Research On The Fusion Of Stereo Vision And Inertial Sensor On Vehicle

The fusion of visual-inertial sensors have become an important research direction in the field of autonomous navigation with its complementarity and autonomy.At present,the accuracy of stereo vision is relatively high,but it is susceptible to dynamic objects and illumination condition in the environment,which makes the system unstable;while the inertial navigation system collects stable data and is not easily affected by the external environment,but its positioning error accumulates seriously.Therefore,the fusion of stereo vision and inertial navigation is very important.Based on the stereo vision positioning technology and inertial navigation technology,this paper explored the fusion method of the two types of sensors deeply.The main research is as follows:Firstly,the sensor information fusion was carried out by using the unscented kalman filter.The state equation and the observation equation of the loosely-coupled fusion method were constructed by the navigation information of the inertial navigation model and the stereo visual odometry.Then the loosely-coupled fusion method would be compared with the existing monocular-inertial fusion method which used the trifocal tensor constraints in three different movement modes of the vehicle.Finally it was verified that the loosely-coupled method was effective in linear motion mode of the vehicle.Secondly,Due to the existing monocular-inertial fusion method's accuracy in curve movement mode,the trifocal tensor constraints was adopted in this paper,which was applied in the stereo-inertial system.The tightly-coupled fusion method effectively overcame the scale drift of the monocular vision and the inertial navigation prediction error of the scale.It also increased the robustness of the system under the interference of external moving objects.Thirdly,through the advantages of loosely-coupled fusion method and the tightly-coupled fusion method in three different motion modes,this paper designed the switching condition of the loosely/tightly-coupled method to form a new fusion method.And then the effectiveness of the proposed method was verified.Then due to the disadvantages of the loosely/tightly-coupled switch fusion method in height calculating,this paper adopted loosely/tightly-coupled fusion method based on the three view geometry constraints.The method effectively solves the problem of the long time motion's height estimates.Finally,based on UKF observability matrix,an observable analysis of the four fusion methods metioned in this paper was used to illustrate the reasons for the convergence of the stereo visual-IMU odometry using loosely/tightly-coupled fusion method based on the three view geometry constraints.