Research On Structure And Deformation Representation Of Fish Body

Controlling 3D fish model deformation by high-dimensional parameters can quickly build realistic 3D fish deformation pose and movement in computer,saving production cost,which is important for applications in virtual reality,game development,animation film and television.Due to the diversity of fish skeleton structure,it is difficult to design a universal fish skeleton to control the 3D fish model pose deformation.This paper focuses on the 3D fish skeleton model and fish body deformation method,through analyzing the skeleton characteristics and pose deformation law of real fish data,and using the iterative optimization learning method to determine a set of general fish skeleton model,based on which the fish pose is parametrically modeled,realistic fish pose deformation can be quickly generated by inputting high-level parameters,providing data for fish pose estimation and other research fields.It provides data support for research areas such as fish pose estimation.The research in this paper includes:(1)design and implementation of a fish pose data acquisition and processing scheme.The common spindle-shaped grass carp is selected as the research object,and a 3D fish body scanning platform is built.In order to ensure the good generalization of the method in terms of fish body shape and pose,this paper firstly constructs the fish body shape scanning dataset and fish pose scanning dataset respectively by using the above fish scanning platform.For the body type scan dataset,two common indicators for measuring fish body type,i.e.,body length and body width,are used to scan the zero-stance fish of different body types,and then a scan dataset containing various body types can be obtained;for the stance scan dataset,the common fish stance is classified into the following four categories by analyzing the distribution pattern of fish stance:"large angle left tail swing ","small angle left tail swing","small angle right tail swing","large angle right tail",and scan the above four types of fish stance data in different body sizes,and then obtain the data containing multiple stances.Then,a scan data set containing multiple postures can be obtained.Then,a 3D fish body template is constructed based on the above scanned data set,and the template is used for non-rigid alignment of the scanned data set.Finally,the triangle intersection detection and correction algorithm is used to deal with the triangle intersection problem in the non-rigid alignment process,and finally a topologically consistent fish body mesh dataset is obtained.(2)A fish skeleton model generation method based on deformed pose instance data is proposed.The skeleton is extracted and the fish skeleton model is defined from the biological characteristics of the real movement of the fish body;firstly,the optimal solution of the skeleton model parameters is approximately equivalent to the optimal solution of the fish torso deformation by joint segments,and the real deformation posture instance data is used as the constraint to make the rigid deformation posture of the fish body template close to the real fish body deformation posture within a certain error range while ensuring the number of fish body segments is small.With the increase of the number of joints and the increase of fish segments,the error between the rigid deformation attitude of the fish template and the real fish deformation attitude gradually becomes smaller,and the iterative optimization is stopped when the maximum error among all fish segments is less than a preset threshold,so as to determine the final fish skeleton parameters.Then the optimized fish skeleton joints are used as Ground Truth,and the fish skeleton joint features are extracted using deep neural network,so that it can automatically generate the fish skeleton based on the given 3D fish body mesh data.In addition,this paper realizes the fish body model deformation from two aspects:body shape and pose.For the body shape deformation,the principal component analysis is used to extract the body shape features and parametrically model the 3D fish model so as to realize the fish body shape deformation.For pose deformation,the pose parametric modeling method is used to drive the fish body template deformation by the skeleton parameters,which can get the fish body deformation under different poses.The experimental results show that the method in this paper can not only generate fish body data of different body sizes,but also generate fish body deformation of different postures according to the given high-dimensional parameters,and the deformation occurring in the fish body model conforms to the real deformation posture,which in turn verifies the rationality of the skeleton model generated by the iterative optimization method in this paper.(3)A prototype fish deformation system is built.Based on the fish skeleton model generation method and the body shape or posture deformation method,the functions of data processing,fish skeleton joint solution,posture deformation parameter learning and body shape deformation parameter learning are realized to meet the requirements of visualization and interactivity in the process of parametric fish modeling.