The　depth　map　sequence　collected　by　Kinect　can　be　transmitted　through　the　existing　video　compression　coding　platform,　which　saves　a　lot　of　computing　resources　and　saves　a　lot　of　computing　resources　than　the　conversion　of　the　point　cloud　data.　However,　the　depth　map　sequence　collected　by　the　Kinect　has　an　unstable　characteristic　in　the　time　domain,　so　the　prediction　of　the　residual　error　is　not　accurate　after　entering　the　video　encoder,　which　will　consume　unnecessary　coding　resources.　In　this　paper,　through　the　supplementary　analysis　of　the　work　in　the　literature　[10],　an　approximate　three-dimensional　volume　algorithm　is　introduced　to　generate　a　stable　reference　frame　for　the　Kinect　depth　map　sequence.　In　addition,　this　article　also　distinguishes　between　dynamic　macroblocks　and　static　macroblocks　in　the　depth　map　to　determine　whether　to　skip　the　search　steps　in　the　HEVC　(efficient　video　coding)　interframe　prediction;　finally,　compared　with　the　unprocessed　Kinect　depth　map　sequence　in　the　bit　mice.　The　performance　of　E,　run　time,　PSNR　(peak　signal-to-noise　ratio)　and　3D　reconstruction　are　better　than　those　of　the　latter.　©　2018　IEEE.