Research On Indoor Robot Positioning System Based On Monocular Vision And IMU Fusion

At present,the main indoor robot positioning methods are dead-reckoning with wheeled odometer or IMU,visual positioning and indoor pseudo satellite positioning.Dead reckoning has good positioning accuracy in short time and short distance,but there will be cumulative error.The accuracy of indoor pseudo satellite positioning is very good,but the cost is too high.Visual localization is a new research direction,it works well in static background.But it’s not so good in dynamic background.Currently,positioning schemes combining different methods have emerged to improving the positioning accuracy of indoor robots.In this paper,the localization scheme combining monocular vision odometry and IMU is chosen to study the dynamic obstacle uncertainty in the working environment of indoor mobile robots from the perspective of cost and performance.In this paper,the following aspects are investigated:(1)To address the root cause of degradation or even failure of visual odometer positioning accuracy in dynamic environments,the matching speed and matching accuracy are optimized by improving two mechanisms,which are based on ORB;Loop detection works well in the positioning scheme of the integration of monocular vision odometer and IMU proposed.(2)IMU sensor data is not accurate,low pass filtering and complementary filtering are used to optimize the original data of IMU.Aiming at the problem that the monocular vision positioning algorithm lacks absolute scale and can’t feedback the real pose information,according to the inherent drift characteristics of IMU,a strategy to restore the absolute scale of monocular vision positioning is proposed.(3)Considering the efficiency of the fusion positioning system,the Maximum Correntropy Kalman Filter(MCKF)algorithm is applied to the robot positioning,and the data of the monocular vision odometer and the IMU sensor are filtered and fused through Mc KF to improve the accuracy.Under the robot working environment,aimed at the performance question that nondeteminacy of dynamic obstacle influences vision positioning system.Improve complex events processing and matching system of feature points.Enhance positioning accuracy for vision positioning system and tolerance of dynamic barrier.Based on the feature of IMU accumulates error,come up with the monocular vision milemeter recovery measure strategy,and use Maximum correlation entropy Kalman filtering and loopback detection majorization based on the word bag model to improve the accuracy of coalesce data and remove accumulate error of positioning system.By experience verify.Based on the feasibility and positioning accuracy of IMU accumulates error and the monocular vision milemeter.Meanwhile,raise the tolerance limitation of dynamic barrel and robot sport rate for this project.