it is most likely the tank.