Optimization On Visual Localization Under Constraint Of Road Map

Localization denotes the ability of mobile robot such as an autonomous vehicle to accurately determine its own position.Robust and accurate localization is a prerequisite for mobile robot navigation.In order to solve the task,classical navigation systems such as GPS and wheel odometry are widely used.Due to the defects of the classical localization methods,the visual localization methods have been developed rapidly in recent years.Visual odometry(VO)is a classical visual localization method,which has high cost performance and strong environmental adaptability.However,as a dead reckoning method,VO inevitably suffer error accumulation.Especially for mobile robots such as autonomous vehicles that require pose estimation over long distances,the problem will become more prominent,which will directly affect the wider application of VO.In this paper,an optimization method based on road map and particle filtering framework for autonomous vehicle in urban area is proposed to tackle this problem.To effectively match the ego-trajectory to various complicated road in map,a new representation based on anchor point(AP)which captures the main curving points on the trajectory is presented.Aiming the trajectory matching in large uncertainty,a flexible Multi-Position Joint Particle Filtering(MPJPF)method based on particle filtering framework is proposed to correct the accumulation error.Compared with traditional methods,the method in this paper has the following characteristics and advantages:(1)Only a relatively simple road map is needed to effectively correct the cumulative error of VO;(2)The method features the capability of adaptively estimating a series of APs jointly and only updates the estimation at situations with low uncertainty,which explicitly avoid the drawback of obliging to determine the current position at large uncertain situations;(3)The method is computationally inexpensive and has a certain real-time nature,which is suitable for application in unmanned vehicles.Experiments were carried out on the KITTI benchmark and the 10 km long-distance dataset,and compared with other similar methods.Experiments show that the proposed method can correct the error accumulation problem of VO and has superiority in accuracy and robustness.