Automatic visual surveillance for industrial robotic environments based on the temporal model of moving objects

The implementation of a safety device by using computer vision is discussed. Image sequence processing brings power to the study of dynamic scenes. The use of direct information coming from a camera fixed above the robot is convenient for modeling purposes. In this situation the model is created automatically with the real scene. Modeling is divided into two stages, acquisition of characters, and modeling of moving objects. The synchronized model is obtained by a special real-time tracking system in a preliminary cycle and is compared with the real scene at each sequence. The prediction of future displacements of moving objects obtained with the model facilitates the segmentation of the scene and, in the same way, helps the safety device reach decisions. An application to the surveillance of a five-axis robot and some statical objects is discussed