Document Details

Document Type : Thesis 
Document Title :
Abnormal Event and Behavior Detection Using Scene-Based Domain Generalization Technique
كشف الأحداث والتصرفات الغير عادية في المشاهد باستخدام تقنية تعميم النطاق
 
Subject : Faculty of Computing and Information Technology 
Document Language : Arabic 
Abstract : In the last few years, surveillance cameras have been massively distributed in public and private facilities due to security concerns and ensuring the safety of the society, creating a need to monitor unexpected events and behaviors in a scene. An intelligent automated approach is highly required to detect anomalies from the video scene to save the time and cost required by laborers to detect the anomalies manually from monitor screens. In this research, a deep learning-based approach proposed to detect abnormal events and behaviors from surveillance videos in crowd scenes using scene-based domain generalization technique. Thus, by using the keyframe selection method to extract the keyframes that contain essential information from video frames. Then the selected keyframes are used to generate a spatio-temporal entropy template that represents the motion region. After that, the obtained template is supplied to the pre-trained 'AlexNet' network to extract deep high-level features. Eventually, the Relieff feature selection method is applied to select the appropriate features, then based on these features; the study generates a classification model via Support Vector Machine (SVM) classifier. The model is evaluated on six different public datasets, along with two datasets constructed in this study, named 'Collected Dataset' and 'Validation Dataset' where both datasets consist of normal and abnormal events and behaviors videos. The results showed that the proposed method with domain generalization outperformed other existing methods by achieved detection accuracy ranging from (87.5% to 100%) and demonstrated the generality of the proposed model for the detection of anomalies from different domains at a high accuracy rate of 97.13%. 
Supervisor : Dr. Salma Cumin 
Thesis Type : Master Thesis 
Publishing Year : 1441 AH
2020 AD 
Added Date : Saturday, June 13, 2020 

Researchers

Researcher Name (Arabic)Researcher Name (English)Researcher TypeDr GradeEmail
آلاء عطالله المزروعيAlmazroey, Alaa AtallahResearcherMaster 

Files

File NameTypeDescription
 46386.pdf pdf 

Back To Researches Page