Vehicle Automatic Positioning Method Based On Multi-sensor Information Fusion

With the increasing traffic problems and the development needs of various industries,the technology of intelligent vehicle is changing rapidly.As a major aspect of intelligent vehicle technology,environmental perception has attracted more and more attention and research,and accurate positioning is the premise of environmental perception.Traditional positioning technology,such as Global Positioning System(Global Positioning System,GPS),is used widely.However,GPS has many problems,such as reflection multipath and low signal update frequency,which leads to inaccurate positioning in some cases.Therefore,it is very necessary to use other positioning methods to assist positioning.This thesis is focus on the localization method based on multi-sensor information fusion for intelligent vehicle,to solve the positioning accuracy problems caused by cumulative error and Signal shielding of traditional localization methods.Firstly,this paper introduces the localization method based on Dead Reckoning(DR),which applies the vehicle kinematics model,to design the odometer positioning algorithm based on ROS system.For the defects of error accumulation,this paper using inertial measurement unit(IMU)to revise the positioning result,which reduces the influence of cumulative error.Secondly,the visual odometer positioning algorithm based on feature points is introduced.In the design of the visual front end,this paper using image processing to pre-process the image data,such as image segmentation,median filtering.Due to the nonuniform distribution of the feature points.this thesis proposes a method of extracting feature based on image Gaussian pyramid,which extracts feature points by reducing image's resolution.The improved visual front end has a greater improvement in stability,Based on the stable visual front end,this paper completes positioning of the visual odometer by using PNP and other algorithms.For the defects of using single sensor to locate,such as inaccurate positioning and unstable positioning.This topic fuses the information of visual sensor and inertial sensor to design a visual-inertial odometer(VIO)positioning method based on multi-sensor information fusion.First,the VIO models and calibrations the error of IMU,the IMU pre-integration method is used to match the IMU and the visual sensor in time,the IMU error term and the visual reprojection error are used to establish a optimization objective function,and the objective function is nonlinearly optimized,Completed the positioning of the visual-inertial odometer.In order to compare the advantages and disadvantages of several positioning algorithms,this paper conducted simulation tests and platform tests on the above three positioning methods,and evaluated the test results according to relevant standards.The test algorithms are implemented in C ++ language.The results show that the positioning method based on the wheel odometer has the lowest positioning accuracy and the maximum cumulative error.The positioning method based on the visual sensor has higher accuracy,but it is easy to failure in high-speed and high-frequency movement.Vision-inertial odometer positioning algorithm has the highest positioning accuracy,its absolute positioning error is only about 50 mm,and it can adapt to high-speed,high-frequency movement.