Comparative Analysis of CNN-Based Object Detection Models: Faster R-CNN, SSD, and YOLO

Abstract

Target detection is widely used in the current environment. With the rapid development of deep learning, innovative models like Convolutional Neural Networks (CNNs) were born. CNNs have been widely used in many practical applications for object detection, since CNNS outperform traditional models in terms of speed and accuracy. This paper first introduces three well-known CNN-based target detection models: Region-based Convolutional Neural Network (Faster R-CNN), Single Shot Multibox Detector (SSD) and You Only Look Once (YOLO). Then this paper gives data on speed, accuracy and resource consumption. Based on these data, the advantages and disadvantages of these three models in different scenarios are systematically and comprehensively evaluated and analyzed. The purpose of the paper is to give researchers a deep understanding of the different characteristics of the three models and which model should be used best in what situations by examining their strengths and limitations. Finally, this paper analyzes the different characteristics of the three models to facilitate the researcher's possible subsequent improvements.