Performance Comparison of Object Detection Neural Network Models Based on Accuracy and Latency

Aim: The purpose of this study is to compare the object detection performance of You Only Look Once V4 (YOLOv4) and Single Shot Multibox Detector (SSD) algorithms with respect to metrics like accuracy and latency. Materials and method: Twenty sample photos in all, from different classifications and labels, were gathered. These samples were divided into training dataset (60 %) and test dataset (40 %). To measure the performance, values for accuracy and latency were computed for YOLOv4 and SSD with G power 0.8. Result: The accuracy in prediction of the object in the image was higher in the YOLOv4 algorithm (97 %) compared to the SSD algorithm (84 %). After running a t-test on an independent sample of the two groups under consideration. It is observed that YOLOv4 reported greater preference than the SSD algorithm having p value 0.166 (p>0.05). It was proven that the YOLOv4 reported greater preference than SSD in terms of accuracy.