In this paper, we present a method to reconstruct the configurations of kinematic trees of rigid bodies not using measurements of relative angles (such as, e.g. rotary encoders at joints) but absolute posture sensors (such as IMUs) along with suitable filter algorithms. We argue that the relatively larger inaccuracies shown by absolute sensors can be compensated by suitable processing, such as a passive complementary filters exploiting the Mahony-Hamel formulation. The proposed method is applicable to systems where measurements of relative angles is not feasible or convenient, or where the joint kinematics are not lower pairs: for example, human body parts or soft robotic devices. In the paper, we make explicit reference to the reconstruction of posture of the compliant, underactuated Pisa/IIT SoftHand. Quantitative comparisons with ground truth data in grasping tests are used to validate the proposed method. The resulting hardware design is mechanically robust, cheap and can be easily adapted to robotic hands with different structures, as well as to sensorizing gloves for studying human grasping strategies.