Research On Monocular Visual SLAM System

Simultaneous localization and mapping(SLAM)is one of the key technologies in computer navigation.The problem of robot locating and mapping in unknown environment is solved by SLAM which is a precondition of finishing tasks by robots themselves.So SLAM has great potential of research and application.To improve computing efficiency and robustness of monocular visual SLAM,a label based extended Kalman filter SLAM(EKF-SLAM)system and a dual view image fusion SLAM system are built according to two popular methods that are called as filter and optimization in SLAM.The label based EKF-SLAM system takes quick response code encoded with serial numbers as labels and uses EKF algorithm to deal with state estimation of the system's back end.The observation of EKF is the distance and angle between labels and vehicle heading.The control of EKF is the driving distance of the two rear wheels.The system can correct the error of dead reckoning and improve the accuracy of positioning.But its computational efficiency is decreased rapidly with the increase in the number of labels.To solve the problem above,a label screening strategy is proposed in this paper which only preserves labels within observation ranges and retains one best labels outside at regular intervals.Simulation results show that this strategy can save nearly 90% computation time and hardly affect the accuracy of pose estimation in EKF-SLAM.So it's an effective strategy for improving the real-time performance of SLAM systems.The dual view image fusion SLAM system is an extension of the classic monocular vision SLAM system.Two cameras can obtain different angle of view's environmental features than one within the same time period which makes the dual camera fusion SLAM system more robust.In this paper,the way of processing two cameras' pose update,pose tracking and map optimization are transformed into the way of processing one's by the fixed spatial relationship between the two cameras.It's easy to extend monocular SLAM to multi camera SLAM.Practical driving experiments show that the dual view image fusion SLAM system can achieve accurate positioning in the outdoor dynamic environment.