INDOOR OBJECT DETECTION COMPARISON USING YOLOV8, NANODET-PLUS, AND DETECTION TRANSFORMER MODEL

The development of object detection algorithms keeps on moving at a faster speed every year. This occurence is unique because a robust object detection method will undoubtedly play a significant role in various implementations such as mobile robots and humanoid robots, especially for indoor objects, as the robot has more narrow and limited space to move. In addition to that, the poses of the camera and entities contin-ue to change following the robot’s motion so that the detection becomes less accurate. Therefore, the prerequisite for creating such robots is detecting the correct object at the right place. Thus, robust object detection is required. This research compared some of the newest object detection algorithms, specifically YOLOv8, Nanodet-Plus, and Detection Transformer (DETR). All methods are trained under the same number of epochs, which is 20. Moreover, each method also used pre-trained models, namely YOLOv8x, NanoDet-Plus-m-1.5x-416, and detr resnet50 in the same number of datasets, namely 2213 images of objects in a room divided into eight classes. The experiment results show that YOLOv8 has a slight advantage with achieving mAP50 or a mean Average Precision calculated at Intersection over Union (IoU) 0.5 of 99.1% compared to Nanodet-Plus with 92.9% and Detection Transformer with 93.2%.