Continuum robots represent an actively developing and fast-growing technology in robotics. To successfully implement control and path planning of continuum robots it is important to develop an accurate three-dimensional shape and position sensing algorithm. In this paper, we propose an algorithm for the three-dimensional reconstruction of the continuum robot shape. The algorithm is performed during several steps. Initially, images from two cameras are processed by applying pre-processing and segmentation techniques. Then, the gradient descent method is applied to compare two-dimensional skeleton points of both masks. Having compared these points, it finds a skeleton of the robot in a three-dimensional form. Additionally, the proposed algorithm is able to define key points using the distance from the robot base along the center line. The latter allows controlling the position of points of interest defined by a user. As a result, the developed algorithm achieved a relatively high level of accuracy and speed.