Adaptive threshold for robust segmentation of mobile robots from visual information of their own movement

In this work, a solution for robust motion segmentation of mobile robots is presented. Motion segmentation is obtained from the images acquired by a calibrated camera which is located in a fixed position in the environment where the robots are moving, and without incorporating invasive landmarks on board the robots. The proposal is based on the minimization of an objective function that depends on three groups of variables: the segmentation boundaries, the 3D rigid motion parameters (components of linear and angular velocity) and depth (distance to the camera). For the objective function minimization, we use a greedy algorithm which, after initialization, consists of three iterative steps. The accuracy in the results and also the processing time are closely related to the initial values of the involved variables. GPCA technique is used for curve initialization, comparing the reconstruction error with a threshold. Two approaches (fixed and adaptive) are proposed to set that threshold. The experimental tests carried out have proved that the proposed adaptive threshold increases, notably, the robustness of the system against lighting changes.