Research On Classification And Recognition Technology Of Traffic Targets Based On Millimeter Wave Radar

In today’s complex and congested traffic road environment,automatic driving and assisted driving systems have become an important technology to ensure the safety of drivers and passengers and improve driving comfort.As one of the core sensing devices used in today’s assisted driving systems,millimeter-wave radar has the characteristics of all-weather high reliability in target detection.Studying the classification and recognition technology of traffic road targets using millimeter-wave radar can help the decisionmaking unit of the assisted driving system to fully perceive the driving environment,so as to obtain an effective assisted driving strategy.The thesis studies the method of classifying and identifying traffic road targets using millimeter-wave radar.The main traffic road targets of concern include pedestrians,nonmotorized vehicles（e.g.bicycles,motorbikes）,small and medium-sized vehicles（e.g.everyday sedans,SUVs）,and large vehicles（e.g.buses,trucks）.It mainly studies the method of using high resolution range profile for recognition and the method of using point cloud to realize recognition.The details are as follows.First,the method of target information acquisition using LFMCW millimeter-wave radar is studied.Based on the principle of radar signal,the mathematical principle of realizing target detection and target location parameter acquisition is deduced.Based on the deduced results,the "Range-Doppler-Angle" signal processing framework is studied,so as to realize the acquisition of target position information and motion state information.This study aims to provide data support for the following research on traffic target recognition methods.Secondly,a method for classifying and identifying road objects using the high resolution range profile is studied.The differences of common road objects in high resolution range profile are studied,and corresponding statistical features are extracted to construct feature vectors.A classifier was designed using the support vector machine algorithm.It realizes the classification and recognition of road targets under the premise of low computational complexity.The high resolution range profile acquisition process is relatively simple,and is less affected by the target distance.The classification and recognition method based on this can still perform well when the target is at a distant location,so it can be applied to complete target recognition in long-distance scenes.Finally,methods for object classification and recognition using point clouds are studied.On the one hand,the classification and recognition method based on the statistical characteristics of point cloud is studied,not only the method of using a single-frame point cloud for target recognition,but also the method of improving the classification accuracy by combining the multi-frame point clouds on the time series.All methods have achieved high accuracy in road target classification and recognition experiments.On the other hand,a classification and recognition method for extracting deep-level features from point clouds using Point Net++,which is designed based on the convolutional neural network,is studied,which further improves the effect of road target classification and recognition.The classification and recognition effect of the above algorithm is verified by using the data obtained in the actual traffic road scene.The results show that the algorithm designed in the thesis can well complete the classification and recognition task of the main traffic road targets.The results of the thesis also provide a new idea for further research on the method of using radar to realize road target recognition.