Gaussian pyramid based acceleration of optical flow reconstruction

This contribution treats modification of Horn&Schunck's optical flow reconstruction algorithm. The main goal of proposed adjustment is acceleration of original iteration method to be useable in the case the images are captured by camera in real-time. Algorithm modification is based on multilevel relaxation scheme. Optical flow calculation begins on most top level with rescaled pictures of scene and image derivates. The results from this level are used as calculation initial state on next lower level. The pictures for each level are generated by Gaussian pyramid. The calculation accuracy is compared by evaluation optical flow's vector angular error of proposed modification and original algorithm. The practical use of optical flow intensity representation is demonstrated by application allowing analyzing road traffic situation due to car counting in traffic lanes.