Design of IOT based violence detection systems using lightweight virtualization mechanism

Today, violent events have become more common in public areas, schools, and sports arenas. The early detection of these events, as well as catching the perpetrators, is vital for social order. Today's Internet of Things (IoT) technologies help to overcome these problems, and for this purpose, this thesis proposes a cloud computing-based IoT violence detection system, which is able to help security forces and related institutions. This violence detection system uses edge computing technologies, and in addition, the edge devices use Docker, which is a lightweight virtualization technology, for both software execution and system sustainability. The artificial intelligence (AI) module, which detects violence, uses deep learning models that are suggested in the literature. The deep learning module includes (CNN) with LSTM, BiLSTM, Gated Recurrent Unit (GRU) and Recurrent Neural Network (RNN) models that use a pre-trained MobileNet. The images taken by the edge devices are locally evaluated, and the urgent cases are forwarded to the relevant institutions, or to the cloud environment. Computing at the edge leads to a great reduction in the increase of cloud load, and the Docker technology allows the installation and management of related AI modules on edge devices in a fast and flexible way. The training dataset consists of violent and non-violent images. In the thesis, AI modules were trained outside the edge layer, and then the trained modules were loaded on edge devices, using container technologies. The experiments were done in local environments and the performance of the system was then analysed. In addition, the resources consumed by the edge devices were examined, with an additional discussion regarding edge device performances. In practice, conventional violence detection systems may have disadvantages in terms of installation and execution costs. Moreover, these costs can increase even more with the use of IoT technologies such as cloud computing. Attention should always be paid to optimizing the cloud system load and customer cost, which was proven in this work, since the proposed IoT system model has several gains and advantages, with a clear practical and commercial potential.