The occurrence of falls is an urgent challenge in our aging society. For wearable devices that actively prevent falls or mitigate their consequences, a critical prerequisite is knowledge on the user's current state of balance. To keep such wearable systems practical and to achieve high acceptance, only very limited sensor instrumentation is possible, often restricted to inertial measurement units at waist level. We propose to augment this limited sensor information by combining it with additional knowledge on human gait, in the form of an observer concept. The observer contains a combination of validated concepts to model human gait: a spring-loaded inverted pendulum model with articulated upper body, where foot placement and stance leg are controlled via the extrapolated center of mass (XCoM) and the virtual pivot point (VPP), respectively. State estimation is performed via an Additive Unscented Kalman Filter (Additive UKF). We investigated sensitivity of the proposed concept to model uncertainties, and we evaluated observer performance with real data from human subjects walking on a treadmill. Data were collected from an Inertial Measurement Unit (IMU) placed near the subject's center of mass (CoM), and observer estimates were compared to the ground truth as obtained via infrared motion capture. We found that the root mean squared deviation did not exceed 13 cm on position, 22 cm/s on velocity (0.56-1.35 m/s), 1.2°on orientation, and 17°/s on angular velocity.Keywords: human gait and balance, wearable sensors, state estimation, virtual pivot point, extrapolated center of mass, capture point, additive unscented kalman filter, fall detection