Research On 3D Human Body Measurement System Based On Deep Learning

Along with the improvement of material living standards,people’s demand for clothing no longer stays in keeping warm from the cold.Clothing has gradually become a window for modern people to reflect the trend.Personalized clothing customization came into being,but the shortage of professional tailors hindered the development of this industry.In recent years,artificial intelligence has made breakthrough progress and has begun to be widely used in people’s daily life.The introduction of artificial intelligence into the measurement industry is bound to bring new vitality to the clothing customization industry.In response to the above problems,this paper designs and implements a three-dimensional human body measurement system based on deep learning,which has the advantages of low cost,high speed,high precision,and small size.And research on the following points:(1)Aiming at the problem of multi-view 3D point cloud registration,this paper proposes a multi-view point cloud registration algorithm based on the matching score.This method uses deep learning-based features for point cloud registration,and uses an iterative closest point algorithm to optimize the registration effect,then introduces the matching score to monitor the effect of point cloud registration.(2)Aiming at the problem of human body key points detection,this paper applies the 3D human key points detection algorithm based on a deep neural network to detect the spatial coordinates of multiple key points.Compared with traditional geometric methods and 2D methods,the coordinate accuracy is higher and the spatial position information can be directly applied to model segmentation and body measurement.(3)Aiming at the problem of the girth calculation method,this paper proposes an improved convex hull algorithm for the P-spline curve.The P-spline curve is used to fit the discrete points of the girth to reduce the girth measurement error caused by outliers.Finally,this article tests the measurement error and running speed to verify the accuracy and practicability of the system.The experimental results show that the system designed in this paper has high accuracy in most measurement items,and the running speed is ahead of similar products on the market.Simultaneously,the production cost is much lower than the selling price of similar products.Therefore,the system designed in this paper will have broad application prospects in practice.