<![CDATA[Obesity is known to disrupt motor control and biomechanics; however, detailed gait alterations in individuals with obesity remain underexplored, particularly in dynamic and real-world walking conditions. This study aims to quantitatively characterize gait differences between individuals with obesity and those of normal weight by analyzing postural and temporal gait parameters. The investigation focuses on pitch, roll, and cadence dynamics using body-worn inertial sensors, with phase transition modeling via Hidden Markov Models. This work proposes a novel framework that integrates multi-location Inertial Measurement Unit (IMU) sensors and a Hidden Markov Model–Supervised Marginal (HMM-SM) approach to detect and classify gait phases with high accuracy, offering practical value for clinical gait assessment and personalized rehabilitation. IMU sensors were placed on the waist, thigh, calf, and heel to record gait data from participants in both obese and normal-weight groups. Gait segmentation and phase modeling were conducted using 4-, 5-, and 8-state HMMs. Quantitative analysis revealed significantly greater postural variability in the obese group during slow walking, with standard deviations in roll and pitch reaching 20.68° and 9.23°, respectively—much higher than the normal-weight group (0.60° and 0.26°). Hidden state transitions from 5-state pitch HMMs showed a very strong effect size for the obese group (Cramér’s V = 0.72) compared to a moderate effect for the normal-weight group (V = 0.33). Similar patterns were observed for roll and cadence. In terms of segmentation accuracy, the 4- and 5-state HMMs outperformed the 8-state model, achieving accuracy levels above 99%, while the 8-state model reached only ~93%. The findings demonstrate that obesity significantly alters gait dynamics, particularly in postural stability and gait phase transitions. The proposed IMU-based HMM-SM framework effectively captures these changes, offering a reliable tool for gait analysis in clinical and biomechanical applications.]]>
