Gender And Age Group Recognition Of 3D Human Shape Via Deep Learning

Gender and age recognition are one of the hotspots in the fields of computer vision and pattern recognition,and have important applications in actual scenarios such as identity authentication and security monitoring.Previous work mainly focus on 2D images and videos representation of human bodies.In recent years,with the advancement of 3D scanning technology and 3D parametric human shape reconstruction methods,the number of 3D models has gradually increased.However,the research on gender and age recognition of 3D human shapes has not resembled as wide as that images.At the same time,the gender and age recognition of 3D human shapes have the following problems: the representation is complex,and the topological is inconsistencies.To solve the above problems,this thesis constructs a database with gender and age group labels respectively,and uses deep learning methods to conduct gender and age group recognition research on the 3D human shape.The main research work is as follows:(1)Propose a 3D human gender recognition method based on bi-harmonic distance distribution and deep learning methods.Firstly,for each 3D human shape,find the bi-harmonic distance from each point to all points,and compute the histogram distribution of the distance,the gender feature descriptor.Then,the gender descriptor is input to the fully connected network to perform the 3D human shape gender recognition.The experimental results show that the 3D human gender recognition method in this paper is better than previous work,and has noises robust.(2)Propose a 3D human age recognition method based on point cloud attention network.Firstly,the point cloud convolution network is used to extract intermediate features from the 3D human body point cloud.Then,the age features are obtained through the point cloud attention network.Finally,the multi-layer perceptron network is used to identify the age of the 3D human shape.Experiments demonstrate that the point cloud neural network integrated into the attention mechanism has better performance in the recognition of 3D human age.(3)Construct 3D human gender and age datasets with labels respectively.They are applied to recognize gender and age groups of 3D human shapes.Firstly,download the existing model from the Internet,and manually label the unlabeled ones.Secondly,the models with labels are generated through the 3D model generation software.