This thesis is dedicated to the 3D reconstruction of thin tubular structures, such as cables or ropes, from a given image sequence. This is known to be a challenging task, mainly because of self-occlusions of the structure and its fine details. This new approach combines image processing tools with physics simulation to faithfully reconstruct jumbled and tangled cables in 3D. This method estimates the topology of the tubular object in the form of a single 1D path and also computes a topology-aware reconstruction of its geometry. This method is evaluated on both, synthetic and real datasets and demonstrate that this method favourably compares to state-of-the-art methods.