The equilibrium shape of a continuum robot is resulted from both its internal actuation and the external physical interaction with a surrounding environment. A fast and accurate shape estimation method (i) can be used as a feedback to compensate for more accurate motion; and (ii) can reveal rich information about physical interactions (e.g. instrument-anatomy contacts / forces during a surgery). From a prior work that demonstrated an offline calibration of continuum robots, we adopt its shape modal representation and error propagation models that include identification Jacobians. In this work, we present an iterative observer approach to enable online shape estimation. We develop a dual Extended Kalman Filter (EKF) to estimate both the robot state and the shape modal parameters. The dual EKF provides robust estimation on (i) the configuration space variables that are controllable and driven by internal actuation; and (ii) the modal coefficients representing homotopies of shape families that are governed by the physical interactions with the environment. We report results from simulation studies in this work, and plan to investigate methods in the future to use the proposed approach for predicting physical interactions.