Development of specific area intrusion detection system using YOLO in CCTV video

Abstract

Recently, many things are monitored with CCTV. It needs manpower to monitor. Therefore, we propose a system that stores the object’s data when the object enters a specific area. Generally, You Only Look Once (YOLO) method is used for object detection. We propose a system that increases the recognition rate by using new dataset because the recognition rate may not be accurate when using existing VOC weight, and can apply to intrusion detection. The Proposed model with the training data created on the particular conditions precisely detects proper objects and reduces effect by a variance of frame and brightness. The object was found more precisely with newly trained weights instead of using VOC weight. The dataset is made to fit to the specific situation, and it can be used for genera situation. When using the VOC weight, object detection was affected by the frame change and the brightness change of the light, but the effect was reduced by using new weight.