Anomaly detection and alerting video surveillance system for home security using deep learning techniques

Abstract

The usage of video surveillance systems has been increasing rapidly, and the demand for embedding analytical capabilities to the systems has also been increasing accordingly. The development of intelligent surveillance systems with less human interaction and more reliability has become highly valuable. However, developing a video surveillance system for home security that can detect and alert suspicious and abnormal incidents in real-time has been a research challenge due to the complexity of identifying anomalies from real-time video footage and developing a real-time alerting system. This thesis proposes a deep learning and IoT-based anomaly detection and alerting system for home security to address this challenge. The proposed solution aims to produce an alert of defined suspicious incidents in a home environment by analyzing surveillance video data and sending it to the owner. The system comprises two key components: an anomaly detection system and an alerting system. The anomaly detection system build with deep learning techniques and best model is deployed identify any pre-defined anomalies in the video data. The alerting system, which uses suitable IoT techniques, sends an alert via a MQTT protocol to the homeowner when an anomaly is detected. In conclusion, this thesis proposes an innovative deep learning and IoT-based anomaly detection and alerting system for home security that is expected to provide an accurate alert of any pre-defined anomaly in a home environment by analyzing the video data. This thesis contributes to the existing literature by exploring the evolution of intelligent video surveillance systems, anomaly detection in surveillance videos, alerting systems, and trends of anomaly detection and alerting video surveillance systems in home security. The limitations and technology/methodology used in this thesis are also discussed, and the research problem is defined.