Abstract:

Objective Anthropometric measurement is the primary step in garment production, where the lower body circumferences of the human body have high precision requirements and are difficult to measure. However, current prediction methods for lower body circumferences suffer from poor accuracy and limited applicability. Therefore, a highly accurate, convenient and rapid prediction method suitable for various circumferences required in trouser manufacturing is needed.

Method Five circumference measurements were considered for male youth aged 18 to 23, i.e. circumferences of waist, hip, thigh, knee and calf. By analyzing 176 manually collected samples in the training dataset, linear relationships among different parameters were identified. Prediction models were established based on these varying degrees of linear relationship different.

Results A data correlation analysis was conducted on various parts of the lower body required for trousers manufacturing, and the strength of the linear relationships between each part and the corresponding parameters was obtained. Based on the strong linear relationships between waist circumference, hip circumference, body weight, body mass index (BMI), width, and thickness, linear regression prediction models for waist and hip circumferences were established. Considering the strong linear relationships between thigh circumference, body weight, BMI, and the proportional coefficient K (height/weight ratio), a linear regression prediction model for thigh circumference was developed. In response to the insufficient fitting of the linear regression model for calf circumference and the weak linear relationships between knee circumference and various parameters, a method combining the random forest RF prediction model optimized by the whale optimization algorithm (WOA-RF) with the multiple linear regression prediction model was proposed. Hybrid prediction models for calf and knee circumferences were established respectively. Finally, the prediction models for each part and 44 samples were validated. A comparative discussion was carried out on the classified and unclassified body types for waist and hip circumferences, as well as the single prediction models and hybrid prediction models for calf and knee circumferences. The results showed that in the validation of waist and hip circumferences, the prediction accuracies of the unclassified method were 98.56% and 98.91%, respectively, higher than the results after classification by the width-to-thickness ratio. In the validation of thigh circumference, the prediction model achieved an accuracy of 97.80%. In the validation of calf and knee circumferences, the prediction accuracies of the hybrid model with the introduction of the proportional coefficient K were 97.89% and 97.80%, respectively, which were better than those of the single prediction model and the hybrid model without the introduction of the proportional coefficient K. The results meet the production requirements of garment enterprises.

Conclusion Taking the five circumference measurements required for customized male trousers in a garment enterprise as the main parameters, an adaptive circumference prediction method for each body part is proposed. This method can quickly predict the lower-body circumferences with a small amount of human body information. Compared with conventional methods, it has higher applicability and accuracy and can provide theoretical references for human body circumference prediction and two-dimensional non-contact anthropometric systems.

Key words: anthropometric measurement, circumference prediction, prediction model, non-contact body measurement, male trouser design