In　order　to　complete　the　task　of　vehicle＇s　autonomous　driving　in　structural　road,　it＇s　necessary　to　detect　obstacles.　This　paper　proposed　a　method　of　obstacle　detection,　taking　use　of　the　four-line　laser　radar.　This　algorithm　combined　the　improved　DBSCAN　with　K-Means　and　overcame　DBSCAN＇s　defect　that　it　couldn＇t　divide　obstacles　with　similar　density.　At　the　same　time,　this　method　can　eliminate　noise　points　effectively.　Based　on　clustering,　obstacle＇s　information　can　be　acquired,　such　as　angle,　distance　and　size,　to　achieve　the　task　of　obstacle　detection.　The　algorithm　is　applied　to　obstacle　detection　in　intelligent　vehicle.　The　test　proved　that　it　is　consistent　and　reliable,　which　complies　with　the　requirement　of　intelligent　vehicle＇s　autonomous　driving.