Abstract:
All across the world, surveillance systems are facing many challenges especially in
monitoring large public spaces, detecting threats, and promptly and accurately
responding. Due to problems such as high latency, poor real-time analysis, and threat
detection, conventional systems are less reliable in critical situations. The IRIS: AI-
Based Surveillance System uses TensorFlow and YOLO v8 for real-time object
detection which makes it a reliable surveillance system. This hybrid solution can
manage multiple video streams simultaneously due to its low-latency performance and
scalability. To enhance the performance of the system and ensure the model is trained
on diverse datasets, strict pre-processing methods like frame shrinking, noise reduction,
and data augmentation were used. Also, a special way of changing the frames we input
and filtering them made detection more accurate and brought down computing
overhead. The system has been tested extensively and is accurate at 83% for fire, 80%
for people, and 68% for violence detection making it suitable for many applications.
IRIS system is a remarkable progress in automated monitoring and offer a
comprehensive and effective solution to the current secure without any human
involvement.
