The　main　objective　of　this　paper　is　to　develop　3D　point　cloud　alignment　technique　by　using　an　improved　absolute　orientation　algorithm　based　on　unit　quaternion.　In　this　method,　Scale　Invariant　Features　Transform　(SIFT)　is　used　to　find　corresponding　feature　points,　and　Random　Sample　Consensus　(RANSAC)　is　used　as　robust　estimator　to　remove　false　matches　in　the　point　cloud　group.　The　unit　quaternion　solution　is　employed　for　initial　registration.　After　the　initial　registration　of　point　clouds,　this　cannot　meet　the　requirements　of　registration　accuracy.　Therefore,　we　need　to　achieve　accurate　registration　on　the　basis　of　initial　registration.　The　Iterative　Closest　Point　(ICP)　algorithm　is　one　of　the　widely-used　methods　to　cope　with　3D　registration.　However,　ICP　is　vulnerable　to　outliers　and　missing　data,　which　severely　compromises　its　performance.　In　this　paper,　Sparse　L-p-norm　based　ICP　algorithm　is　developed　to　achieve　precise　registration.　Experimental　results　verified　accuracy　of　the　presented　algorithm　in　point　cloud　registration.