Vision Based Intelligent Systems to Monitor People's Behaviour and Perception

This work presents the development and analysis of three information systems used to obtain insights about human behaviour in three different contexts. The first one is applied in school context, as a system designed to acquire multimodal data from students performing practical tasks related to STEM (Science, Technology, Engineering, Mathematics) subjects. System architecture is reported as long as multimodal data analysis and a specific analysis based on body poses used to estimate the level of collaboration among students. The second system finds its application in work context. It is a system that is designed to monitor crowded outdoor working areas where humans and vehicles work in proximity. The system uses video features only and it is composed by a set of calibrated cameras. Camera calibration in large outdoor areas was indeed one of the key problems to tackle in order for the system to be functioning. An overview of possible solutions and a novel algorithm for tracking 2D landmarks is reported. The last system reverts the perspective of the other two. Instead of getting information about human behaviour using mainly artificial vision, It exploits human interaction in order to get insights on how the human visual system works. The study of the human visual system is indeed meaningful both for improving the design of future perceptual systems and for the design of systems that require human robot interaction based on visual features. Experiments performed with such a system reported evidence on adaptive properties that become measurable in response to natural statistics, both with relation to local stimulation and to contextual modulation from the surrounding scene.