Due　to　the　emergence　of　deep　learning,　face　recognition　has　made　remarkable　achievements.　But　in　the　traditional　video　surveillance　scene,　because　the　monitoring　object　is　not　controlled　and　the　face　posture　changes　frequently,　the　actual　face　image　may　be　a　negative　face　image,　which　greatly　reduces　the　accuracy　of　face　recognition.　And　the　traditional　monitoring　system　does　not　have　an　automatic　warning　function,　which　wastes　a　lot　of　human　resources.　This　paper　focuses　on　the　realization　of　a　real-time　and　efficient　face　matching　system　in　multi-pose　changing　scenes.　The　system　mainly　uses　the　MTCNN　algorithm　for　face　detection　and　alignment　and　uses　a　generative　confrontation　network　to　make　face　positive　to　eliminate　the　influence　of　posture　change.　The　feature　extraction　network　uses　lightweight　MobileFaceNet　as　the　backbone　network　for　feature　extraction,　and　the　feature　comparison　is　carried　out　by　cosine　similarity.　Experiments　show　that　the　multi-pose　face　matching　system　can　greatly　improve　the　speed　of　feature　extraction,　and　has　good　accuracy,　real-time,　and　robustness.　©　2021　IEEE.