Real-time Decentralized Monocular SLAM

This paper presents a real-time Decentralized Monocular SLAM process. It is the first time, to our knowledge, that a decentralized SLAM with vehicles using only proprioceptive sensors and a single camera is presented. A new architecture has been built to cope with the problems involved by a decentralized scheme. A special care has been given to the data incest problem. It is solved thanks to a substate system. Network aspects and computational time are also considered. By using an Extended Kalman Filter and sending only essential information, we are able to make our decentralized algorithm suitable for an important number of vehicles. We also introduce a way to retrieve the distance between vehicles and so to put the different maps built in a common frame. This approach was tested using a realistic simulator with different trajectories.