Research On Method Of Traffic Lights Detection, Tracking, Mapping And Recognition

Traffic lights detection is one of the most basic technologies for a driverless vehicle.This dissertation presents a system of detection, tracking, recognition and online mappingusing different kinds of information, including the image sequence captured by a cameramounted on the vehicle and the position and attitude information provided by GPS/INS orSLAM.The sequential results of detector, which is treated as observations with uncertaintyunder the influence of distraction, are associated and filtered by data association algorithmand multi-target tracking algorithm. Then, the results of tracking will be classified intodifferent shapes. To support this process and improve the tracking performance, a motionmodel of traffic lights is designed. Because the localization of the traffic lights affects theprediction of its motion when the vehicle is close to the intersection, an online mappingmodule that estimates3D location of traffic lights is designed base on multi-view analysis.Because the motion model and the mapping module are related to the measurements of thevehicle position and attitude, a simultaneous localization and mapping (SLAM) module isdesigned in order to substitute GPS/INS. Since a lot of different kinds of information areincluded in the system, the method of calibration is described too.