Detecting a moving object in a video feed is easy if the camera is still. Just look for the pixels that are changing. Detecting a moving object from a moving camera, however, is hard. It is even more difficult if the target is just a pixel in size, and you have just monochrome imagery to work with.
Below is a single frame from a video feed taken with an off-the-shelf webcam. The filming location was the western hill of Fort Reno Park, near the highest location in Washington, DC. The camera was aimed Southwest to image airliners landing at Reagan National Airport (DCA), about 8 km away “as the crow flies”. The webcam was hand-held, after drinking coffee so as you can imagine the video was quite jerky.
Where is the aircraft? Using proprietary algorithms invented at Centeye, we could detect the airliner even as it flew behind the tree, where it was partially obscured by visual clutter (the branches). We developed these algorithms with “see and avoid” applications in mind, but the techniques are usable to the general problem of providing vision to an air vehicle, and has a lot of other cool uses as well. This algorithm is patent pending. Contact us for details.
