Analisa Kemampuan Algoritma YOLOv8 Dalam Deteksi Objek Manusia Dengan Metode Modifikasi Arsitektur

Abstract

Object detection is a skill that can be taught to a machine with the help of a camera sensor to capture a digital image. By using the YOLO algorithm we can teach machines to detect, for example, humans. Much research in object detection has been carried out previously using different algorithms and methods and also on different objects and images. In this research, a method was carried out to modify the architecture of YOLOv8 in the head section to be used to detect human objects in grayscale images. The training process was carried out 4 times using the default architecture, Model 1, 2 and 3 architecture. With the default model results, the mAP value was 76, Model 1 had an mAP value of 66, model 2 had an mAP value of 81 and model 3 produced an mAP value of 80. From the research carried out modifications The YOLOv8 architecture in the head section can influence the training results and produce a better model than the default architecture which only produces an mAP value of 76. The best results were obtained in model 2 with layers used of 40x40x512xW resulting in a model with an mAP value of up to 81.