Research On Some Key Techniques Of Stereo-vision Localization With IMU In Dynamic Environment

With the development of robotics,robots will enter more areas and make people's life more convenient.The key technology to make robot into more areas is autonomy of Robot,and the precise positioning is the basis of autonomous navigation.Although the traditional global positioning system has high accuracy,it has received many restrictions.And the use of binocular stereo vision can get the robot's autonomous positioning,but a single visual sensor,can perceive the movement of the amount of change,do not know their own and the relationship between the geodetic coordinate system,while the gyroscope can give the geodetic coordinate system,so this paper uses the gyroscope assisted visual positioning.In this paper,we studied the mobile robot autonomous positioning algorithm,analyzed the existing system used technical's advantages and disadvantages.Based on the above research?analysis and some questions in the actual application,we carried out the following work:(1)Stereo vision localization process can be divided into the following steps:image acquisition,feature extraction,feature matching,feature tracking and motion estimation.The traditional visual localization method can only get the camera's changed movement,and can't know the true position relationship between the localization coordinate system and the geodetic coordinate.In this paper,we uses IMU information to assist the visual positioning,and provides the geodetic coordinate system for vision localization system.(2)when the system is used in dynamic environment,because visual localization system get the camera's location through the feature point position changed,feature point which on the dynamic object would affect the accuracy of motion estimation.In this paper,we using scene flow delete the feature point which on the dynamic object,improve the accuracy of this system.(3)When the robot back to the starting position or the position,if robot can detect this information through visual information,which can reduce the accumulated error.This paper realizes closed-loop detection method through bag of visual words,when the robot back to the starting position,our system can detect closed-loop.Finally,we designs a binocular stereo visual localization system,which can improve the accuracy of the data in vertical direction,through ground plane by gyroscope provided,and it can return accuracy result in dynamic environment.It can detect closed-loop,when the robot back to the starting position.In this paper,the localization algorithm is realized in Ubuntu operation system,and verification of this system through practical experiments.The accuracy of visual localization is improved by analyzing the experimental data.