Research On Low-Observable Multi-Pedestrian Tracking Based On Millimeter-Wave Radar For Urban Autonomous Driving

All-weather and high-precision pedestrian target detection and tracking are important guarantees for road traffic safety in autonomous driving urban road scenes.Compared with traditional pedestrian detection system based on vision,millimeter-wave radar detection distance,strong penetrability,privacy security,etc.Therefore,this dissertation studies pedestrian target detection and tracking methods based on commercial automotive millimeter-wave radar sensor.However,for radar detection systems,pedestrians are typical low-observable targets,which are mainly reflected in:(1)The small radar cross-section(RCS)of pedestrians is only about-5d Bsm,which leads to weak radar reflection signal.(2)In urban road scenes,the radar signals reflected by pedestrians are submerged in the strong clutter signals brought by vehicles and road surfaces,which brings challenges to all-weather pedestrian detection and tracking.In this dissertation,challenges encountered in radar pedestrian weak target detection and tracking are studied.The main work and contributions include the following aspects:1)Research on joint estimation method of pedestrian target motion parameters based on millimeter-wave radar.Radar has the advantages of all-weather,small size,and low cost.This dissertation adopts millimeter-wave radar for studying the pedestrian target detection and tracking method.Based on the space-time characteristics of Frequency Modulated Continuous Wave(FMCW)radar Chirp signal and antenna array,the motion parameters of pedestrian target are estimated by time-frequency analysis method,including distance,velocity,and azimuth.In addition,this dissertation focuses on Multiple Input Multiple Output(MIMO)virtual antenna extension technology to achieve high-resolution estimation of pedestrian target motion azimuth under the condition that the number of receiving channels of low-cost radar is limited.2)Research on pedestrian target detection and tracking algorithm based on RFT coherent accumulation.Aiming at the problem that the radar echo signal of pedestrians on urban roads is interfered with by other clutter signals,a pedestrian detection and tracking algorithm based on phase-coherent accumulation is developed.According to the Range-Slow time signal model of radar,the phase difference caused by the coupling of velocity and slow time is compensated,and the target signals are accumulated in the slow time dimension.On this basis,the Constant False Alarm Rate(CFAR)method is used to achieve pedestrian target detection.Finally,based on pedestrian motion parameter information,algorithms such as Kalman filter and particle filter are used to track pedestrian targets in automatic driving scenarios.3)Research on integration algorithm of pedestrian multi-object detection and tracking based on SMC-TBD.Aiming at the problems of weak pedestrian echo signals and difficult detection in urban road scenes,which lead to poor pedestrian tracking performance,research on integration algorithm of pedestrian multi-object detection and tracking based on Sequential Monte Carlo tracking before detection(SMC-TBD).Based on recursive Bayes idea,Markov state transition matrix is introduced to estimate the motion state of pedestrian target by posterior probability approximation.On this basis,the sequential Monte Carlo method is used to solve the posterior probability of the target state.Among them,the radar range-velocity-azimuth spectrum without threshold processing is directly taken as the measurement data,avoiding the false alarm or missing alarm caused by the unreasonable setting of detection threshold,as well as the wrong association between measurement and known target trajectory.The motion state and detection probability of pedestrian targets are represented by random samples with presence state,thus realizing the integration of pedestrian target detection-tracking in low signal-to-noise ratio(SNR)scenarios.Finally,Extensive simulation and experimental results are carried out based on commercial automotive millimeter-wave radar.The results show that the SMC-TBD-based pedestrian multi-target detection-tracking algorithm proposed in this dissertation exhibits better performance than the traditional methods in tracking multiple pedestrian targets in low SNR scenarios.