A self-targeting missile system using computer vision

The use of visual sensing to acquire and track targets has the potential to significantly reduce cost in comparison to active sensors. We present preliminary results for a system that combines a webcam with a toy foam missile launcher on an altitude-azimuth mount. The controller uses foreground segmentation to identity new elements in a scene. It then applies SURF feature detection and a pinhole camera model to determine their position in the real world. Based on our analysis of the ballistic flight characteristics for the foam missiles, the system can thus generate the altitude and azimuth launch angles that will allow a missile to intercept the target. We interface with the missile launcher via USB, and implement a second vision system based on gradient detection to provide pose feedback for the launcher barrel. The system is consistently able to intercept targets, within the restricted parameters of our testing.