Research On Intelligent Perception And Recognition Of Traffic Objects Based On Millimeter Wave Radar

With the continuous development of industrial sensors,AI technology and 5G communication technology,intelligent vehicles,autonomous vehicles and other technologies have been rapidly developed.To improve the reliability of autonomous vehicle by processing and analyzing the data collected by vehicle sensors has become a hot research topic at present.By modeling traffic objects with artificial intelligence,sensor data can be effectively utilized,so as to mine the characteristics of different traffic objects and identify different traffic objects.By combining the data collected by the sensor with the proposed target motion model and Kalman filtering algorithm,the entire established system model is closer to the actual traffic scene,and the accuracy of prediction and tracking of traffic objects are improved.This thesis designed a traffic object perception and recognition system by analyzing the data collected by millimeter wave radar sensor: millimeter wave radar echo signal of target feature extraction algorithm in the complex environment,object recognition algorithm based on Support Vector Machine and deep learning network and target tracking algorithm based on UKF combined with target motion models.Based on the three models and algorithms that make up the system,the research results are as follows:(1)MMW radar signals were studied by 2D-FFT and CFAR,the characteristic parameters of targets were extracted by FFT,and the possible radar clutter interference was eliminated by CFAR.The simulation results show that the algorithm can accurately and effectively extract the necessary features from traffic target radar echoes.(2)A traffic object recognition algorithm based on SVM and deep learning network was proposed.Based on the data set of millimeter-wave radar traffic objects collected by the selfdesigned radar acquisition experiment,a new variable,namely DRCS,is designed by statistically analyzing the data of different objects.By mining the new radar data features,this study establishes the dynamic characteristics of different traffic objects in different motion states,thus providing a basis for the identification of these traffic objects.After introducing DRCS,the previous traffic object recognition algorithm is optimized.The simulation results show that by employing support vector machine and other machine learning algorithms,our method performs better than previous studies.(3)This thesis proposed a traffic target tracking algorithm based on millimeter wave radar,through the establishment of target motion model of a nonlinear CTRV(Constant Turn Rate and Velocity magnitude)to simulate the traffic object motion and UKF(Unscented Kalman Filter)algorithm is completed to track traffic object.By employing track initiation and data association algorithm,this thesis realized the target and the allocation of the measured radar data in the FOV(Field of View)of the radar sensor,which provides the correct data for tracking.The effectiveness and stability of the whole traffic target tracking system are verified in two scenarios of uniform linear motion and complex turning motion utilizing the target radar motion data collected in the actual traffic scene.Finally,this thesis not only summarized the research achievements,but also pointed out the shortcomings of the whole research.Finally,it looks forward to the future research work.