Fused robot pose estimation using embedded and external cameras

Recently, the interest for autonomous driving and unmanned aerial vehicle has drastically increased in the field of robotics. This very clear trend is now expending to everyday life devices, such as lawn mower. In this context we present an automatic lawn mower vision system capable of estimating its absolute pose using both a monocular camera embedded on the robot and an external camera observing the environment where the robot is functioning. This configuration is particularly desirable since the pose estimation of the robot on the lawn is challenging. This can be explained by the fact that the lawn contains only a few feature points which can be tracked reliably. In order to overcome this problem, we propose the use of an external camera to simplify the localization problem using a marker on the robot. Nonetheless, the pose estimation from the external camera alone is not sufficiently accurate for far distances. To improve its precision, we fuse the pose estimation from the surveillance camera with the relative motion of the robot computed from an on-board fisheye camera. We demonstrate the efficiency of the proposed method using real data captured by a wireless synchronized camera network.