Интеллектуальная технология выявления нестандартного поведения людей в системах видеонаблюдения
Аннотация
A video surveillance system monitors the behavior, activities, or other changing information, usually, of people from a distance through electronic equipment. This thesis presents an intelligent framework video surveillance in an academic environment that takes into account the security and emergency aspects. The framework designed to consist of the two processes: the first one is a tracking system that can follow targets with identifying a set of features to descriptive information of each target. The second one is a decision system that can realize if the activity of the target track is normal or abnormal then energizing a real-time alarm when recognized abnormal activities.
The system proposes an abnormal human activity classification by the detection motion algorithm based on the Gaussian mixture model (GMM) followed by the Fuzzy C-Means (FCM) segmentation algorithm. Combined (HARRIS-SIFT) algorithms together to extract features and the Kalman Filter for tracking targets. Finally, the K-Nearest Neighbor (KNN) algorithm used for the classification of the activities that belong to three different datasets tested: (1) Weizmann standard dataset, (2) KTH standard dataset.
The results show the efficiency of the system. The results of the test show that the datasets have the accuracy ratio is (97%), the detection ratio is (%97) and the false alarm ratio is (4%).
The system proposes an abnormal human activity classification by the detection motion algorithm based on the Gaussian mixture model (GMM) followed by the Fuzzy C-Means (FCM) segmentation algorithm. Combined (HARRIS-SIFT) algorithms together to extract features and the Kalman Filter for tracking targets. Finally, the K-Nearest Neighbor (KNN) algorithm used for the classification of the activities that belong to three different datasets tested: (1) Weizmann standard dataset, (2) KTH standard dataset.
The results show the efficiency of the system. The results of the test show that the datasets have the accuracy ratio is (97%), the detection ratio is (%97) and the false alarm ratio is (4%).