The large majority of legged robots currently employ ball-feet or flat-feet. More recently soft feet have been introduced, to improve walking performance on uneven grounds. Nevertheless, their novel adaptability requires sensor systems beyond traditional Force/Torque sensors to estimate the distribution of forces on the contact surface. This letter shows how a perception layer realized with Inertial Measurement Units allows a soft foot to reconstruct not only the shape of the foot - hinting at the shape of the ground beneath - but also, under precise hypotheses, the contact force distribution. The problem is theoretically formalized and analysed with a quasi-static approach in the Sagittal plane. Then, theoretical results are experimentally validated in a simplified foot-ground interaction scenario. The force reconstruction provided by the proposed method allows to correctly identify the sole contact location arising from obstacles with radius down to 1 cm.