Research On Recognition Algorithm Of Lateral Positioning And Complex Markings In Long Tunnels By Autonomous Vehicles

Autonomous vehicle and intelligent vehicles have become the development trend of the automobile industry at present.Vehicle positioning and vehicle environment awareness technology are two important research directions in the technology of driverless vehicles.It is the top priority of autonomous vehicle technology to make the vehicle have high-precision positioning and enable the vehicle to accurately capture and analyze the surrounding environment information.Starting from the two aspects of autonomous vehicle positioning and lane line detection,this paper proposes a vehicle lateral positioning system with multi-sensor fusion and a sliding window lane line detection algorithm with steering wheel angle sensor fusion respectively.The main research contents of this paper are as follows:(1)A multi-sensor fusion vehicle lateral positioning system is established,which mainly provides high-precision lateral positioning when the GPS signal is locked for a long time in a long tunnel.When the vehicle changes lanes or overtakes,the lateral displacement distance of the vehicle can be well calculated through this system to calculate the lane of the vehicle after the lane change.The whole system includes strapdown inertial navigation,monocular camera,binocular camera and odometer.The binocular camera measures the width of the lane through dynamic distance measurement,providing an input value for the whole system;Strapdown inertial navigation is mainly responsible for providing the vehicle’s heading angle information,judging whether the vehicle has lane change behavior through the difference of the heading angle in the adjacent time,and calculating the difference of the vehicle’s heading angle before and after the lane change;The odometer provides the distance traveled when the vehicle changes lanes;The monocular camera is responsible for assisting in judging whether the lane change occurs.(2)A vehicle lateral positioning algorithm is proposed.The difference between the vehicle’s heading angle at the beginning and end of lane change can be calculated by using the heading angle data calculated by SINS,and then the lateral displacement of the vehicle during lane change can be calculated by using the odometer data to finally complete the vehicle lateral positioning.The simulation analysis of the vehicle lateral positioning algorithm shows that the lateral positioning accuracy of the system has been greatly improved compared with the single strapdown inertial navigation positioning system under the same inertial components.The positioning accuracy is at the centimeter level,and the positioning error does not accumulate with the passage of time.The algorithm can still provide high positioning accuracy under the condition that the accuracy of inertial components is not high.(3)Aiming at the fact that the existing lane line detection algorithm can not recognize and track the lane line well in the curve with large curvature,this paper proposes a sliding window lane line detection algorithm combined with steering wheel angle sensor.First,the image taken by the front camera is converted into the perspective of the aerial view through the perspective transformation of the image,and then the radius of the curve where the vehicle is located is calculated through the steering wheel angle sensor information.The distance under the world coordinate system is converted into the pixel coordinate distance of the aerial view through the distance matching method,and the position of the sliding window in each aerial view is preliminarily predicted,If white pixel points are detected in the sliding window,the mean abscissa value of these pixel points is taken as the abscissa value of the center point of the sliding window.Otherwise,the predicted position is taken as the position of the sliding window.Finally,the lane line is merged and the aerial view is converted into a normal perspective through inverse perspective transformation.In this paper,the proposed lane line detection algorithm has been simulated and tested on a real vehicle.The simulation results show that the algorithm can track the lane line well in small radius curves with radius of 80 m,60m and 40 m,and the lane line finally fitted is highly consistent with the reality;The algorithm still performs well in the experimental road with the radius of 60 m and 42 m.The binocular camera is used to assist in locating the position of the first sliding window.When the vehicle is driving on the dotted lane line,the pixel statistics method may not detect the pixel points,resulting in that the first window cannot be set.Therefore,the binocular camera is used to assist in locating the position of the first window to make up for the shortcomings of the pixel statistics method.In this paper,the formula for locating the first sliding window with binocular camera is derived.The experiment shows that the pixel distance of the error in the aerial view is only 1 pixel,which can meet the positioning requirements of the sliding window.