Overview of visual object detection technology based on deep learning

Thesis link ：http://www.cjig.cn/jig/ch/reader/view_abstract.aspx?flag=2&file_no=202201170000006&journal_id=jig

Visual object detection aims to locate and recognize objects in images , It belongs to one of the classic tasks in the field of computer vision , It is also the premise and foundation of many computer vision tasks , On autopilot 、 Video surveillance and other fields have important application value , It has been widely concerned by researchers . With the rapid development of deep learning technology , Great progress has been made in target detection . First , This paper summarizes the basic process of deep target detection in the process of training and testing . The training phase includes data preprocessing 、 Detection network 、 Label allocation and loss function calculation , In the test phase, the trained detector is used to generate the test results and post process the test results . then , This paper reviews the visual object detection methods based on monocular camera , It mainly includes the method based on anchor box 、 The method without anchor box and the method of end-to-end prediction . meanwhile , Some common sub module design methods in target detection are summarized . After the visual object detection method based on monocular camera , A visual object detection method based on binocular camera is introduced . On this basis , The research progress of monocular target detection and binocular target detection at home and abroad are compared , The development trend of visual object detection technology is prospected . Through summary and analysis , It is hoped that it can provide reference for relevant researchers to carry out relevant research on visual target detection .

Visual object detection is a classical task of computer vision , Aimed at Locate the position of the existing object in the image and identify the specific class of the object other . Object detection is the focus of many computer vision tasks and related applications Basis and premise , Directly determine the performance of related visual tasks and applications Good or bad . therefore , The technology of visual object detection has been widely accepted by the academic circles 、 Industry And other fields 、 And even the wide attention of governments all over the world . learning Technical field , Object detection has always been a major computer vision conference and journal It's one of the research hotspots in the world , There are a large number of papers on target detection every year publish . According to Google academic , Researchers are close to 10 Over the years Published papers on standard detection 15 000 Yu Wen . In industry , At home Foreign technology giants ( Such as Google 、 facebook 、 Huawei, Baidu, etc )、 Startups department ( Such as soup 、 Open eyes, etc ) They have invested in the field of target detection A lot of human and financial resources . meanwhile , Target detection technology is a new generation Important common key technologies of artificial intelligence , Countries all over the world are competing .