Communication between end to end is facilitated by the recent development in the area of wired and wireless networks. The end to end devices like mobiles, handheld and palm size PC's have limited capability for computation, display capacity and bandwidth. This augment, for ways for communicating large data files like video, in highly compressed and acceptable quality over the limited available bandwidth. We propose a reduced complexity quad tree based recursive motion estimation algorithm. It is best suited for video transmission over a very low bandwidth network. The scheme involves the preprocessed pruned quad tree decomposition of frame providing a better level of adaptation to scene content compared to fixed or variable block size approach in gray domain. Advance detection of all zero DCT (AZB) blocks in the ‘P’ frame is considered. The proposed technique utilizes the difference of the sum of absolute values in the block under consideration in the current frame and blocks in the reference frame for a priori determination of all zero blocks. Results demonstrate that the technique is able to outperform the method of the sum of absolute difference of the motion compensated blocks. An early detection of all zero blocks (AZB) in the P frame shall eschew the need of DCT significantly reducing the processing requirements and also allow effective bit control. The partition criterion is based on segmenting the blocks recursively into active (non AZB) or passive regions (AZB). Only the active regions (object) blocks, are partined to lower dimension (8*8 & /4*4). The active region is very small region compared to total frame size thus we are able to reduce the number of blocks for motion estimation and encoding. Our result shows that it reduces the prediction error and computational complexity