Panoramic　video　is　a　video　collected　by　multi-channel　fisheye　lens　and　obtained　by　splicing　and　fusion.　Compared　with　traditional　video,　panoramic　video　has　better　integrity,　immersion　and　presence.　Therefore,　panoramic　video　has　been　widely　used　in　the　video　field.　In　order　to　meet　the　client＇s　personalized　tracking　needs　in　the　panoramic　broadcast　system　and　panoramic　monitoring　system,　this　system　designs　and　implements　a　people-tracking　system　based　on　personalized　distribution　in　the　panorama,　and　provides　personalized　services　for　the　client　to　select　target　objects.　Compared　with　the　traditional　manual　sliding　screen　for　viewing,　this　automatic　live　broadcast　mode　with　the　target　person　as　the　video　center　greatly　improves　the　user　experience.　This　system　uses　the　method　of　deep　learning　to　process　the　panoramic　video.　After　the　video　is　processed　by　the　detection　unit　based　on　YOLOv3　and　the　recognition　unit　based　on　ResNet50,　the　coordinates　of　the　target　person　in　the　panoramic　video　will　be　obtained.　According　to　the　different　objects　selected　by　each　client,　the　local　video　of　the　corresponding　target　object　is　distributed　to　each　client　through　a　streaming　media　server　to　meet　the　personalized　tracking　needs　of　different　clients.