In　order　to　solve　the　problem　that　the　visual　observation　actions　of　the　tower　controller　during　the　take-off　and　ground-off　phases　of　the　aircraft　can　only　be　recognized　manually　by　the　management　personnel,　the　visual　observation　action　recognition　software　of　the　tower　controller　is　established.　In　the　PyCharm　compilation　environment,　based　on　Python,　call　functions　in　the　OpenCV　library　to　obtain　video　frames,　use　functions　in　the　Dlib　library　to　implement　face　detection,　implement　face　recognition　based　on　the　Caffe　framework,　and　implement　face　rotation　tracking　recognition　based　on　CamShift　functions.　Display　the　results　of　the　visual　observation　action　recognition　of　the　tower　controller　in　graphic　form.　The　results　show　that　the　software　has　the　advantage　of　extracting　the　deep　features　of　the　image,　realizes　the　function　of　automatically　identifying　the　visual　observation　actions　of　the　tower　controller,　with　high　recognition　accuracy,　and　provides　a　powerful　automated　tool　for　tower　site　management　and　incident　investigation.