Research Of Video-based Abnormal Behavior Recognition In Exam

College entrance examination as a very important exam in China, its fairness has increasingly become the focus of public attention. By2009,electronic video monitoring system has widely covered more than60thousand test sites. To some extent, the system worked. However, current e-examiner can only provide simple functions such as video capture, video callback. It cannot reduce human workload efficiently and it lags behind from the true e-examiner. Based on the background, this thesis aims to analyze the reasons resulting in the lack of intelligent video monitoring system used in the examination environment. This thesis develops a recognition rule that can recognize video-based exam abnormal behavior. Besides, the thesis tests the accuracy of the rule.Firstly, the thesis proposes an extraction algorithm based on the differences of RGB color distribution. The algorithm calculates easily and can get high real-time. Aiming to exclude pseudo desktop areas effectively and ensure the accuracy of extraction of desktop areas, it proposes a filtering rule based on space and area.Secondly, the thesis proposes the improved skin color segmentation method based on YCgCr color space. The new color space can improve the problem that YCgCr color space can be easily affected by the light. It can minimize the bad effect with high real-time. The thesis gives the skin color distribution of elliptical model in new color space and applies the model to the skin color segmentation and ensure the accuracy and real-time. The model is robust under good light conditions. Aiming the results of skin color segmentation, this thesis proposes criterion based on connectivity judgment and the number of hand areas to achieve the recognition of abnormal behavior.Thirdly, by studying the artificial examiner monitoring process and the features of behavior of candidates, the thesis proposes strong real-time criterion based on the number of abnormal behavior of the hands on the desktop area. Meanwhile, by setting the frequency of abnormal behavior variables to achieve the recognition of suspicious candidates, the criterion can both ensure the comprehensiveness of finding suspicious candidates and reduce the artificial workload. Finally, aiming to test the accuracy of the criterion, the thesis designs a video-based exam abnormal behavior recognition demonstration system. Under the Window XP operating system environment, the system applies the Microsoft Visual Studio development tools and uses C++programming language and OpenCV graphics library development to achieve. Using the video datas to test,it finds that the rules can be better when finding the suspicious candidates in college entrance exam environment and can get high real-time.