Grid based fusion of off-board cameras

In this paper, we describe the perception system, based on a grid environment model, developed within the French project PUVAME. This system consists of several off-board cameras observing an intersection to detect objects (i.e. pedestrians, cyclists and vehicles). We present a generic and new method to design a sensor model for off-board cameras where each of the camera video stream is processed independently by a dedicated detector. In addition, to obtain tolerance to miss detections and false alarms, we model the failure of each sensor. Then, we detail how to build an occupancy grid, fusing the information from the different cameras. Experimental results showing that our approach is well suited to build an environment model are provided