Vehicle's Pose And Position Estimation Using Monocular Camera Based On Road Sign

Automobile intelligent driving technology develops fast in the past decade.As the basic component of intelligent driving technology,the vehicle ego-localization has become the key technique in this field.With the rapid development of computer vision and image processing/recognition technology,the vision-based vehicle ego-localization has become a research hotspot in both academic and industry.Moreover,the fast and accurate localization technology based on computer vision approach has potentially a wide range of applications in the fields of robot navigation,virtual traffic stream scene simulation and mixed virtual reality as well.At present,the speed of urban development is accelerating,the scale of the city is expanding day by day,and the road environment is becoming more and more complicated.The vehicle ego-localization in complex road environment puts forward higher requirements for intelligent driving technology.The precise ego-vehicle localization,the lane recognition and the vehicle attitude estimation at the complex traffic intersection are significant for the intelligent driving of the vehicle.Using a single sensor,such as a traditional GPS or inertial measurement unit(IMU),the positioning accuracy cannot meet the requirements of intelligent driving,and multi-sensor cascade system is sensitive to road congestion,so its positioning accuracy declines with the increase of cumulative error.In this article,we focus on the complex road environment of the city and propose a pose and position estimation method based on the plane perspective transformation.Combining with the multi-sensor cascade system,this method can provide another stable and reliable position and pose data reference while the precision of multi-sensor cascade system decreases.As the inherent facilities of urban roads,the road signs are visible commonly besides or above the roads,which are used to indicate road information and lane guidance information.The road signs are widely distributed,brightly colored and fixed in shape(rectangular),which are easily detected and recognized by visual methods.First,a simple database is built with the index of the precise positions of the road signs,including the length and width of the signs,and the number,width and direction of lanes under the signs.Secondly,the road signs are extracted from the images captured by the on-board monocular cameras,and the vertexes of the road signs are obtained by a method of edge lines detection and fitting.Then the homograph matrix is calculated,the position and pose parameters of the on-board camera in the road sign coordinate system are obtained.Finally,the experimental data are compared with the real measurement data.The experimental results show that the proposed method can accurately detect the road sign in different speed and light environment.Within 100 meters from the roadsigns,the vertex error of the sign is less than 3 pixels,the pose error is less than 1 degree,and the position error is less than 1 meter,which can reach the lane-level positioning accuracy.Through the comparison with the Compass high-precision positioning system L202,this method determines the lanes vehicle driving on more accurately.In this article,the background of vehicle ego-localization and the current research situation at home and abroad are introduced and expounds the research contents and algorithm flow are expounded firstly.Then,the image preprocessing and precision extraction of the quadrilateral vertex algorithm are described,as well as the method of calculating the position and pose parameters by the homograph matrix.The final of the article is the experimental analysis and discussion.