Computer vision in scheduling of industrial operations under uncertainty in large-scale and dynamic environments

• Authors: Anastasios Doulamis;Nikolaos Matsatsinis
• Venue: ACA'12 Proceedings of the 11th international conference on Applications of Electrical and Computer Engineering
• Year: 2012

This paper researches on-line and self adaptive learning strategies blending with computer vision tools to visually observe and then schedule industrial operations that are executed dynamically, concurrently and under uncertainty in large-scale and complex industrial plants. Our methods emulate humans' learning; they recursively recognize objects and industrial processes, from visually observed data, combining innovative "look-ahead" learning techniques, able to estimate future states of objects/events, with dynamic model evolution approaches to make robust identification which are resilient to environmental changes. It also continuously improves objects/operations' learning by transferring knowledge within a network of distributed and active cameras so that what is learnt from one confident task will improve learning of other uncertain but related tasks. All these self adaptive strategies are framed with reverse learning methodologies (unlearning) which forgets erroneous or even contradictory visually observed and uncertain industrial operations. Reverse learning resembles humans' brain activity during sleep sessions; it clarifies mistaken samples to improve knowledge generalization.