关键词: 机器视觉, 曲面枕, 个性化定制, 神经网络, 轮廓提取

Abstract:

Objective Curved surface pillows, as innovative bedding products designed to provide comfortable support and promote quality sleep, are characterized by their customization of pillow curvature based on individual features and needs. This customization aims to accommodate the shape of the head and neck, improve sleeping postures, and alleviate pressure from the neck and shoulders. With growing consumer concerns on health, the demands towards personalized customization of curved surface pillows continue to rise. To enhance the adaptability of curved surface pillows and meet the demands for large-scale personalized customization, a method for personalized customization of curved surface pillows based on machine vision is proposed.

Method In the experiment, photographs of head, neck, and shoulder areas of participants were captured using a mobile phone. MINet saliency object detection was employed to segment human subjects and extract their contours. 4 key reference points and 32 feature points within the human subjects were extracted. Based on this information, contour curves were fitted and dimensions were measured, resulting in the construction of a sample dataset. These data were then imported into the SPSS system for data analysis. The classification and summary function to select variables were used to calculate the mean and variance and to find the relationship between variables for classification.

Results In the experiment, 65 valid sample datasets were collected to verify the method's discriminative capability. These datasets included measurements of participants' head and neck areas and sleeping posture information. Measurements revealed irregular curves and unevenness in the back of the neck for each individual. The degree of curvature and unevenness varied significantly among participants. However, if a one-person-one-version customization method was to be adopted, there would be problems such as long cycle time and high cost. Therefore, the collected head and neck samples were classified into 12 categories based on the different degrees of curvature of the posterior neck, which was primarily a classification and summary of the measured sample's (shoulder width - width between ears)/2 data and peak and valley values. Through calculation and data association, it was found that the lateral lying height of the pillow was divided into 12 categories with an interval length of 0.5 cm, the peak curve was divided into an interval length of 0.35 cm, and the valley curve was divided into an interval length of 0.3 cm. Such classifications allowed for the selection of the appropriate category based on each participant's head shape and size, enabling the customization of a personalized curved surface pillow tailored to their individual needs. Consequently, each individual would have a pillow customized to their unique head features, ultimately enhancing sleep comfort and quality. By extracting and classifying sample data, models of curved surface pillows that fit individual data were created. Pillows customized based on individual measurements were found to better accommodate the shape of the individual's head. On the one hand, further expanding the shape adaptability of curved pillows would effectively reduce the occurrence of neck pain and discomfort, and improve sleep quality, and on the other hand, it would also provide a classification method reference for the customized market of curved pillows.

Conclusion Associating shape data of personalized pillows with user information, the complete data for personalized pillows is formed, and the system uses three-dimensional modeling software to create models of curved surface pillows based on individual head shapes and sleeping postures. This approach takes into account the neck morphology and curves, providing better neck support, making individuals more comfortable during sleep, and enhancing sleep quality. At the same time, it also provides a new design method for pillow customization, improving customization efficiency, and saving production costs. However, this experiment lacks analysis of actual neck pressure. Due to long-term use and compression, the height and shape of the pillow may change. In future research, the introduction of pressure sensing technology to monitor actual neck pressure in real time could optimize pillow design parameters.

Key words: machine vision, curved surface pillow, personalized customization, neural network, contour extraction