The Research On Human Motion Alignment Using Motion Segment Based Method

Motion capture technology has been widely used in many fields such as digital entertainment,motion simulation and so on.Compared to the traditional animation production technology,motion capture technology has significant advantage in the fidelity of character and convenience of production.But at the same time,the cost is relatively high,both in the motion capture equipment and every single time's capture.Despite the huge human motion capture database has been established,how to effectively use the existing data to generate more abundant sports samples has become a hot research topic in the field of computer animation.Motion alignment aims to find the corresponding relationship between different motion clips,the different motion clips can have a high degree of logical similarity after alignment,therefore it become a necessary step of much subsequent work.As a classic method of human motion alignment,original Dynamic Time Warping algorithm can cause poor effects while applying in the field of human motion synthesis,such as pause and distortion.To address these problems,we propose a new human motion alignment method which is based on human motion segmentation.We first make use of Isomap which is a classic algorithm in manifold learning to reduce dimensionality of all the motion frames in the original motion segment.Then we can obtain the cycle of different motion segments by analysing the corresponding component of low-dimension coordinates.Finally we apply Dynamic Time Warping method in the corresponding cycles of different motion segments respectively to achieve refined alignment results.We conduct contrast experiment between our algorithm and traditional method on different types of human motion,the result shows that our method can eliminate pause and distortion phenomenon caused by traditional method,and make significant improvement in the application of human motion synthesis.