Exploring Object Detection: Datasets, Metrics and Algorithms

Abstract

Object detection, as one of the basic issues in the computer vision (CV) field, has huge application value in realm such as autonomous driving and face recognition. Since the object differences in images in different scenes, angles, and lighting environments are too great, the number and size of objects are also different, those greatly increase the difficulty of detection. Therefore, how to detect objects quickly and accurately has attracted widespread attention from researchers. This article provides a review from the perspectives of data and algorithms in accordance with the order of technological development. It mainly introduces commonly used data sets, evaluation indicators, traditional object detection algorithms, and object detection algorithms based on deep learning. Finally, this paper will discuss future research directions, explore what difficulties still exist in the field of target detection, and how researchers should further improve the performance and generalization capabilities of target detection to cope with more complex and changeable scenarios and needs.