Target Tracking And Recognition Based On Millimeter Wave Sensors

In recent years,autonomous technology has gradually become a hot field of research with the development of artificial intelligence.Autonomous vehicles try to perceive the surrounding environment through multiple sensors,and the aim of environmental perception focuses on target tracking and recognition.Therefore,a high-precision tracking algorithm and a highaccuracy recognition algorithm have a great impetus for the research of autonomous vehicles.In the target tracking process,due to non-line-of-sight propagation of echo signals and environmental noise interference,the outliers often appear in the observation information,which affects the tracking performance of the algorithm.In addition,traditional recognition algorithms are more sensitive to class imbalance problems in target recognition.However,the problem of class imbalance is widespread in the application scenarios of autonomous vehicles,especially on the highway,where there are far more vehicles than pedestrians to form a typical class imbalance problem.Therefore,how to reduce the impact of outliers on the tracking performance of the algorithm and solve the problem of class imbalance are important research topics that need to be solved urgently.With respect to outlier suppression in target tracking and the problem of class imbalance in target recognition,this thesis carried out research on target tracking and recognition based on millimeter-wave sensors.The main work and innovations of the thesis are as follows:1.The reason for the outliers during the target tracking process of the millimeter wave sensor is analyzed,and an outlier suppression method based on particle filtering is proposed for the non-linear mapping relationship between the target tracking observation and the state.This method reduces the impact of outliers in the observation data on the state estimation by introducing feedback into the covariance matrix of the observation vector.2.Aiming at the problem of instability of sampled particles in particle filtering,the grid filtering is further studied.The state space is limited to a certain area by introducing the a priori condition of state transition,thereby avoiding the problem of tracking precision degradation caused by the large deviation of sampled particles.An outlier suppression method based on multi-channel grid filtering is proposed.This method reduces the mutual influence between observations by introducing feedback to each observation,so as to make full use of the observation information containing outliers,and further improve the tracking performance of the algorithm.Finally,simulation experiments verify the effectiveness of the above methods.3.The existing target recognition technology based on millimeter wave radar and related research on class imbalance are studied,and a cost-sensitive convolutional neural network(CNN)is proposed.This method guides the learning of network parameters by introducing the evaluation indicators of class imbalance into the loss function,thereby making the network more sensitive to misclassification of the minority classes.At the same time,the network convergence speed is improved without changing the data set.4.Aiming at the problem that the convolutional neural network is insensitive to position information,a target recognition method based on a hybrid Support Vector Machine and Convolutional Neural Network(SVM-CNN)classification technique is proposed.This method makes full use of the typical physical characteristics of targets in the Range-Doppler image and recognizes the vehicle samples with typical characteristics through the improved SVM classifier.On the one hand,it can effectively alleviate the problem of class imbalance between vehicles and pedestrians;on the other hand,it can make full use of the location information of the target in the Range-Doppler image,thereby improving the accuracy of target classification.After that,the remaining samples are input to CNN for re-classification.Finally,simulation experiments verify the effectiveness of the above method.5.Based on 77 GHz millimeter wave radar sensor IWR1443 to build a test system,design the relevant parameters of the radar system.At the same time,the experimental data is collected,and the effectiveness of the method proposed in this thesis is verified by the measured data.