Scene understanding and objects detection and identification with a perceptual system based on the network-symbolic models for industrial robots

Abstract

One of the major problems of modern industrial robots is a lack of reliable perceptual systems that are similar to human vision in its abilities to understand visual scene and detect and unambiguously identify objects. The traditional linear bottom-up "segmentation-grouping-learning-recognition" approach to image processing and analysis cannot provide a reliable separation of an object from its background or clutter, while human vision unambiguously solves this problem. The modern computer vision can only recognize certain features from visual information, and it plays an auxiliary role, helping to build or choose appropriate 3-dimensional models of objects and visual scene. As result, designers of robotics systems must create for industrial robots artificial environments, which allowing for precise computations of 3-dimensional models within such environments. However, outside of such an artificial environment, the robot is dysfunctional. Biologically-inspired Network-Symbolic models do not compute precise 3-dimensional models, but convert image information into an "understandable" Network-Symbolic format, which is similar to relational knowledge models. Feature, symbol, and predicate are equivalent in the Network-Symbolic systems. A linking mechanism binds these features or symbols into coherent structures, and image converts from a "raster" into a "vector" representation that can be better interpreted by higher-level knowledge structures. Logic of visual scenes can be captured in the Network-Symbolic models and used for the disambiguation of visual information.