Dimension Measurement For Dressed Human Bodies Via Orthogonal Image

In the context of the "Internet +" era,body size measurement technology plays a key role in various fields such as virtual fitting,remote non-contact body size measurement technology and the establishment of a human body database.Among them,with the development of clothing information technology,in order to meet the needs of people for remote clothing customization,a non-contact body size measurement technology has been proposed.However,the traditional non-contact body size measurement technology cannot completely get rid of large-scale 3D scanners.Equipment,therefore,there are unavoidable disadvantages such as high cost and poor measurement flexibility.Therefore,this paper proposes a human body size measurement technology based on orthogonal images.The human body size information is obtained through two images of the human body in a specific posture and a series of image processing operations.It is verified through experiments that this technology is accurate,fast,and accurate compared to traditional methods.Advantages such as simplicity and low cost.First,A two-dimensional human body size measurement system based on orthogonal image is proposed in this paper,which realizes the user image acquisition,storage,human contour extraction and two-dimensional human body size measurement.Before collecting the image,the distortion of the mobile phone camera is eliminated according to "Zhang Zhengyou image calibration method",then the collected image is transformed into HSV color space,and the s channel is extracted.Based on the improved multi-directional Sobel operator edge detection,the complete and clear human contour is obtained.Next,an algorithm based multi-features points extraction and dimension measurement of dressed human bodies(Human pesm-abss)via adaptive body structure segmentation(ABSS)is proposed in this paper.According to the differences between the heterogeneity and body shape of Eastern and Western,the critical parts of human body structure are segmented by ABSS.To address the problem of poor adaptability and low robustness of classical algorithm to extract human neck and shoulder feature points,the maximum distance method and local maximum curvature method are proposed.The experimental data of 210 samples with large standard deviation are compared with the real size information.Our empirical study demonstrates that the average error of Human pesm-abss algorithm reduced 2.2cm and 0.26 cm respectively compared with Unclosed Snake model and Simple-FCN-ASM model.And the time consumption of algorithms reduced 1.098 s and 3.552 s respectively.Then,In the aspect of extracting three-dimensional dimensions of human body,Due to lack of depth information in 2D pictures,it is difficult to obtain the three-dimensional size information of the human body.The size information fitted via classical linear regression method is the mean value of the threshold interval to which the human body belongs.Because the heterogeneity of the human body is ignored,the size error of the fitting is very large.Manikin reconstruction can improve the accuracy of size acquisition,due to the large scale of computations and parameters,it is difficult to deploy in mobile devices.Therefore,a model of predicting 3D human dimension via Upgrade GA-BP-MC(UGA-BP-MC)neural network is proposed,This model is used to optimize the BP network structure,weights and thresholds by upgrading the adaptive crossover and mutation probability of genetic algorithm.In addition,the Markov residual network is used to improve the prediction accuracy of the UGA-BP model.The experimental data of 210 samples are compared with the real size information.Our empirical demonstrates that the average prediction error of UGA-BP-MC is reduced2.8cm,1.62 cm and 0.94 cm respectively compared with Hyperelliptic curve method,multivariate function and GA-BP model.Finally,based on the improved WOA-ENN,an improved human body girth prediction method is proposed.The better optimization breadth and accuracy are obtained by optimizing the whale algorithm with the optimized convergence factor,the joint optimal inertia weight and the Archimedes spiral updating strategy.Then the optimal output of improved WOA(IWOA)is used as the initial weights and thresholds of Elman neural network(ENN).It avoids the classical model falling into local optimum in the late iteration and reduces the convergence time of network training.In addition,the Markov residual model is used to optimize the predicted values of IWOA-ENN model,which further improved the prediction accuracy of the model.The measured values of 240 samples from the CAESAR(Civilian American and European Surface Anthropometry Resource)database are compared with the predicted values.Our result prove that IWOA-ENN-MC model has more obvious advantages in optimization capability and accuracy of fitting and prediction than GA-ENN,PSO-ENN and WOA-ENN.In Summary,The overall research route of this paper follows the human body's two-dimensional size measurement and then three-dimensional size prediction.Experiments have verified the superiority of the Human pesm-abss algorithm,UGA-BP-MC and IWOA-BP-MC.