Real-time Localization in Outdoor Environments using Stereo Vision and Inexpensive GPS
ICPR '06 Proceedings of the 18th International Conference on Pattern Recognition - Volume 03
Constructing dependable certainty grids from unreliable sensor data
Robotics and Autonomous Systems
Bridging the gap between feature- and grid-based SLAM
Robotics and Autonomous Systems
Variable sized grid cells for rapid replanning in dynamic environments
IROS'09 Proceedings of the 2009 IEEE/RSJ international conference on Intelligent robots and systems
Mechatronics and Intelligent Systems for Off-road Vehicles
Mechatronics and Intelligent Systems for Off-road Vehicles
| Hi-index | 0.00 |
The presence of automation and information technology in agricultural environments seems no longer questionable; smart spraying, variable rate fertilizing, or automatic guidance are becoming usual management tools in modern farms. Yet, such techniques are still in their nascence and offer a lively hotbed for innovation. In particular, significant research efforts are being directed toward vehicle navigation and awareness in off-road environments. However, the majority of solutions being developed are based on occupancy grids referenced with odometry and dead-reckoning, or alternatively based on GPS waypoint following, but never based on both. Yet, navigation in off-road environments highly benefits from both approaches: perception data effectively condensed in regular grids, and global references for every cell of the grid. This research proposes a framework to build globally referenced navigation grids by combining three-dimensional stereo vision with satellite-based global positioning. The construction process entails the in-field recording of perceptual information plus the geodetic coordinates of the vehicle at every image acquisition position, in addition to other basic data as velocity, heading, or GPS quality indices. The creation of local grids occurs in real time right after the stereo images have been captured by the vehicle in the field, but the final assembly of universal grids takes place after finishing the acquisition phase. Vehicle-fixed individual grids are then superposed onto the global grid, transferring original perception data to universal cells expressed in Local Tangent Plane coordinates. Global referencing allows the discontinuous appendage of data to succeed in the completion and updating of navigation grids along the time over multiple mapping sessions. This methodology was validated in a commercial vineyard, where several universal grids of the crops were generated. Vine rows were correctly reconstructed, although some difficulties appeared around the headland turns as a consequence of unreliable heading estimations. Navigation information conveyed through globally referenced regular grids turned out to be a powerful tool for upcoming practical implementations within agricultural robotics.