Biologically Similar Algorithms for Motion Evaluation in Machine Vision Systems

In paper algorithms of objects motion evaluation on noisy images are developed. For this purpose power and phase-power spectra are used. Additive noise influence on informative sections of power and phase-power differences is observed. It is shown that for perfecting of detection characteristics it is necessary to reduce peripheral noise influence, i.e. to reduce image resolving power from the centre to edges. It can be carried out by setting to zero a part of image vision signal reports. The person visual analyzer also has the variable resolving power, therefore, it is possible to name the suggested algorithms as biologically similar.

It is shown that frequency characteristics single differences do not possess stability at noise level increase. Therefore it is necessary to use frequency characteristics double differences. They are gained by formation from each video frame of a frame series, different in the length of setting to zero. It is deduced from the experiments that double differences are more tolerant to noise agency than the single differences of frequency characteristics or frame-to-frame differences of vision signals.

Experiment results on self-sustaining detection of moving objects on 50 various noise implementation are quoted. The probability of moving object correct detection by suggested algorithms in comparison with frame-to-frame difference of vision signals depending on noise level at different sizes, location and magnitudes of moving object is considered. It is shown that the suggested algorithms possess much more efficiency than frame-to-frame differences.