A two-dimensional (2-D) machine-vision safety method and apparatus performs
high-integrity, high efficiency machine vision. The system digitally
filters a configurable number of time-sequenced camera images of a viewed
area to create a filtered image. Objects within view in a training phase
are detected and compared to objects detected in a runtime phase. Each new
image is then compared to the previously created model and a
pixel-oriented difference image is calculated. The difference image is
then processed for segmentation features attributable to an intruder
object. Alarm conditions are determined from threshold parameters related
to location of the intruder objects. For segmentation results that exceed
a predetermined threshold, an alarm condition is set. Setup parameters are
automatically generated by analysis of distinctively marked physical
boundaries in an image. Automatic tools assist in optimization of setup
configurations, based on size and speed of intruding objects, system
response time, and camera geometry.
