Fast visual search: from inspection to guidance and surveillance

Roy Davies

(Royal Holloway)

This talk considers a range of vision problems that can best be summed up
as ‘fast visual search’. After airing what the human eye can and can’t achieve,
the talk looks at the problems of real-time inspection of products on product
lines, and considers the location of insects in batches of cereal grain. It also
analyses how the grains themselves can be located at the fastest possible rates
by judicious sampling of the input images. The idea is extended to general
location of any objects, and a parallel is drawn with the sampling technique
employed by the human visual system (HVS). This inspiration is taken further,
in a vehicle guidance application, with the aim of locating pedestrians before
accidents can occur. It is usual for 3D scene analysis to proceed via the
identification of salient features in the image, but there is some doubt whether
this is necessary. Further work employing a ‘form + motion channel’ model to
the vision system, similar to that hardwired in the HVS, also suggests that
salient features may not be so necessary, and an integrated theory of guided
sampling will be discussed.
Tuesday 6th May 2008, 14:00
Robert Recorde Room
Department of Computer Science