Pillows have a significant impact on sleep quality and cervical spine health, and neck socket depth and shoulder width are the most critical parameters in their design and selection. Traditional measurement methods require subjects to maintain a standard posture for a long time and are done manually by the surveyor, which is time-consuming and prone to errors due to subjective factors. In this paper, we first propose an automated non-contact measurement pipeline based on the combination of deep learning and region of interest (ROI) geometric analysis, which can calculate the neck socket depth and shoulder width directly from a partially backside body point cloud wearing clothing. The core of the method is the proposed Internal Body ROI Recognition Network (IBROI-RNet), which realizes ROI label prediction and backside internal point cloud reconstruction through a two-branch cascade architecture. On this basis, the accurate computation of target dimensions is accomplished based on geometric analysis. Experimental results show that the method demonstrates excellent measurement accuracy and practicality while protecting user privacy, providing a feasible solution for the design and selection of personalized pillows. The code and model are available for research purposes at https://github.com/kumori97/neck-and-shoulder-measurement