Comparative study of convolutional neural network object detection algorithms for vehicle detection in image processing

Abstract

The introduction of Convolutional Neural Networks (CNNs) has revolutionised the field of computer vision, particularly in the domain of object detection. This thesis presents a comprehensive comparative study of CNN-based object detection algorithms for vehicle detection, aiming to explain their intricate architecture, differentiation in methodology, and characteristics in performance. Beginning with an exploration of the theoretical underpinnings of CNNs and object detection, a solid foundation is established upon which the comparative analysis is built. Practical implementation and experimentation play a pivotal role in this study. The three different object detector algorithms being evaluated in this study within the MATLAB® development environment are Single Shot MultiBox Detector (SSD), Faster Region-Based Convolutional Network (R-CNN), and You Only Look Once (YOLO v3). Through research and experimentation, the strengths and limitations of each algorithm are provided. The findings of this comparative study not only contribute to the academic understanding of CNN-based object detection but also offer practical guidance to practitioners and researchers in selecting appropriate algorithms for specific domain applications such as vehicle image processing. Furthermore, this thesis serves as a roadmap for future research endeavours, highlighting areas for further exploration and improvement within the realm of object detection using convolutional neural networks.