Research On Autonomous Parking System Based On Driving Space Construction

With the development of society and the continuous improvement of science and technology,autonomous driving technology has been booming in recent years.As an important part of the autonomous driving scene,autonomous parking technology can effectively relieve the traffic jam in the core area of the city and greatly facilitate people’s life.It is of great research significance.In this paper,aiming at the actual demand of autonomous parking system,a self-parking method based on driving space construction,which consists of measurement space,semantic space and behavior space,is proposed.This method can effectively improve the vehicle’s ability of perception and cognition in the parking environment,and ensure the safety and efficiency during the parking process.The specific research contents are as follows:Firstly,a SLAM system based on stereo vision is designed to meet the demand of vehicle localization in autonomous parking mission.The proposed SLAM system uses the keyframe technology to construct a sparse map of the environment and uses the local keyframe-based bundle optimization to maintain the local map.In addition,the proposed SLAM system uses a loop-closure detection module based on the bag-of-words and optimizes the global map using pose-graph optimization and global bundle adjustment.The SLAM system enables robust tracking of camera pose in indoor and outdoor environments,providing reliable positioning information for autonomous parking systems.Secondly,aiming at the demand of navigation map building of autonomous parking system,a 3D dense map construction method based on stereo vision is designed and implemented.The method uses the classical SGM dense stereo matching algorithm to obtain the depth map.Drawing on the idea of multi-view stereo(MVS),we proposed a probabilisticbased depth map fusion method to obtain more reliable depth information of the scene.Then,we use the refined depth map to update the status of a octree-based 3D grid map,which is used for vehicle navigation.Thirdly,in order to extract semantic information from parking environment,we propose a parking slot lane mark detection method based on ground segmentation and inverse perspective mapping algorithm.With the aid of depth map,6D pose of the parking slot in 3D space can be inferred.Then,a method of association between semantics and measurement map constructed by SLAM system is proposed,so the semantic information can be updated with the metric map without requiring further human labelling effort.Fourthly,we present a path planning algorithm for autonomous parking system.We focus on the kinematics of the car-like vehicle and improve the conventional geometric parking algorithm by proposing a new curve element named linearly steering spiral.A path planning algorithm based on smooth path searching and optimizing are presented.This method can generate smooth paths incrementally,and the reference control signals can be deduced directly once the path is determined.A simple closed-loop controller is designed in order to deal with the uncertainty from various aspects.Finally,experiments on the proposed algorithms are carried out,including SLAM algorithm,3D map construction based on SGM and depth filter,and parking path planning based on continuous curvature.Moreover,we present a real-time measurement-semantic hybrid map construction experiments to show the effectiveness and feasibility of the proposed method.