Web-Based Intelligence Surveillance System to Detect Area Intrusion Using RT-DETR with Limited Dataset

Property intrusion remains a global concern, often driven by motives such as theft. While human-monitored CCTV systems are commonly employed for prevention, they are inherently limited by human capabilities and challenged by factors such as the increasing number of areas requiring surveillance and other factors. Advancements in Machine Learning (ML) offer promising solutions to overcome the limitations of human-monitored CCTV systems. Object detection, a potent ML method, enables real-time identification and tracking of objects through CCTV cameras. RT-DETR is an example of the excellent object detection model in terms of both accuracy and inference speed. The intention of this research is to implement RT-DETR on a web-based application to prevent property intrusion with such limited dataset quantity, for live video stream using NVIDIA RTX 3050 Mobile Laptop, which is a common and often considered as a mid-range GPU hardware. In addition to customizable surveillance zone definition, the application incorporates features to minimize false alarms, ensuring a more reliable and efficient security solution. Additionally, the application is equipped with automatic video recording as evidence of intrusion evidence.