A　hybrid　vision　system　is　constructed　by　a　monocular　camera　and　an　omnidirectional　camera.　The　epipolar　geometry　of　this　system　is　analyzed　and　a　method　to　determine　the　common　view　filed　of　this　system　is　proposed.　Scale　invariant　feature　transform　(SIFT)　features　are　extracted　in　the　common　view　field.　During　the　feature　points＇　initial　matching　process,　the　angle　between　the　SIFT　descriptor　vectors　is　considered　as　the　constraint　condition.　In　order　to　eliminate　the　error　matching　in　initial　results,　the　epipolar　constraint　is　used　to　obtain　the　correct　feature　matching.　Finally,　3D　reconstruction　of　the　local　scene　is　realized.　Experimental　results　show　that　the　reconstruction　method　is　feasible　and　practical.