Research On The Method Of Predicting Height And Weight Based On Plantar Pressure Images

Due to the differences in the development of innate bones and the influence of labor conditions in the acquired life,everyone's footprint characteristics have different reflections.The barefoot footprint contains a wealth of personal information,such as the habits of individuals who usually walk and the structural characteristics of congenital feet.These are concealed and easy to collect.There have been many domestic and foreign researches on footprints,especially It is of great help to the criminal investigation work.Based on the application of footprints at home and abroad,this thesis studies the relationship between footprints and height and weight,and considers a variety of methods to predict height and weight based on plantar pressure images.The specific research contents are as follows:(1)According to two research methods with different characteristics,the pressure images and characteristics of the barefoot are analyzed.For the multiple regression analysis method and the BP neural network prediction method,18 kinds of footprint characteristics including various length information of the footprint,foot sole area,each pressure surface area and their distribution information were extracted;for the convolutional neural network prediction method,Two kinds of feature images are extracted: HOG feature images and Contour feature images,so as to better train the network and improve prediction accuracy.(2)For the multiple regression model,BP neural network model and a combination model of the two,this chapter considers the relationship between 18 kinds of footprint feature information and height and weight.In the analysis of multiple regression models,by analyzing the correlation between variables and the significance of the regression coefficients of the model,the fitted model is obtained and the prediction experiment is carried out.In the BP neural network,based on the above characteristic parameters,models for height and weight are established respectively.By selecting the best model,the prediction model is finally tested.Considering that the two methods have their own advantages and are not related to each other,the two methods are combined in a weighted manner,and the resulting combined model is tested to obtain the best experimental results in this chapter.(3)For the convolutional neural network,this thesis selects three network models CNN,VGG16 and ResNet34 with classic structure,based on the plantar pressure image and its characteristic image to predict height and weight.Try to improve the VGG16 model,analyze and compare the adjusted model with the prediction results of the other three network models.The three network models CNN,VGG16 and ResNet34 are currently widely used models with different network structures and characteristics.The three image data sets were trained and predicted separately.The adjusted VGG16 model obtained the best experimental results in this chapter.