Three Dimensional Body Measurement Based On CNNs And Body Silhouette

With the rapid development of Internet technology and E-commerce platforms,consumers prefer buy their own clothes through Internet.Moreover,people seek more suitable clothes for them as the personalization culture develops,which results in high demand for human body precise measurement.To recommend clothes and realize three-dimensional virtual fitting,E-commerce platforms need to acquire consumers's human body dimensions.In order to meet the aforementioned requirements,human body dimensions should also be acquired conveniently and quickly.However,current the three-dimensional human body measuring instruments are expensive,which prevents their uses for ordinary people.To this end,this thesis investigates a convenient and cheap human measurement method for ordinary customers.We formulate reconstructing three-dimensional models from orthogonal human body silhouette images as a regression problem and innovatively apply the GoogLeNet architecture to human reconstruction and use the human skeleton for anthropometric measurements.Using a deep learning method to train a network,the contour images are transformed into its corresponding shape parameters,and then the human model is recovered.After that,we extract the skeleton of the human body model in order to obtain the preprocessing feature points of the human body surface.Based on these coarse feature points and the morphological principle of human body,we can accurately locate the surface characteristic points of human body,and finally calculate the necessary human body dimensions for human body measurement.