PENERAPAN METODE YOLO DALAM DETEKSI OBJEK MANUSIA PADA ROBOT PENERIMA TAMU

Abstract

The guest reception robot is designed to enhance user interaction in various environments such as hotels, offices, and shopping centers. One of the main challenges in developing this robot is achieving fast and accurate human detection for real-time operation. This study implements the You Only Look Once (YOLO) method, specifically YOLOv5, for human object detection in the guest reception robot. YOLOv5 was chosen due to its advantages in detection speed and accuracy. The research stages include system design, data acquisition, image processing, model implementation, and performance analysis using mean Average Precision (mAP) and frames per second (FPS) metrics. Based on the research results, it can be concluded that the YOLOv5 method is capable of detecting human objects with high accuracy and fast inference time. The model shows optimal performance under bright lighting conditions, achieving 95% precision and 92% recall, with an inference speed of 30 FPS. Although accuracy decreases by approximately 4% under dim lighting, the model remains consistent with 88% precision and 85% recall. Additionally, the YOLOv5 model was tested in various scenarios, including changes in lighting, object distance, and crowd density within the frame. The results indicate a slight decrease in accuracy when objects are more than 5 meters away or when there are more than three people in a single frame, leading to an increase in false positives. Architecturally, YOLOv5 divides the input image into a 7x7 grid, where each grid cell predicts class probabilities and bounding boxes, enabling efficient and accurate detection. Thus, YOLOv5 is an effective solution for human object detection in various environmental conditions, though there is still room for improvement in specific scenarios such as low lighting and high density