Generation And Application Of Navigation Map For Intelligent Vehicle Based On Stereo SLAM

The mainstream intelligent vehicles navigation algorithms highly dependent on high-precision GPS signals at present.It is difficult for intelligent vehicles to accurately locate itself,which further affects the self-vehicle's path planning based on current road information.What's more,the effectiveness and robustness of the decision control algorithm adopted by the self-driving vehicle is also difficult to guarantee.Therefore,in order to ensure the stability and reliability of the intelligent vehicle navigation system,it is necessary to develop a new type of navigation system to ensure that the intelligent vehicles can still operate stably and reliably even in an environment with poor GPS signals.This paper takes Chongqing Jiaotong University's Xingyuan intelligent vehicle as the research object,and researches on the key issue of Stereo-SLAM algorithm to construct a high-precision positioning system for intelligent vehicles.The main work is as follows.(1)The mainstream architecture of the current visual SLAM algorithm is analyzed in detail from three aspects: front-end motion tracking,back-end nonlinear optimization,and loop closure detection.The principle of ORBSLAM2 algorithm front-end feature extraction and positioning,back-end BA(Bundle Adjustment)optimization and loop closing detection intervention were deeply analyzed.(2)In response to the problem that ORBSLAM2 is prone to tracking loss when the vehicle speed is fast and the lighting changes are obvious,feature extraction optimizations have been put forward.For large-scale map applications,it is difficult to detect loops in weak texture areas,a new loop closure detection algorithm is designed.(3)In order to cope with the complex environment faced by autonomous driving and navigation,on the premise of retaining the important pose nodes of the ORBSLAM2 map,the offline and online scenes of the high-definition map are enriched,and the information inclusion capability of the high-definition map is further improved(4)In order to test the navigation practicability of the high-precision map established in this paper,on the basis of relying only on the high-precision map navigation,a vehicle autonomous driving navigation application algorithm based on preview control and fifth degree polynomial path planning is designed and transplanted ROS platform,which realizes efficient and hierarchical packaging of algorithms;(5)ORBSLAM2-based navigation map construction test and autonomous driving navigation test were carried out at the Automated Driving Test Base of Chongqing Automobile Testing Institute,which successfully achieved the tracking and obstacle avoidance tests of automatic driving tracking navigation.The experimental results show that the binocular vision intelligent car navigation map algorithm designed in this paper can be applied to parks and small and medium-sized automatic driving navigation,and can provide efficient and accurate positioning services for the realization of automatic driving in limited scenarios,as well as the static and current roads.The dynamic road condition information provides powerful support for the self-driving vehicles to perform good positioning and navigation.