Acclimatization Calculation Research Based On Multiple Stereovision

One of the main factors which leads to a low robustness in human-computer interaction is that human must adapt themselves to computers, and computers have no ability to adjust to environment. For example, when light and perspective don't meet the requirement of the computer, it will lose information or increase the uncertainty of judgment. In the application of face recognition, light, posture and other factors will affect the recognition rate of the system in certain degree, which lead to a bad result. To avoid the bad influence caused by posture change in face recognition, acclimatization calculation framework is presented in this paper to increase the robustness of computing, it has a small amount of computing and a simple calculation process. Simulation results show that the presented concepts and method in this paper are correct and practical. The work of this dissertation can be summarized as below:First, multiple stereovision method is presented to solve the problem of perception in face recognition. Traditional monocular and binocular vision are researches centered in central projected camera model, under which there are still limitations for face recognition. So, we propose to take multiple stereovisions in face recognition system, by configuring without overlapping to solve the problem caused by posture change or other complex environment and increase the accuracy of perception.Second, a new idea of concept acclimatization is put forward. Acclimatization is defined as the process of computers adjust themselves to changing environment, it takes key factor detection, computing and make selection based on certain decision to reduce the influence caused by circumstance.Third, a new information fusion method based on acclimatization calculation framework is proposed. In this paper, decision level fusion method is one which combines General Regression Neural Network and Dempster Shafer evidence theory. General Regression Neural Network is used to assign the basic probability of images obtained by cameras, then a judgment about the same target can be computed for each camera, finally, Dempster Shafer evidence theory and decision rule presented in our paper are taken to make an overall judgment of the goal.