Abstract: Accurate, real-time acquisition of 3D head coordinates is important for immersion and user experience in standing virtual reality (VR) interactions. Current solutions for head localization are limited by cumbersome spatial anchor setup, high hardware costs and underutilization of human body geometric priors. To address these issues, a low-cost method for acquiring 3D head coordinates in indoor VR scenarios is proposed. First, target detection and character recognition are adopted to obtain the coordinates of human foot anchors in order to establish the reference coordinates of the human body in the indoor environment. Second, measure- ments from inertial measurement units (IMUs) are combined with body geometric parameters to construct a simplified trunk-head kinematic prior constrained by foot anchors. Third, an adaptive regression model is built based on a dual-branch residual structure and physical consistency constraints. The 3D head coordinates are then inferred by the adaptive regression model. Experimental results show that the average acquisition cycle of foot anchor coordinates ranges from 68 ms to 87 ms with an average error below 1.5 cm. The average output interval for head coordinates is 29.4 ms (34 Hz) , and the Mean Euclidean Distance error (Mean_ED) is less than 5 cm. Personalized calibration experiments show that a small amount of target-subject data can improve model adaptation; IMU mounting error and visual anchor offset experiments show that the proposed method still maintains certain stability under typical deployment disturbances. The results validate that the proposed method is capable of centimeter-level 3D head localization using low-cost hardware. By striking a favorable trade-off among localization accuracy, deployment cost and real-time performance, this work offers a feasible technical route for lightweight 3D head localization in indoor VR systems.