Multimodal Classification of Sense-of-Direction in Healthy Individuals Using Trajectory and Head Motion During a Walking-Based Spatial Task

Sense of direction (SoD) is a stable individual trait that influences spatial behavior in everyday environments. While commonly assessed via self-report, this paper presents a multimodal, low-burden sensing framework to objectively estimate SoD in healthy individuals during a walking-based spatial task. Using only a wall-mounted camera and a head-worn inertial measurement unit, we reconstruct walking trajectories via pose estimation and capture head kinematics during the Walking Corsi Test (WalCT). Trajectory and head-motion features are fused at the participant level and filtered using a correlation threshold. This work introduces a multimodal, low-burden framework combining monocular pose-based trajectory reconstruction with IMU-derived head kinematics for objective estimation of sense of direction. SoD scores from the Santa Barbara Sense-of-Direction (SBSOD) questionnaire were discretized into three ordinal classes (weak, moderate, or strong) for classification. Multiple base classifiers were combined using weighted soft voting across the four trials of the WalCT. Using the selected feature set, the proposed approach achieved a weighted mean accuracy of 77% for three-class SoD classification, indicating that multimodal movement signatures during a simple walking task can support SBSOD-derived SoD classes estimation with minimal sensing burden.