Use this URL to cite or link to this record in EThOS:
Title: Algorithms for low cost VLSI stereo vision systems, with special application to intruder detection
Author: Findlay, Kevin William John
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1993
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
The work described in this thesis is concerned with the development of hardware efficient, image processing and machine vision algorithms for implementation, using recently developed low cost CMOS cameras. These allow the integration of processing on the same silicon substrate as the imaging sensor. The general approach differs from other image processing research in that algorithms are being developed for a target architecture, rather than hardware being developed for a particular image processing function. A particular application, namely intruder detection and tracking, has been chosen, to demonstrate this approach. The use of image processing in alarm systems has many advantages over active electronics: the main ones being installation costs and reliability. In particular, stereo vision has the potential of providing an invisible wall and estimates of intruder significance. However it is also desirable that alarm systems have wide angle lenses. Wide angle lenses create particular problems for stereo vision, in relation to pixel quantisation. Techniques to provide a low cost sub-pixel estimate of disparity are presented. Further, an original stereo matching algorithm is described which solves the stereo correspondence problem, in a computationally simple manner. Adaptations are also made to the low level segmentation stages which would allow an efficient implementation using CMOS sensors and processing. Other savings have been made by eliminating digital floating point calculations, multiplications and divisions at the lower levels of processing. Also, due to the reduced data rates required for global frame to frame computation, higher level calculations can be performed on an associated microprocessor. Thus, a Kalman tracking filter has been applied to integrate the three possible disparities from three cameras, with experimentally calculated error covariance matrices. A results chapter describes the extraction of these matrices, together with simulations of the algorithms applied to twelve different sequences. These show that the system could be effective as an alarm system. Also described, at various stages in the thesis, are possible hardware implementations of the algorithms and partitions between analogue and digital circuitry. The thesis finishes with some general conclusions.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available