Application of multichannel vision concepts and mechanisms in an artificial industrial vision system

Abstract

The design of visual processing systems in very demanding industrial environments is a technical field in which bioinspiration has not been explored as a developing tool. The need of extremely quick, accurate and real time responses needed in industrial applications is not usually seen as compatible with the "messy", "slow" or "inaccurate" methods and algorithms inspired in the information processing mechanisms underlying neural activity in the visual pathway. We are trying, thus, to explore the practical possibilities of interaction among concepts from both worlds: the "real" vision system designed for a real time quality control of a production line, and the "inspiration" taken from multichannel biological vision. In previous papers [1,2] a biologically plausible parallel system for visual detection of form, movement, shape and size has been developed. The system, working off-line and skipping real time restrictions, was tested for a variety of situations, yielding very good results in estimating the mentioned visual characteristics of moving objects. Furthermore, a second parallel-computing version was designed introducing the concept of parallel channel processing, e.g., the discrimination of different visual characteristics by mean of multiprocessors and multithread computing. The architecture we present here, which includes certain concepts developed in the previously explained results [3,4], is intended to work in the production line of a beverage canning industry where cans with faulty imprinted use date and lot number have to be immediately discharged from the line.