Research On Traversability Mapping Methods Based On Multi-source Perception In Outdoor Environments

Traversability mapping is one of the basic and important modules in the navigation system.Accurate,efficient and stable acquisition of traversability information in the environment is the basis for ensuring the safe driving of unmanned vehicles.The current mainstream traversability mapping methods mainly use LiDAR to perceive the environment.However,due to the large amount,sparseness and uneven distribution of LiDAR point cloud data,it usually brings the accuracy and real-time performance problems,which poses a certain challenge for unmanned vehicles to obtain environmental traversability information accurately and efficiently in the outdoor environment.In addition,with the increasing complexity of application scenarios,the limitations of single source perception,for example,single-LiDAR perception is difficult to identify environmental information with insignificant structural features.Single-visual perception usually performs better on the training set,but problems such as poor generalization performance and insufficient robustness to environmental changes are prone to occur in real-world environments,which also brings some challenges for unmanned vehicles to obtain traversability information stably in large-scale outdoor scenarios.Therefore,how to use the complementary characteristics of multiple sensors to fuse multi-source observation information to overcome the limitations of a single perception is also particularly important in the traversability mapping of ground unmanned vehicles.This dissertation focuses on the problems of insufficient accuracy,real-time performance and robustness of the current traversability mapping methods in outdoor environments.Through the efficient modeling of multi-sensor environmental observation information,this dissertation carries out the research on LiDAR-based and LiDARvision-based traversability mapping methods,etc.The main work and contributions are as follows:(1)Aiming at the problems of low accuracy and efficiency in the existing traversability mapping methods based on sparse LiDAR point cloud,a traversability mapping method based on multilevel B-spline surface approximation is proposed.Multilevel B-spline surface approximation is applied to support descriptive terrain modeling.The Kalman Filter is introduced to solve the problem of terrain map fusion between consecutive multiple scans.Then,a geometric features based traversability analysis process is conducted to construct a traversability map.The proposed method can realize large-scale traversability mapping in outdoor scenes,and effectively improves the accuracy and real-time performance of traversability mapping.(2)In order to further improve the accuracy and robustness of the traversability method,a traversability method fused with environmental semantic perception is proposed.Visual semantic information is introduced to correctly identify the cluttered areas that are difficult to be identified by LiDAR,and at the same time,the misclassifications in semantic segmentation are further eliminated through LiDAR perception.The proposed method can achieve more accurate and robust traversability mapping results in outdoor scenes.(3)Based on the previous two parts of work,a path planning method integrating traversability information is proposed to further verify the effectiveness of the proposed traversability mapping method at the navigation level.By introducing the traversability information into the state sampling and path searching process of path planning,this method realizes the measurement of the traversability mapping performance at the perception level and the navigation level,and verifies the practicability of the proposed traversability mapping method and the path planning method in mobile robot navigation.