Research Of Target Tracking Method For Intelligent Vehicle Based On The Information Fusion Of Millimeter Wave Radar And Camera

In the environment perception system of intelligent vehicle,the target tracking method based on the information fusion of millimeter wave radar and monocular camera is an optimal solution considering the performance and cost of sensors.The key and difficulty of target tracking system lies in accurate sensor measurement and data association method with small calculation.Because the detection performance of a single radar or monocular camera is easily affected by the specific traffic environment,it often leads to the error of target association in the scene of multiple adjacent targets.Therefore,it is of great theoretical significance and engineering application value to study the target tracking method of information fusion of millimeter wave radar and monocular camera in general road scene.Based on the information fusion of two kinds of heterogeneous sensors,millimeter wave radar and monocular camera,an intelligent vehicle target detection and tracking method for general road scene is proposed in this thesis.The specific contents are as follows:1.The research background,current situation and existing problems of target detection and target tracking based on information fusion are analyzed systematically.The principles and basic methods of target detection and target tracking for intelligent vehicle are introduced.2.The method of target detection based on millimeter wave radar and monocular camera is studied under the specific condition of vehicle and road in traffic environment.For on-road target detection based on millimeter wave radar,a method of target detection using constraints of road width and vehicle speed for radar data processing and target feature analysis is proposed.For on-road target detection based on monocular camera,a deep neural network is used to recognize the target from the image sequence of monocular camera.The constraint conditions among the vehicle,pedestrian and road are set.The horizontal planes can detect the distance and spatial position of the road target with monocular camera.The real vehicle experiment verifies the validity of the above method in the on-road targets detection of intelligent vehicles.3.The method of target tracking of multi-sensor fusion based on pure angle association is studied.Aiming at the problem that the target distance information obtainedby monocular camera is inaccurate,which easily cause error in data association.This thesis proposes a data association method only using angle information,which transforms the one-step prediction information of target state from Cartesian coordinate system to polar coordinate system,and obtains more effective measurement information through sector tracking gate.On this basis,the threshold of data association is designed,the angle information is used for data association,and the measurement information successfully associated is transformed into Cartesian coordinate.The parallel filtering framework based on Kalman filter is used for fusion filtering,so as to realize the target state estimation and target tracking of multi-sensor information fusion.The simulation results verify the effectiveness of the proposed method in terms of estimation accuracy and computational efficiency.4.The requirement analysis,software and hardware design and implementation of the target tracking system based on the above methods are completed.And the simulation analysis and real vehicle experiment are carried out to verify the feasibility and effectiveness of the new method applied to the actual vehicle driving environment.