An attitude estimator provides non-causal attitude estimates for real-time motion compensation of sensed images on a moving platform. A non-causal filter processes uncompensated attitude samples received with a latency from an IMU at a high rate clock to provide an attitude estimate that is strictly non-causal but satisfies a just-in-time (JIT) criteria for real-time motion compensation of images captured at a low rate clock. On-average the error of the non-causal attitude estimate is less than the error of a causal attitude estimate. If the lag added by the non-causal filter is greater than the latency, the effective transfer function of the non-causal filter has a negative gain slope that attenuates high frequency noise of the uncompensated attitude samples. The attitude estimator may also include a causal filter to generate a causal attitude estimate for real-time active stabilization of the image sensor at the high rate clock.