Research On Location Based On MAP Matching And Macro Path Planning In Intelligent Vehicles

Since early 1950 s,the intelligent vehicle has become a hot topic in the modern intelligent transportation systems.However,how to precisely position a commercial autonomous vehicle still remains open,since the accurate position is also one of the most important preliminaries for vehicle control.Therefore,high precision vehicle positioning and tracking techniques are the main motivation of this thesis.Precision 3D point cloud maps are popular to solve the positioning issue during driving.However,due to the high cost and lack of robustness of generating high precision maps,a reconstructing method of the 3D point cloud from some views is proposed in the thesis.It intends to simulate and construct the high precision maps with the method of 3D reconstruction based on multiple view.Three steps are included as follows,feature points detecting and matching,camera calibrating and 3D reconstructing.Meanwhile,in consideration of existing point cloud maps,their matching with the real-time point cloud data for the intelligent driving positioning is required.An improving measure is put forward and simulation results validate that our proposed method achieves less errors.Since the accurate position of an individual vehicle can be achieved,we then try to perform intelligent path planning based on the real time traffic prediction and evaluation.The traditional traffic flow data collection mostly depends on the large and fixed infrastructure with very apparent defects.At present,a replaced method appears based on remote sensing,by either satellite image or unmanned aerial vehicles.In the thesis,three methods to obtain vehicle density parameter in traffic information flow are mainly introduced.The first method is the prediction mode based on off-line with the time series analysis model of the combination of the SARIMA model and the GARCH model.The second method is based on the on-line observation,by extracting the histogram feature of directional gradient of positive and negative samples from remote sensing satellite images and classifying,thus to extract the vehicle targets.The third method using a Kalman filtering framework tries to combine the off-line prediction with the on-line observation.Simulation results show that the Kalman filtering method can efficiently enhance the estimated precision of vehicle density parameter.We then adopt the fuzzy inference system to evaluate the traffic flow information,through which the weight of each road section is output,and finally with a Q learningbased method.The weight of each road section is applied to complete the path planning of an intelligent vehicle.According to the proposed path planning method,we can achieve much better performance than traditional methods,such as the shortest path planning.