To reduce the huge computation cost of motion estimation is indispensable in practical encoding system. Utilizing the motion activity of video sequence to adapt the search window is proposed in this thesis. The proposed scheme provides a precise estimation of two motion activities in horizontal and vertical directions to refine the search window in a shape of rectangular, instead of a square search window. Hence, we can find a simpler yet more appropriate search window. Furthermore, an efficient pipeline architecture designed for PLS algorithm, which is hybridized with our proposed algorithm, is discussed in this thesis and is implemented in our experiments. Experimental results show that 40% to 50% performance improvement in speed is gained while yielding nearly the same quality measured in PSNR compared to the PLS.