We have developed a two-stage model of motion perception that identifies moving spatial features and computes their velocity, achieving both high spatial localisation and reliable estimates of velocity. Features are detected in each frame by locating the peaks of the spatial local energy functions, as for stationary images (Morrone MC and Burr DC. Proc R Soc Lend 1988;B235:221-245.). The energy functions are calculated for different scales and orientations, and integrated within a temporal Gaussian window. The velocity of features is determined by the direction of maximal elongation of the energy in space-time, evaluated by calculating the three characteristic curvatures of the energy at each feature point. To circumvent the aperture problem, the energy maps are blurred in space by various amounts. and velocity is computed separately for each spatial blur. The Weber fraction of the local curvatures (curvature contrast) describes the spatio-temporal energy elongation at each feature point, giving a reliability index for each velocity estimate. For each point, the velocity of the spatial blur that yielded the highest curvature contrast was selected, with no further constraints, such as rigidity of motion. Dynamic recruitment of operators of different size allows maximum flexibility of the analysis, allowing it to simulate human visual performance in the detection of noise images, transparent motion, some motion illusions, and second-order motion. (C) 1998 Elsevier Science Ltd. All rights reserved.
Motion analysis by feature tracking
MORRONE, MARIA CONCETTA
1998-01-01
Abstract
We have developed a two-stage model of motion perception that identifies moving spatial features and computes their velocity, achieving both high spatial localisation and reliable estimates of velocity. Features are detected in each frame by locating the peaks of the spatial local energy functions, as for stationary images (Morrone MC and Burr DC. Proc R Soc Lend 1988;B235:221-245.). The energy functions are calculated for different scales and orientations, and integrated within a temporal Gaussian window. The velocity of features is determined by the direction of maximal elongation of the energy in space-time, evaluated by calculating the three characteristic curvatures of the energy at each feature point. To circumvent the aperture problem, the energy maps are blurred in space by various amounts. and velocity is computed separately for each spatial blur. The Weber fraction of the local curvatures (curvature contrast) describes the spatio-temporal energy elongation at each feature point, giving a reliability index for each velocity estimate. For each point, the velocity of the spatial blur that yielded the highest curvature contrast was selected, with no further constraints, such as rigidity of motion. Dynamic recruitment of operators of different size allows maximum flexibility of the analysis, allowing it to simulate human visual performance in the detection of noise images, transparent motion, some motion illusions, and second-order motion. (C) 1998 Elsevier Science Ltd. All rights reserved.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
