COMPARISON OF OBJECT DETECTION METHODS IN COMPUTER VISION

Abstract

Object detection is a fundamental task in computer vision, with applications ranging from autonomous driving to surveillance systems. This article presents a comprehensive comparison of various object detection methods. The methods evaluated include traditional methods such as logistic regression and SVM, as well as state-of-the-art deep learning architectures such as YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), FPN (Feature Pyramid Network), RetinaNet. YOLO prioritizes real-time processing speed, making it ideal for applications demanding swift detection, such as self-driving cars. However, this emphasis on speed might compromise accuracy when compared to other methods. SSD offers a compelling balance between speed and accuracy, achieving faster processing than some methods while maintaining good detection capabilities. FPN solves the problem of detecting objects at different scales in an image. Using the Feature Pyramid Network function, it can effectively analyze both small and large objects in the same structure. RetinaNet, on the other hand, focuses on improving accuracy by introducing a focal loss function that mitigates the class imbalance problem, a common obstacle in object detection tasks where certain classes significantly outperform others. For object classification, YOLO utilizes the cross-entropy loss function. This function measures the difference between the predicted probability distribution of an object's class and the actual class distribution. Minimizing this loss during training guides the model to make more accurate class predictions.

The paper analyzes the existing object detection methods and conducts an experiment with the YOLOv5 model trained on the COCO dataset.