Detection of aircraft in video sequences using a predictive optical flow algorithm

This paper presents a computer vision algorithm that segregates spurious op tical flow artifacts to detect a moving object. The algorithm consists of s ix steps. First, the pixels in each image are shifted to compensate for cam era rotation. Second, the images are smoothed with a spatiotemporal Gaussia n filter, Third, the optical flow is computed with a gradient-based techniq ue. Fourth, optical flow vectors with small magnitudes are discarded. Fifth , vectors with similar locations, magnitudes, and directions are clustered together using a spatial consistency test. Sixth, similar optical flow vect ors are extended temporally to make predictions about future optical flow l ocations, magnitudes, and directions in subsequent frames. The actual optic al flow vectors that are consistent with those predictions are associated w ith a moving object. This algorithm was tested on images obtained with a vi deo camera mounted below the nose of a Boeing 737, The camera recorded two sequences containing a second flying aircraft. The algorithm detected the a ircraft in 82% of the frames from the first sequence and 78% of the frames from the second sequence. In each sequence, the false-alarm rate was zero. These results illustrate the effectiveness of using a comprehensive predict ive technique when detecting moving objects. (C) 1999 Society of Photo-Opti cal Instrumentation Engineers. [S0091-3286(99)01603-7].