Mocap data compression: Algorithms and performance evaluation .
详细信息
文摘
The mocap data has been widely used in many motion synthesis applications for education, medical diagnosis, entertainment, etc. In the entertainment business, the synthesized motion can be easily ported to different models to animate virtual creatures. The richness of a mocap database is essential to motion synthesis applications. In general, the richer the collection, the higher quality the synthesized motion. Since there exists limitation on network bandwidth or storage capacity, there are constraints on the size of the mocap collection to be used. It is desirable to develop an effective compression scheme to accommodate a larger mocap data collection for higher quality motion synthesis. In order to synthesize natural and realistic motion from existing motion capture database particularly in the context of video game applications, the compression procedure enables efficient management. In this research, we explore the characteristics of the mocap data and propose two real-time compression schemes that allow a flexible rate-distortion trade-off. These two compression schemes are designed to optimize different objective functions for different application purposes. Unlike previous work, the proposed schemes do not demand any prior knowledge of the motion type. The first one aims at preserving nearly lossless content. It encodes prediction residuals, which allows more flexibility in bit allocation. We study the relationship between the mocap data, coding parameters, and coding performance. To be more specific, we use temporal and spatial features to characterize a clip of mocap data, and propose a rate control algorithm to adjust coding parameters adaptively to deliver a result that matches the target bit rate or the error bound. With proper temporal and spatial information handling, this scheme can achieve a coding gain of about 45:1 with low coding complexity and good visual quality, which is 2.5 times better than the state-of-the-art techniques. The second one is perception based compression, which aims at further compression with visually pleasant quality. In the second scheme, we partition the motion into segments with positive and negative velocities and then compress each segment separately. This scheme can achieve a coding gain of at least 100:1, and can be used to provide a quick preview of the content of the database.