Comparative Analysis of YOLOv3-320 and YOLOv3-tiny for the Optimised Real-Time Object Detection System

The aim of the proposed work is to detect multiple objects effectively of different classes using YOLOv3-320 in comparison with YOLOv3-tiny algorithm. A sample size of 91 acquired for the study using G power by considering factors effect size, standard error rate, algorithm power as 0.21, 0.05, 0.80 respectively. The data used for this analysis are obtained from the COCO dataset. The configuration file as well as a weights file related to YOLOv3-tiny and YOLOv3-320 consisting of different variables, measurements and architectures for the objects available in the COCO dataset. Non maximum suppression values are found to be significant (p<0.05) leading to efficient detection of the multiple objects. The YOLOv3-tiny algorithm detects the objects with a faster speed of 220 FPS and achieved an accuracy of 70% whereas the YOLOv3-320 detects multiple objects at a time with a speed of 45 FPS with an accuracy of 95%. The proposed YOLO v3-tiny algorithm has performed significantly better for the detection of objects in terms of speed whereas the YOLOv3-320 algorithm has performed significantly better for the detection of objects in terms of accuracy.