Font Size: a A A

Key Technologies Of Positioning System Of Unmanned Autonomous Vehicle In Urban Environments

Posted on:2017-04-16Degree:DoctorType:Dissertation
Country:ChinaCandidate:J M KangFull Text:PDF
GTID:1312330536951955Subject:Intelligent Transportation Systems Engineering and Information
Abstract/Summary:PDF Full Text Request
The real-time and accurate positioning of the autonomous vehicle is an important guarantee for its safe running on road.But in the urban environment,there are some semienclosed areas surrounded by tall buildings,dense trees,resulting in the global navigation satellite system(GPS)positioning error increases or completely fails to work.To solve the above problems,this paper proposes an integrated navigation and positioning system,and depicts its key technologies.The main innovations of this paper are as follows1.An autonomous vehicle platform meeting the urban environment is proposed.In the platform,the vehicle can sense all the objects within a bounded spherical space,which are on different heights and distances.In the decision planning system,many prior rules and scenarios set in the knowledge base according to China's traffic laws and regulations,which helps change the serial traffic incidents in time domain into networks.The above strategies reduce the time consumption and computational resources of the vehicle during the process of decision planning.The motion control systems of vehicle are connected with CAN bus,which is superior to the add-on control system in terms of response sensitivity,control precision and safety performance.2.A demodulation scheme of Neumann-Hoffman(NH)codes is proposed.After capturing the satellite signal,the fast Fourier transform(FFT)is applied in the frequency domain and time domain to search the NH code of the terminal and the signal accuracy.The hardware resource is fully used,and the processing speed is fast.The experimental results show that the proposed scheme can be applied to receive both GPS and Beidou signal,and it provides a new idea for the integration of the global navigation satellite system.3.A segmentation method of solid objects of the 2D laser radar cloud data is proposed,which is a kind of ”first segmentation,and then fusion”.Different segmentation methods with traditional data,the first segmentation,segmentation method in fusion ”is divided into” on the basis of the maximum detection range segmentation method,based on the adjacent observation point spacing segmentation method ”and” on the basis of observation point distance fusion method ”in three parts,respectively,to achieve environmental entity between” geographical segmentation ”,” outlook,the background segmentation ”and”environmental entity ”fusion of three different needs.The segmentation method is based on the judgment of the distance between the point cloud clusters in the observation point cloud,and the segmentation method is to segment the cluster,which is especially suitable for the data segmentation of the ”crown” and other loose entities in the urban environment.4.A neural network based on the environmental characteristics of the target classification method.Different from the common environmental characteristics of ”first identification and classification” extraction method,the method in the analysis of types of data samples using neural network,not only by the maximum output node type neural network classifier to determine the input samples,the neural network classifier is analysis of the weights of the output node,to determine the effectiveness of classification the.Finally,the feature extraction of the effective classification is carried out.The method can effectively restrain the proportion of ”false positive” and ”false negative” in the classification result,and improve the stability of feature classification.5.A method based on Delaunay Triangulation is proposed,which is based on the unique representation of environmental characteristics.In the observation data of the2 D laser radar,there is no data to describe the environmental entity.Therefore,it is not unique to identify the environmental features only after the classification.In this study,we use the Delaunay Triangulation method to give the unique geometric relationship between the environmental characteristics.The method opens up the distance barrier of environmental characteristics in data association,and extends the correlation between features from the local space of the ”nearest neighbor” method to the global space.However,due to the use of triangle matching method,the number of environmental characteristics required must be greater than or equal to three.6.A fast path loop closure detection method for the autonomous vehicle is proposed.The method includes two steps: 1)the possibility of detecting loop path.In order to detect the error as much as possible to reduce vehicle pose error is introduced,the method of observation data will be mapped to the two-dimensional symmetric Gauss space,the accumulation of all observation points Gauss mapping value and scene recognition as one of the elements,combined with the distance between the vehicle and the number of features,the vehicle heading pose is calculated on the basis of the possibility of the occurrence of frame loop path the selected loop higher possibility of observation frames of ”circle of observation frame set.The first step calculation is simple,can achieve fast screening of the observation frame.2)path loop accuracy.Method to extract each group of loop observation frames between ”environmental characteristics of triangular geometry using Delaunay Triangulation,instead of the full amount of data frame matching with the matching of the two most similar triangles,and save all” and the matching error of the matching results of the loop observation frame ”.The matching result with the minimum matching error is chosen as the constraint condition of the optimal cumulative error.The second step do not use the full amount of data to match the frame,which avoids the interference of the clutter data to the matching result,and improves the matching precision.Experimental results show that the proposed method can stably and accurately detect the environment features,and can quickly and accurately identify the path loop points;the environment features map built by the proposed method has high precision,the building profiles in the point cloud map have a high coincidence with the real ones.
Keywords/Search Tags:Unmanned/Autonomous Vehicles, Positioning System of Intelligent Vehicle, Global Navigation Satellite System(GNSS), Simultaneous Localization and Mapping(SLAM), Feature extraction, Loop Closure Detection, Triangulation
PDF Full Text Request
Related items