Characters Animation Editing Technology Research Based On Story Text Semantics

With the rapid development of computers,animation editing technology has attracted more and more attention.In traditional animation editing,a professional animation producer creates character and scene models through 3D animation software,then artificially act on the K-frame of the character and gradually place the scene model.This not only imposes high demands on the producer's technology,but also makes it more difficult for general users to participate in animation editing.In order to make the editing of the animation more convenient and quick,this thesis chooses the character animation editing technology based on the story text semantics.The main research work of this thesis is as follows:First of all,this thesis proposes a combination of natural language processing and nonverbal behavior generation,and appropriately modifies and expands the rules of nonverbal behavior generation,so as to extract character animation semantics and generate behavior description language;Then analyze the spatial relationship of the scene objects in the text according to the dependency relationship and store them in triples.Extract the default scene in the story text,match the spatial relationship diagram template in the template library,and expand the default spatial relationship diagram template according to the triplet.This thesis proposes a method of scene placement based on spatial relation graph templates,transforms the spatial relationship of scene objects into concrete coordinates,and generates scene description language.Combine the two languages to get a behavior-scene description language and parse.Secondly,for the joint mapping in action retargeting,this thesis proposes an automatic mapping method based on the feature of joint tree hierarchy.This method can automatically map joints between the captured data skeleton and the target skeleton,and obtain better joint mapping results for most human-like structural skeletons.The constraint adjustment of motion data uses a combination of geometric scaling and Jacobi pseudo inverse matrix,which effectively avoids the distortion that occurs when the existing motion capture data is redirected to the target skeleton,ensuring the consistency of the target character and the original action.For the realization of the character movement and the arm movement behavior,an action mixing method based on multi-segment interpolation fusion is proposed.This method constructs different parameter spaces for the two types of actions and can generate different action effects in real-time according to the input parameters.Finally,based on the above methods and techniques,this thesis designs a semantic-based editing system for animated characters and scenes.The system takes animated story text as input,extracts text semantics to generate behavior-scene description language,and then parses the labels and attributes of this language to drive the automatic placement of the scene and the execution of the character animation.Based on this,combined with the simple manual adjustment function of common game engines,a real and convenient animation editing system is realized.