Qualitative Reasoning Engine for Visual Scene Understanding in Cognitive Vision Systems

In the process of development of cognitive vision systems, which can be embedded in real situations and that can interact with the environment in an autonomous manner should be capable of learning from visual scenes in a manner closer to human visual learning. Yet this has not been realized fully. A research has been conducted to incorporate a qualitative reasoning engine into cognitive vision systems to learn from visual scenes. The reasoning engine is capable of analyzing the observing visual scene in terms of qualitative object-object spatial and temporal relations. By considering the learnt relations as examples, generalized rules are learnt using inductive logic programming. These learnt rules for a specific qualitative relation depict the conditions that should be satisfied in general terms. Hence these rules can be used in any similar scenario to identify the qualitative relations that exist between any two objects. This work has a great potential in developing agents that can learn autonomously from visual scenes in a manner closer to human visual learning and thereby produce agents that can learn to cognitively mimic the object manipulations in the observed scene.