Research And Implementation Of Character Motion Blending Based On Data-driven

Character animation has become more and more important in the modernentertainment industry. As the basis of character animation, how to achieve abundantanimation resources is an important demand for character animator. Motion capture is asuccessful technique for character motion generation because of its high authenticity ofsynthesizing data, but motion capture technology is primarily limited by itself andrequires expensive hardware equipment. Motion editing technologies emerge to meetthe requirement of reusing captured motion data. Motion blending is an importantmotion editing technology, which produces new motion and creates seamless transitionby combining two or more motion clips according to blending weights, and alsoenriches animation library and generates natural character behaviors.A further research about motion blending technology is developed, and obtains anabundant research achievement, as shown in the following aspects:1) The abstraction of motion data. Generally we represent motion data usingtranslation values and a rotation value of character’s all joints on each moment which isbased on joint levels. We represent motion data based on bone levels instead of jointlevels. Because most motion is closely associated with four limbs, so we focus on eightbones of four limbs and the center bone. The motion data of each moment is abstractedby eight angles between eight bones of four limbs and the center bone, and the changeof angles shows character’s motion characteristics. The abstract motion data structure isconvenient and simple without more pre-processing, which has been applied widely inmulti-target sample motion blending and motion transition.2) About multi-target sample motion blending algorithm, based on motion datastructure of bone levels, we apply curve simplified algorithm to time synchronization ofmotion blending, then extracted time synchronization points are used for timesynchronization for multi-target sample motion to find those similar key-frames. We usequaternion algebra instead of Euler angles to interpolate similar key-frame valuesinteractively. Our method can not only overcome the complexity of parameter zingmovement data, but also get good blending effect in real time without much preprocessing. Besides the position and orientation of root joint are recreated, virtualcharacter will travel along random path specified by users. This is flexible and practical.3) About motion transition algorithm, we generate transition frame based onmotion transition graph, and we create lookup table of similar postures for extractedtime synchronization points, then apply posture alignment algorithm and createtransition phase. In the end blending interpolation algorithm is used for motiontransition. We get postures in transition period through four steps linear interpolation online. It is simple and efficient with few calculated amount. At the same time it producesa true and natural effect for motion transition.4) We use classes to encapsulate blending algorithms and data structure, anddivide modules and design the details of the modules. We develop a mature motionsynthesis system in the form of plugins based on Anima Motion Interaction System ofdigital media technology lab. Our system improved the utilization of resources andenrich animation express effects, which is high-efficiency and convenient. We will domore researches based on our platform in the future.