With　the　emergence　of　a　large　number　of　artificial　intelligence　technologies,　deep　learning　has　become　the　key　technology　in　computer　vision　area.　Object　tracking　is　one　of　the　most　important　technology　in　the　field　of　computer　vision.　Thus　we　studied　about　tracking　algorithms　and　proposed　a　method　mainly　hopes　to　solve　the　occlusion　problem　in　complex　tracking　scene.　Using　object　detection　algorithms　based　on　deep　learning　to　increase　the　speed　of　associations　and　improve　tracking　effect.　It　can　return　the　position　of　the　tracking　object　unsupervised.　Then　extract　features　to　store　in　features　library,　so　that　the　prediction　of　trajectory　whose　features　can　highly　be　matched　is　more　accurate　and　the　associations　are　more　reliable.　Experiment　shows　our　tracking　algorithm　combines　with　detection　algorithm　based　on　depthwise　separable　convolution　networks　not　only　has　a　smaller　and　faster　model　but　also　achieved　a　robustness　and　real-time　tracking　in　scene　where　objects　are　under　occlusions.　©　2019　IOP　Publishing　Ltd.　All　rights　reserved.