A STATISTICAL CORRELATION TECHNIQUE AND A NEURAL-NETWORK FOR THE MOTION CORRESPONDENCE PROBLEM

A statistical correlation technique (SCT) and two variants of a neural network are presented to solve the motion correspondence problem. Sol utions of the motion correspondence problem aim to maintain the identi ties of individuated elements as they move. In a preprocessing stage, two snapshots of a moving scene are convoluted with two-dimensional Ga bor functions, which yields orientations and spatial frequencies of th e snapshots at every position. In this paper these properties are used to extract, respectively, the attributes orientation, size and positi on of line segments. The SCT uses cross-correlations to find the corre ct translation components, angle of rotation and scaling factor. These parameters are then used in combination with the positions of the lin e segments to calculate the centre of motion. When all of these parame ters are known, the new positions of the line segments from the first snapshot can be calculated and compared to the features in the second snapshot. This yields the solution of the motion correspondence proble m. Since the SCT is an indirect way of solving the problem, the princi ples of the technique are implemented in interactive activation and co mpetition neural networks. With boundary problems and noise these netw orks perform better than the SCT. They also have the advantage that at every stage of the calculations the best candidates for corresponding pairs of line segments are known.