This paper proposes a new fast multi-reference picture motion estimation algorithm which uses the relation of motion vector with distance of each frame. The proposed algorithm reduces search areas of motion estimation using linear feature of object's motion for fast motion estimation. Search areas in reference picture 2, 3 and 4 are minimized using a distance of each frame and motion vectors of reference picture 0 and 1. Additionally, the proposed algorithm does not require a threshold value which makes processing time irregularity Therefore, the proposed algorithm can be efficient hardware implementation. Experimental results show that the number of computation is reduced 52.5% and PSNR drop of encoded sequence is negligible when compared with standard which uses 5 reference pictures.