Study On Detection Of Preceding Vehicle Based On Data Fusion Of Millimeter Wave Radar And Vision

With the increasing of car ownership and the rate of traffic accidents in China,technologies of autonomous vehicle and driver assistance have received extensive attention and research.Detection of preceding vehicle is one of the key technologies for the intelligent driving system.Accurate and real-time vehicle detection can provide effective decision-making basis for autonomous driving assistance system,and is of great significance for improving driving safety and driving comfort of vehicles.Millimeter wave radar and machine vision are two kinds of sensors commonly used for vehicle detection.Millimeter wave radar detects the position and speed of obstacles accurately,and has strong environmental adaptability,but it has difficulty in identifing different types of obstacles and being too sensitive to noise.Machine vision provides abundant information and has a great advantage in target recognition and classification with low cost,but its detection rate is not stable and is easily affected by the bad weather.Therefore,the fusion technology of millimeter wave radar and machine vision has been considered as an effective way to improve the accuracy and real time of vehicle detection.Accordingly,the road environment perception technology in the vehicle automatic driving system was considered as the research object in this thesis.The information in front of the vehicle was obtained by the camera and the millimeter wave radar mounted on the vehicle in real time,and the data receiving,processing and fusion algorithm of both sensor were proposed to achieve timely,accurate,reliable and environmental adaptability detection of preceding vehicles.The main research work of this thesis is as follows:1.Effective target determination of millimeter wave radar.The jamming targets were classified into invalid targets,stationary targets and non-dangerous vehicle targets by analyzing the target features detected by millimeter wave radar.The target life cycle,target area and signal-to-noise ratio threshold,target lateral distance and velocity threshold were used to eliminate each kind of jamming target.Thus,the vehicle targes were effectively extracted from the interference targets.2.Real time vehicle detection based on machine vision.The vehicle detection classifier was trained by combined the Adaboost machine learning algorithm and the Haar-like rectangle feature.A multi-scale optimization and region-of-interest segmentation method was proposed to reduce invalid searching area and invalidsearching windows,to improve the detection efficiency.The compressed tracking algorithm was used to track the detected vehicle in real time,to enhance the robustness of the algorithm under occlusion and environmental change,furthermore,the false positive rate and missed detection rate was reduced.3.Data fusion of millimeter wave radar and machine vision.The integration of millimeter wave radar and machine vision in space and time is realized by built the millimeter wave radar and vision data fusion model,which established the conversion relationship of millimeter wave radar coordinate,world coordinate,camera coordinate,image coordinate and pixel coordinate,as well as calibrated the internal and external parameters of the camera and unified the sampling time of sensors.The test results showed that the method proposed here could accurately identify the location of preceding vehicles,and had high detection rate and environmental adaptability under the condition of good illumination and clear vision.It could provide an effective decision basis for the intelligent driving system with the collision avoidance function.