Object Detection in Real-Time Surveillance Using Deep Learning-Based YOLO Framework

The continual growth of the computer vision and artificial intelligence fields overtime have provided the basis for more efficient and accurate surveillance. This paper examines the use of YOLO, a deep learning model that is used in the detection of objects, for use in real-time surveillance systems. The added benefits of identifying and labeling multiple objects in a single pass forward makes YOLO ideal for real-time surveillance. Unlike other approaches to object detection such as region-based methods or methods with multiple stages of detection, YOLO cast object detection as a regression problem where boxes and class probabilities are predicted jointly. This leads to a far improved performance when compared with their slower counterparts without the sacrifice of precision. In this work, the YOLO model is finetuned on a large database of surveillance videos to improve object detection at low lighting and in the presence of occlusion and crowding. The developed system has a high detection speed and accuracy; the application areas will include uses such as security inspection, traffic regulation, and anomaly detection. This is done to affirm that using YOLO is more efficient and faster than the other object detection algorithms in real-time surveillance tasks. The results indicate directions for further improvements to guarantee public safety and improve efficient surveillance systems.