Research On Fusion Perception Of Monocular Vision And Inertial Navigation Based On Lane Changing Scene

This paper is mainly aimed at the research of smart car in the scene of lane change,fusion monocular camera and inertial navigation information to perceive the road environment.The lane line detected by the camera vision algorithm is integrated with inertial navigation to plan the ideal trajectory of keeping or changing lanes in real time,and at the same time,the position of the vehicle and the ideal trajectory is positioned by inertial navigation.error.In order to reduce errors,the camera is first calibrated to obtain parameters such as distortion coefficients,internal and external parameter matrices,inverse perspective transformation is performed on the road ahead to obtain a top view,and the top view is converted to the geodetic coordinate system.Then use "gradient difference method" and "color space LUV method" to separate the lane lines,and select the vertical integration projection map as the information fusion carrier of the two.For these two figures,the sum of the pixels that meet the requirements in the same column is taken as the vertical coordinate of the vertically integrated projection figure,and the horizontal coordinate of the corresponding column is taken as the horizontal coordinate of the projection figure.Convert the top view into a grayscale map,find the peak position of the lane line in the integrated projection map,locate the search window in the grayscale map,record the position of the internal point of the window,and perform quadratic curve fitting to obtain the lane line equation.In order to ensure the accuracy of the peak position,two adaptive adjustment algorithms are designed in this paper: the dynamic update algorithm of the peak position and the threshold adaptive adjustment algorithm,which can make the algorithm more robust,and can also adaptively adjust the original setting in the image processing algorithm.Set threshold,performance will not decrease under changing road conditions.For the lane change process,this paper has designed two modes: lane keeping mode and lane change mode.The vehicle is in lane keeping mode by default,and the center line of the lane line equation obtained by the visual algorithm is tracked as an ideal trajectory,so that the vehicle keeps driving at the current lane center.At the same time,use the position of the center line of the lane in the previous frame to calculate the approximate position of the peaks represented by the lane lines on the integrated projection map.When the visual algorithm is affected by random errors or the failure tolocate the peak position in the vertical integration projection map,assist the algorithm Make corrections and quickly adapt to the current ground environment.Experimental data shows that under good road conditions,the detection accuracy rate is improved by about 3%,and under bad road conditions,the algorithm accuracy rate can be stabilized at about 95%.When the lane change signal is received,the lane change mode is turned on,and the system immediately plans an ideal trajectory of a quartic polynomial according to the position of the vehicle and the intended arrival position of the target lane.The inertial navigation is used to locate the vehicle in real time,and the distance between the vehicle and the ideal trajectory is combined to form a closed-loop control about the position.The actual vehicle experiment shows that the vehicle can complete the mutual switching between the lane keeping mode and the lane changing mode with the help of inertial navigation based on the visually detected lane line.