Research works

Image Presentation

  • Implicit human computer interaction
  • Geometric Constraints of Cameras And People Locating
  • Experimental method for Computer Vision Based on Virtual Data

Implicit human computer interaction is an emerging research area of human computing. The implicit interaction system inputs from the understanding of users intension via automatic detecting and analyzing user actions. The output of the system is the adaptive service to the user. Dynamic context acquisition and awareness sustains the whole implicit interaction in understanding and service. This project will focus on dynamic context acquisition and awareness for bridging the semantic gap between action and intention, which is a pervasive and challenging problem in implicit interaction. A bottom-up method will be used to acquire the dynamic context of present action and its environment and to search for the relative semantic of the context for guiding the low-level sensor information processing and event detection. The implicit interaction system will approach to the understanding of users intention by capturing semantics of users actions in different levels by updating the context associated with actions. In this project, the classification of contexts, the relationship among contexts, the modeling of contexts and the semantic acquisition of contexts will be investigated for dynamic context aware. The theories of combining bottom-up information processing with top-down context guidance, and the uncertainty reasoning methodologies for context and ontology based semantic will also be studied.

The human body locating has been one of the most popular subjects in computer vision, which is the basis of the human-computer interaction. We proposed four types of geometrical relationship between the camera imaging plan, pround plane and the human targets. Spatial geometric relationship constraints reduce camera erected posture, improving the applicability of the algorithm. Taking the human bodys as locating targets, we used the method of probability, computing the optimal positions at the time of each frame.

In computer vision field, traditional experiments are based on real experimental data, or use shared databases. However, in many cases the real data can not meet our needs, for example we can not get pure ture value without error. We proposed a virtual experiment method based on virtual experimental data. With the rapid developmentof computer graphics int eh past few years, Scenes and equipments required by the experiments can be simulated with the 3D animation designing softwares, to provide the most accurate expeirmental environments.