Title : Object based motion estimation A cost - effective implementation

Advances in multi-media have led to large range of video formats and, more specifically, to an ever increasing number of picture refresh rate frequencies. Video Format Conversion (VFC) therefor becomes a necessity. The more sophisticated VFC methods use motion estimation and motion compensation to prevent motion judder and motion blur. In this report the Object-based Motion Estimation (OME) algorithm is considered for a cost-effective VFC solution. An analysis of the computational requirements and memory bandwidth requirements is made. Multi-level caching architectures are proposed to keep memory bandwidth and power requirements low. Further, algorithmic changes to OME, which are necessary for OME to be mapped on an Application Specific Instruction Processor (ASIP) based on a Very Long Instruction Word (VLIW) template, are discussed. The quality of the new solution is verified through a comparison of the vector fields of OME and 3DRS. An initial implementation of the OME ASIP has been made, and it is compared with other motion estimation solutions. A reduction of 102 times in power requirement and a reduction of 8 times in area is achieved for the ASIP compared to a General Purpose Media Processor (Trimedia). Conclusions: Overall the conclusion can be made that OME delivers comparable motion vector fields to 3DRS, but requires more resources (a 10 % higher computational requirement and 35% more memory bandwidth requirement) to achieve them. OME, as presented in this report, is therefor not suited as a replacement for 3DRS in Video Format Conversion applications. It does however provide a segmentation of the image into objects at relative low cost, which could make it useful for emerging applications. Further study should reveal its usefulness for these applications. c ©Philips Electronics Nederland NV 2003 iii 2003/00573 Unclassified Report iv c ©Philips Electronics Nederland NV 2003