This study aimed at the real-time estimation of the lower-limb joint and torso kinematics during a squat exercise, performed in the sagittal plane, using a single inertial measurement unit placed on the lower back. The human body was modeled with a three-degrees-of-freedom planar chain. The planar IMU orientation and vertical displacement were estimated using one angular velocity and two acceleration components and a weighted Fourier linear combiner. The ankle, knee and hip joint angles were thereafter obtained through a novel inverse kinematic module based on the use of a Jacobian pseudo-inverse matrix and null-space decoupling. The above-mentioned algorithms were validated on a humanoid robot for which the mechanical model used and the measured joint angles virtually exhibited no inaccuracies. Joint angles were estimated with a maximal error of 1.5 deg. The performance of the proposed analytical and experimental methodology was also assessed by conducting an experiment on human volunteers and by comparing the relevant results with those obtained through the more conventional photogrammetric approach. The joint angles provided by the two methods displayed differences equal to 3±1deg. These results, associated with the real-time capability of the method, open the door to future field applications in both rehabilitation and sport.