Target identification from a rolling image sequence

An infrared camera mounted on a high-speed airborne projectile provides imagery information necessary for self-guided target tracking capability after launching. However, due to the interaction of air now above and below the wingspread, the rolling phenomenon occurs. An important feature for the identification of target above water mark is the horizontal reference line separating the boundary between waters and sky. However, due to the rolling effect, this reference line may tilt toward certain direction, thus deviates from its horizontal position. The goal here is how to efficiently compensate the rolling effect and de-rotate the reference horizon to its normal position. Given the rolling angle θ, the rolling effect can be simply compensated by applying one more rotation with reverse angle -θ. However, this floating point geometric transformation, operated on a pixel-by-pixel basis, is highly computation intensive. In this paper, a simpler approach has been developed by recognizing the inter- and intra-frame properties of the image sequence and reducing the above floating point operation into summation of constants