期刊名称：International Journal of Advanced Computer Science and Applications(IJACSA)
印刷版ISSN：2158-107X
电子版ISSN：2156-5570
出版年度：2022
卷号：13
期号：2
DOI：10.14569/IJACSA.2022.0130293
语种：English
出版社：Science and Information Society (SAI)
摘要：Studies on abnormal behavior based on deep learning as a processing platform increase. Deep learning, specifically the convolutional neural network (CNN), is known for learning the features directly from the raw image. In return, CNN requires a high-performance hardware platform to accommodate its computational cost like AlexNet and VGG-16 with 62 million and 138 million parameters, respectively. Hence in this study, four CNN samplings with different architectures in detecting abnormal behavior at the gate of residential units are evaluated and validated. The forensic postures, with some other collected data, are used for the preliminary step in constructing the criminal case database. High accuracy up to 97% is obtained from the trained CNN samplings with 80% to 97% recognition rate achieved during the offline testing and 70% to 90% recognition rate recorded during the real-time testing. Results showed that the developed CNN samplings owned good performance and can be utilized in detecting and recognizing the normal and abnormal behavior at the gate of residential units.
