Sports　injuries　can　be　a　major　problem　for　athletes.　Therefore,　sports　injury　protection　has　become　a　key　focus　of　attention　in　sports　and　medical　circles.　With　widening　participation　in　sport,　related　injuries　can　have　an　impact　at　national　level.　However,　many　areas　around　the　world　lack　adequate　medical　resources.　It　takes　more　time　and　money　for　local　people　to　get　to　the　nearest　rehabilitation　department　or　physical　therapy　studio.　Artificial　intelligence　(AI)　has　undergone　vigorous　development,　leading　to　increased　computing　speed　and　accuracy.　Nowadays,　two-dimensional　image　signals　can　be　used　for　body-posture　recognition.　This　research　is　based　on　the　Openpose　limb-detection　AI　model,　which　has　corrective　exercise　training　elements　and　uses　functional　motion-detection　technology　as　the　diagnostic　basis　(combined　with　physical　therapists’　clinical　knowledge　of　rehabilitation　interventions).　We　propose　a　2-D　imaging　physical　health　detection　system.　The　system　is　divided　into　four　main　parts:　a　mobile　app　user　interface,　the　computing　server,　professional　interface　and　database.　A　user　video　is　recorded　by　the　app.　The　computing　server　then　calculates　keypoints　of　the　human　body　through　Openpose　and　converts　these　data　into　clinical　test　indicators.　Finally,　health　indicators　are　displayed,　and　the　app　registers　action　scores.　The　professional　interface　presents　users　with　health　feedback　and　recommends　suitable　rehabilitation　videos.　The　computing　server　is　divided　into　two　parts:　limb　detection　and　the　detection　system.　The　limb-detection　aspect　is　divided　into　four　parts:　sample　collection,　video　processing,　keypoint　processing　and　results　comparison.　The　detection　system　is　subdivided　into　system　architecture　and　a　health　evaluation　(forming　the　basis　for　video　recommendations).　The　key　contributions　of　this　paper　are　that　the　proposed　system　can　calculate　body　posture　and　automatically　detect　the　physical　condition　and　health　of　the　body.　In　addition　to　reducing　dependence　on　professional　human　resources,　the　system　can　also　save　the　trouble　of　manual　angle　measurement　in　traditional　physical　therapy.　©　2021,　The　Author(s),　under　exclusive　licence　to　Springer　Science+Business　Media,　LLC,　part　of　Springer　Nature.