Reseach On Abnormal Behavior Detection In Videos And Software Implementation

With the improvement of the human’s life, more and more people pay close attention to the safety. And there is a rapid growth in the security system which is used for social harmony and protecting the life and property. As the core component in the field of security monitoring, the abnormal behavior detection plays a role key. Abnormal behavior detection which is to detect the unusual or suspicious behavior in the monitoring video and to alarm, is applied in the airport safety system, subway security, residence monitoring, campus surveillance system and so on.The first stage of abnormal behavior detection is to the automatic detecting some defined behavior by the computer. As the direction becoming matured, some stable products have entered the marketplace. But this kind of system must be trained with some special behavior in advance. In recent years, abnormal behavior detection applied anywhere is growing research direction. Though there are some achievements in the direction, it haven’t been matured.This thesis focuses on the abnormal behavior detection in any scene. Analyze video captured by a fix camera in any scene, and train to get a classifier, which regards frequent behaviors as the normal behavior, others as abnormal. Then use the classifier to detect the behavior from the real-time camera. The work of the thesis is as follows:1. We proposed a new outline descriptor and the corresponding extracting method. The method combines the interframe difference and background difference to extract the region of motion. To solve the uncontinuity of object’s outline, we propose the outer tangent line detecting method. Design a simple system to achieve tracking, object merging and object departing.2. To measure the descriptor with different dimensions, we proposed a markov model whose status space bases on the cluster centers of Bo F. Apply the descriptors to train the model, and adopt the designed corresponding criterion to judge the behavior.3. We used a 3D sliding window to detect behavior, and proposed a gaussian mixture model based on cluster to judge the behavior.4. Base on the algorithm of markov model, we developed a software based on MFC and Open CV, and achieved real-time testing.The experiment results proved the proposed outline sequence descriptor and markov model can accomplish abnormal behavior detection in complicated scene, and the gaussian mixture model based on cluster can detect the abnormal behavior under the multiple objective scene.