An Intelligent Image-Based Human Intrusion Detection using ImageAI and Local SMS Alerting

Abstract

Security remains a critical global concern due to the rising rate of crime and intrusions. In response, Video Surveillance Systems (VSS) have evolved from passive monitoring solutions into intelligent, real-time surveillance platforms, powered by advancements in computer vision, machine learning, and telecommunications. Modern systems increasingly aim not only to record footage for forensic review but also to deliver real-time alerts and facilitate immediate intervention. However, a major limitation of many commercial surveillance systems is their inability to accurately classify detected motion. These systems often fail to differentiate between human intruders and non-human movements, leading to inefficient memory usage and frequent false alarms. This paper presents the design and implementation of an Image-Based Intrusion Detection System Using ImageAI and locally triggered SMS notification. The system captures images upon motion detection via surveillance cameras and stores them using the FTP protocol. The stored images are then analyzed using a pre-trained deep learning model to classify and recognize the detected objects. When ImageAI detects a 'person', a flag is set in the MySQL database, which the Diafaan SMS Server polls at intervals to trigger SMS alerts to pre-registered users without requiring internet connectivity. This hybrid architecture provides both cost-effectiveness and timely intervention during live intrusion scenarios while minimizing false positives and unnecessary data storage.