Human activity recognition in video using diagrammatic reasoning and QSTR

Cognitive Vision is an area of research with emphasis on the development of robust computer vision systems by endowing them with a cognitive faculty, such as an ability to learn, adapt and develop new strategies for analysis and interpretation . Cognitive vision systems are expected to possess capabilities akin to human Vision. Humans perceive spatio-temporal information through Vision, together with other background information. Many a time, such spatio-temporal information is incomplete and imprecise. An intelligent brain manipulates these with cognitive faculties and learns from experiences. Perceived space-time information may be defined with both quantitative values and qualitative descriptions. Every day, spatio-temporal reasoning is mainly qualitative.Consequently, in cognitive vision research, qualitative information processing is extensively applied. Over the years, many approaches have been explored for spatio-temporal abstractions toward varied spatial problem analysis; qualitative spatial and temporal reasoning (QSTR) being the most predominant. Some form of imagination or diagrammatic representation is the basis for any level of human visual reasoning visualization eases human problem-solving efforts . Many successful methodologies involving diagrammatic reasoning (DR) and heterogeneous approaches combining formal and informal reasoning techniques have been explored. However, much more remains to be done to exploit the advantages of the representative power of diagrams and interpretation of situations in visual problem-solving. Gem Stapleton's work on the design of diagrammatic inference rules using insights into what humans find accessible, provides instances for precise modeling of situations using diagrams . The work demonstrated accuracy in problem analysis through formal reasoning techniques based on a situation's actual physical organization using diagrams. However, the diagrammatic inferences are rule-specific.