基于服裝點(diǎn)云的高精度袖窿分割線自動(dòng)識別方法

中圖分類號：TS941.26 文獻(xiàn)標(biāo)志碼：A 文章編號：1673-3851（2025）07-0488-10

引用格式：，等。基于服裝點(diǎn)云的高精度袖窿分割線自動(dòng)識別方法[J].學(xué)報(bào)（自然科學(xué)），2025，53（4）：488-497.

Abstract：To solve the problems that it is dificult to identify the armhole segmentation lines and that the sleeve template does not meet the requirements of plate making，an automatic high-precision recognition method for armhole segmentation lines based on garment point clouds was proposed by taking tops with the shoulder line at the shoulder position as an example. First，a dynamic graph convolutional neural network （DGCNN） was used to coarsely segment the garment point cloud to obtain the armhole point cloud area，and the armhole key point cloud was obtained by denoising. Then，the weighted least squares method was used to fit the key point cloud of the armhole，and the front and rear armhole dividing lines were obtained，and the split lines were optimized by combining the arc interpolation method and cubic spline interpolation to solve the problems of missing and dislocated connections. Finally，the surface deployment algorithm based on angle protection generated the sleeve template，and the sleeve cap arc correction scheme was added to compare the generated template with the real one and that generated by the NeuralTailor model to verify the accuracy of the templates. The experimental results show that the armhole segmentation line generated by this method is highly similar to the real armhole segmentation line，and the RMSE and MAE are both less than 0.200cm . The sleeve template generated by surface deployment algorithm based on angle protection has high precision，with the average deviation of the key dimensions of the sleeve being 0.238cm ，and the average absolute error value of sleeve template being 0.296cm ，which further verifies the accuracy of the armhole dividing line. This provides an effective technological path for the automated generation of garment sleeve templates，and can enhance the efficiency of clothing design and development.

Key words： garment point clouds； armhole segmentation line recognition； dynamic graphconvolutional network;curve fitting; surface deployment； sleeve body template generation

0引言

隨著服裝產(chǎn)業(yè)數(shù)字化和智能化的快速發(fā)展，服裝行業(yè)的生產(chǎn)與銷售模式正在經(jīng)歷深刻的變革。消費(fèi)者對服裝個(gè)性化定制的需求日益增長，傳統(tǒng)的成衣生產(chǎn)模式已越來越難以滿足市場的多樣化要求[1-3]。當(dāng)前服裝制版主要依賴人工操作，不僅效率低，而且難以應(yīng)對快速迭代和個(gè)性化定制的需求，導(dǎo)致生產(chǎn)周期延長和成本增加。同時(shí)，電子商務(wù)的崛起縮短了產(chǎn)品迭代周期，傳統(tǒng)制版模式面臨前所未有的挑戰(zhàn)，對制版效率提出了更高的要求。為此，三維服裝模型向二維樣板轉(zhuǎn)換的技術(shù)[4-6逐漸成為研究熱點(diǎn)，該技術(shù)不僅能顯著提高服裝制版效率，更為個(gè)性化樣板的生成提供了切實(shí)可行的路徑。然而，三維到二維轉(zhuǎn)換過程中的分割線識別仍面臨挑戰(zhàn)，尤其是復(fù)雜的袖窿分割線，它直接影響成衣的合身性和舒適度。因此，精確識別和定位服裝分割線，特別是袖窿分割線，成為自動(dòng)化樣板生成中的關(guān)鍵問題。

針對三維服裝模型向二維樣板轉(zhuǎn)換技術(shù)，國內(nèi)外學(xué)者開展了大量研究，研究方向主要可以分為兩類：一是基于深度學(xué)習(xí)算法直接生成二維樣板；二是基于三維模型進(jìn)行曲面展開來獲得二維樣板。Korosteleva等提出了一種從三維服裝點(diǎn)云中恢復(fù)服裝樣板結(jié)構(gòu)的深度學(xué)習(xí)網(wǎng)絡(luò)—NeuralTailor，用于決策拓?fù)浜皖A(yù)測樣板細(xì)節(jié)，該網(wǎng)絡(luò)結(jié)合了三維點(diǎn)級注意力機(jī)制與循環(huán)神經(jīng)網(wǎng)格（Recurrentneuralnetwork，RNN），顯著提升了制版速度。……