Depth　information　acquisition　is　the　key　foundation　of　panoramic　longitudinal　roaming.　In　order　to　improve　the　accuracy　of　before　and　after　scene　images　matching,　the　belief　propagation　algorithm　for　the　epipolar　matching　is　proposed.　According　to　the　principle　of　epipolar　geometry,　the　epipolar　line　of　scene　images　is　constructed　by　taking　the　image　center　as　the　benchmark　and　diverging　in　the　whole　direction.　The　vertical　matching　cost　component　is　added　to　the　matching　cost　function,　and　the　disparity　map　is　generated　by　the　belief　propagation　algorithm.　Through　the　geometric　relationship　between　the　scene　images,　the　model　of　disparity　and　depth　is　constructed　to　obtain　the　depth　information.　Experimental　results　show　that,　compared　with　the　local　optimization　algorithms,　this　algorithm　is　better　for　obtaining　the　depth　map　of　scene　images.　SSIM　and　PSNR　are　improved　by　0.22　and　24%　respectively.　©　2018,　Beijing　China　Science　Journal　Publishing　Co.　Ltd.　All　right　reserved.