Research Advanced in Object Detection Based on Deep Learning

Abstract

Object detection has always been a hot topic in the computer vision community. The task is easy for humans, while it is difficult for computers to learn the informational concepts to locate object positions, especially for the sense that multiple objects and cluttered backgrounds are mixed together. Traditional object detection methods are based on manual feature extraction. Coupled with the development in deep learning, both the detection accuracy and speed had made a breakthrough. This paper provides a review of deep learning-based object detection architectures. First, a simple summary on deep learning and its applications area are introduced. Then, this paper pay attention to the basic framework of object detection and its performance index. Then, it classify the object detection algorithms into three categories according to their structure: one-stage, two-stage and point-based. And this paper evaluates their performance as well as their practicality. Finally, several current problems and solutions are provided as suggestions for advanced research on object detection and its applications.