Research On Classification Algorithm Of 3D Non-rigid Model

3D non-rigid model classification is an important subject in the study of 3D model classification,The research of this topic is based on the National Natural Science Foundation of China" Measure preserving transformation for 3D deformable objects and recogniton of them by sparse manifold embedding methods".In the thesis,we study the classification algorithms of 3D non-rigid models which have undergoes both posture and shape changes.Our work is summarized as follows:·3D non-rigid model expansion and posture normalization.We collected and consolidated 3D non-rigid models from SHREC'15 and SHREC'16 official databases.Then,we change the posture and shape of the model through Blender software.Finally,we perform pose normalization preprocessing on the data.This experimental study uses this database.·Feature extraction and classification of 3D non-rigid models.We respectively studied the feature extraction and classification of 3D non-rigid Model under both traditional and deep learning methods.Traditional methods:We extract the traditional features of the 3D non-rigid model,and then used multi-class SVM method for classification.we used two main traditional features:Gaussian curvature features and RoPs features.We have improved the detection of keypoints and combined Harris 3D keypoint detection with Gaussian curvature keypoint detection.Finally,a multi-class SVM algorithm is used for classification.Deep learning methods:We use MVCNN to extract features and classify different 3D non-rigid models.·3D non-rigid model classification and decision fusion based on Deep Forest method.Through this new attempt,the classification accuracy of the 3D non-rigid model has been further improved.In the thesis,we use C++,Matlab,PCL,TensorFlow framework,and Deep Forest framework to implement the algorithm in the above study.The total 3D non-rigid models are 6240 and are subdivided into 48 categories.The decision fusion accuracy is about 99.74%.We has achieved a good classification effect.