An Automated Inspection System With Biologically Inspired Vision

The design of artificial systems has long been guided by disciplines like physiology, cognitive science and neurology [15, 16, 17]. Studies have revealed that such systems must have oculomotion, attention and spatiotemporal reasoning capabilities [8, 9, 18, 19, 20]. Vision researchers haven then proposed selective perception in pursuit of developing systems mimicking some of this behavior [11, 10, 4, 21, 22, 23]. Interestingly, few studies have focused on employing selective perception mechanisms in higher level tasks. One such example has been APES [3]. In our work, we extend these ideas to automated visual inspection – a task assumed to be simple, yet still posing problems in real-time and robust applicability in factory manufacturing settings and investigate the possibility of overcoming these problems. The novelty of our vision system is that it can direct its attention to spatial points of interest and both the temporal and spatial nature of the visual information thus gathered is used in the accomplishment of its task.