This thesis investigates the problem of machine and biological motion perception from an adaptive systems perspective. Adaptive temporal filters are incorporated into two established image motion analysis algorithms, one each from machine and biological vision. The correlation-based Reichardt detector (Reichardt, 1961) is equipped with adaptive filters to account for existing electrophysiological data on motion adaptation in the insect lobula plate. The wider applicability of the model is tested by recording electrophysiologically from cells in the mammalian nucleus of the optic tract (NOT), and by investigating rapid adaptation to motion in human psychophysical observers. Adaptive temporal filters are also used to implement a phase-based scheme for image velocity measurement (Fleet and Jepson, 1989). The use of adaptive filters reduces the computational load of the phase-based scheme while maintaining performance on a synthetic test sequence. The adaptive scheme shows an advantage over its non-adaptive counterpart at high levels of noise as adaptation serves to maximise the signal power in the output of the filters.
Adaptive motion analysis in the machine and biological vision
1997-01-01
Doctoral thesis, University College London (United Kingdom).
Hochschulschrift
Elektronische Ressource
Englisch
DDC: | 629 |
NTIS | 1988
|Robotics, Motion and Machine Vision in Manufacturing
British Library Conference Proceedings | 1994
|Robotics, motion and machine vision in manufacturing
Kraftfahrwesen | 1994
|Cognition Helps Vision: Recognizing Biological Motion Using Invariant Dynamic Cues
British Library Conference Proceedings | 2015
|LUH Institutsbestände | 1965
|