Abstract:For decades,image feature detection and matching has been the foundation of computer vision.Without feature detection and matching,there would be no visual tasks such as SLAM,Sfm,AR,image retrieval,image registration,or panoramic images.Based on the review of classic detection algorithms in the past decades,this paper describes the latest progress in image feature detection and matching after the introduction of machine learning algorithm led by deep learning.The survey includes all the key points such as feature points,local descriptor,global descriptor,matching and optimization,and end-to-end framework,and compares the merits and demerits of each algorithm.In summary,facing the requirements of wide baseline,real-time,and low computing load detection from the industrial sector,image feature detection and matching is still a hard task.The multitasking global framework which fuses feature points,local descriptor,global descriptor,matching and optimization has become the trend of future research.