3-D shape recognitions of target objects for stacked rubble withdrawal works performed by rescue robots

In this research, we aim to develop a method to recognize three dimensional shape of stacked rubbles each by each for rubble withdrawal rescue robots. Shapes, masses, states of stacked rubbles and so on are various and unknown at disaster areas. Then, grasping positions on rubbles and ways to remove them have to be considered for not breaking down the stacked rubbles and falling them down on victims. Thus, it is necessary to recognize stacked rubble individually and to identify their features, such as shapes, masses, center of gravity positions and so on. In this paper, we propose a 3-D object shape recognition system with a RGB-D sensor and a 3-D reference marker. Moreover, we also propose an extraction method of rubbles using the SSD (Single Shot Multi Box Detector) of the AI (Artificial Intelligence). Experiments were performed to confirm the validity of the proposed method with our constructed prototype of a rescue robot. Through the experiments, it is shown that target stacked rubbles were recognized individually.