Research And Implementation Of Body Shape Feature Extraction Based On Deep Learning

With the rapid development of technologies such as virtual reality and augmented reality,three-dimensional models have been applied to all aspects of life.The fitting degree of three-dimensional models to the human body to a certain extent depends on the human body shape feature extraction technology,which is convenient and fast.It can well assist relevant personnel in 3D human body modeling.In current life,smart devices are becoming more and more popular,and image data is no longer difficult to obtain.Therefore,the method of extracting human body features based on image data has also become a research hotspot.At the same time,in recent years,deep learning technology has made great achievements in the field of computer vision,and has also pointed out new research directions for the extraction of human body shape features.The research of the thesis introduced a research design scheme based on convolution neural network of the shape feature extraction algorithm,called Multi-Human-Shape CNN.The algorithm mainly solves how to extract body shape features from static pictures to obtain the body shape feature description of the human body in the picture.Compared with simply extracting the body shape feature description from the picture,the algorithm also associates the body shape feature description with SMPL,and designs a The three-dimensional model generation algorithm is used to derive the three-dimensional human body model corresponding to the human body features.The thesis elaborated on a human body feature extraction system designed and implemented based on the above algorithm.The system consists of a picture preprocessing module,a human body shape feature extraction module and a three-dimensional model generation module.The overall system takes pictures as input and various forms of human body features,three-dimensional human body models as output.The thesis first introduces the research significance and related background of the thesis topic,and gives a brief overview of the research content,and then introduces the main fields and related technologies involved;and then introduces Multi-Human-Shape CNN,.And its related 3D model generation algorithm.Then based on the algorithm,the related work of the human body feature extraction system is introduced,including demand analysis,outline design,detailed design and related testing;finally,the overall work is summarized and prospected.