Simultaneous Localization And Mapping On Unstructured Environment Based On Binocular Vision Of Autonomous Vehicles

Vision based simultaneous localization and mapping(SLAM)is the trend of using visual method of intelligent vehicle in the field of navigation,and also in recent years,the hot research problem.The jumping and drifting of positioning which always occurs in the traditional positioning method of integrated navigation when there is a high building nearby can be avoided.Therefore,it has important applications in the automatic driving scene with higher obstacles around.Based on the on-board binocular camera of unmanned vehicle,this paper studies the SLAM problem of unmanned vehicle in a static scene with higher obstacle.The SLAM back-end optimization problem has been divided into open and closed loop problem according to whether the path route will return to the starting point or not.And the optimization model for both the open-loop and close-loop based on graph optimization method has been built.The open-loop optimization model first propose a key frame unit structure composed of multiple key frames and the corresponding 3-dimentional feature as the minial optimization unit.Then the model constructs a dispatcher and a optimizer based on multi thread.The dispatcher is responsible for constructing the key frame unit under a certain size window.And the optimizer optimizes the key frame unit the dispatcher constructed.The model efficiently improves the positioning accuracy and stability under open-loop conditions.In the closed loop optimization model,a closed-loop optimization strategy based on global segmentation is proposed.The key frames are constructed into several key frame units according to certain rules to optimize,and the errors are reduced by multi iterations.Finally,the open-loop and closed-loop optimization are coupled to ensure the consistency of the optimization in the presence of both open-loop and closed-loop.In the end,the open loop and closed loop model have been validated in this paper by using the data in the visual data set KITTI.Then,taking the laboratory unmanned vehicle as the experimental platform,the car experiments were carried out in the campus environment,the sheltered off-road environment and the sheltered forest environment.Experimental results show that the optimization model proposed in this paper can effectively reduce the localization error in off-road environment and occlusion environment,and optimize the SLAM positioning results.