Performance of passive ranging from image flow

The accuracy of passive ranging derived from image flow is investigated for potential use in air vehicles. By using the optical flow constraint, a closed form expression for range at each pixel is obtained directly from spatial-temporal gradients of the image intensity, the pixel location, and the known egomotion. Gradient estimation is performed with a space/time Sobel mask, and a local median filter is employed for rejection of outliers due to algorithm singularities. The filtered range image is smoothed by a two-state Kalman filter. Simulation results employing this range estimator in translation and rotational camera motion situations are promising and appear suitable for air vehicle applications. Scenarios where large inter-frame pixel motion degrades the ranging accuracy are discussed.