Autonomous Navigation Technique Of Intelligent Vehicle In Regional Traffic Environment

As an effective method to solve the urban transportation problem,Intelligent Vehicle Technology is drawing more and more attention. Thisthesis mainly proposes an autonomous navigation technique of IntelligentVehicle in regional traffic environment. There are four essentials: systemscheme designing, road boundary detection, autonomous navigation in roadsection and autonomous navigation in intersection.In the system scheme designing section, the navigation environment isanalyzed. It is divided into road section and intersection. A sensorsconfiguration of Intelligent Vehicle System is presented to solve the problemthat regional traffic environments are so complex that navigation would befailed. In road section, laser radars and camera are used to generate thevehicle's passable region. In intersection, multi-sensors are fused for localpath planning based on passable region method. The sensors scheme must fitseveral navigation requirements: road boundary detection by using the forward-looking laser radar installed on the front-top of the vehicle, lanedetection via camera, obstacles detection by the forward-looking laser radarinstalled on the front of the vehicle, global localization by GPS, deadreckoning by encoder and odometer.In the road boundary detection section, a road boundary detectionalgorithm based on laser radar is proposed to solve the problem that presentmethods cannot be applied in various road conditions. The algorithm can beefficaciously applied to different types of road conditions. Some linesegments are extracted through self-adaptive parameter based onenvironment characteristics after a series of preprocessing on range data oflaser radar. The extracted line segments are combined and further selected.Then unite the selected segments to match the road models and generate theroad boundary points which are tracked by Nearest Neighbor Filter.Extensive experiments conducted in urban and semi-urban environmentdemonstrate the robustness, effectiveness and viability of the proposedmethod.In the part of navigation in road section, a navigation method based onpassable region is put forward to be applied in complex or abominable roadsection. The method employs the Vector Field Histogram+(VFH+) toaccomplish vehicle's local path planning. The passable regions are determined by fusing lane, road boundary and obstacle information. A costfunction is used to select the final direction of the intelligent vehicle. Thismethod effectively figures out the problem of navigation failure due to singleroad detection technique in complex environment. Experiments demonstratethat this method is stable, reliable and robust. Finally, an algorithm ofobstacle avoidance and overtaking basing on double-arc path planning isproposed.In the section of autonomous navigation in intersection, a navigationmethod based on passable region is presented to solve the problem that thereis not sufficient road surface information, which can be used to localizationand navigation. Global localization is used to determine whether the vehicleis near certain intersection. In order to achieve intersection localization, alocal coordinate system is built by the stopline or zebra crossing which isdetected by computer vision. Then dead reckoning is employed for thesubsequent localization. Basing on the finite environmental information andthe geometric structure characteristics of intersection, some fictitiousobstacles are generated to determine the passable region for steer angleplanning. Experiments indicate that this method is effective and credible.