Implementasi Metode Yolo pada Deteksi Objek Manusia

Until now, the problem of theft of motorbikes and livestock in North Sumatra is still quite high. Locations for motorbike theft can occur in many places such as schools, homes, parking lots, offices and so on, while for livestock it can occur on pastures and in pens during the day or night and the perpetrators are men. To make this theft a success, various modes are used in varying human positions, from sitting, squatting to standing. To help overcome this, several object detection methods have been developed such as Background Subtraction, Template Matching, Histogram Oriented Gradient (HOG), Deformable Part-based Model (DPM) and Viola Jones (VJ). Of the many methods that have been used, there are still shortcomings, namely in terms of time, accuracy and various human positions. For this reason, research was carried out with the aim of improving the time and level of accuracy in detecting human objects using the YOLO method. The research stages carried out in this research include literature study, collecting data, determining training and test data, creating programs, training, and testing. From the trials carried out, it is known that YOLO can detect humans in various positions with a mAP value of 0.99 and an average detection time of 810.01 ms.