Efficient Implementation Of Multi-Frame Detectionbefore-Tracking Technology For Millimeter-Wave Radar Based On GPU

In recent years,millimeter-wave radar,as one of the most critical sensors in autonomous vehicles,has gradually become the focus of attention and research.The detection performance of millimeter-wave radar for faint targets is degraded due to the detection threshold setting during the Detect Before Track(DBT)procedure,while the Multi-frame Track Before Detect(MF-TBD)technology can effectively improve the detection performance of millimeter-wave radar for faint targets by jointly processing the echo data in multiple frames.The improved detection performance is essential for the early detection of targets.However,most of the existing research on MF-TBD algorithms are based on simulation validation,and there is a lack of practical engineering application research demonstration due to the problem of the algorithm taking too time-consuming to meet the radar real-time requirements.we focus on the engineering of efficient GPU implementation of MF-TBD algorithm in the context of millimetre wave radar,and develop a set of MF-TBD processing system based on embedded GPU,the specific research content is as follows.1.The hardware working mechanism of millimeter-wave radar transceiver signals is studied,and the theory of millimeter-wave radar range,speed,and angle measurement is derived with the received LFMCW signal model,and the signal processing procedures of millimeter-wave radar based on DBT processing structure is analysed by means of real measurement data.2.To address the problem of poor detection performance of millimeter-wave radar for faint targets,a measurement model applicable to multi-frame energy accumulation under millimeter-wave radar is established,and the MF-TBD algorithm in the context of millimeter-wave radar is studied,and the analysis of simulated data concludes that the method has better detection performance compared with the traditional single-frame detection algorithm.3.To address the problem that the MF-TBD algorithm takes a long time to meet the demand for radar real-time,the characteristics of the algorithm are analyzed,a GPU is used as the algorithm implementation platform for the algorithm characteristics,an MFTBD algorithm based on GPU parallel architecture is designed,and an algorithm acceleration optimization strategy is proposed,which can effectively improve the realtime performance of the algorithm on the radar system.4.Based on the above research,a low-power embedded GPU-based millimeter-wave radar signal processing system is designed and developed for practical engineering needs,which implements the raw echo data acquisition function,signal processing function and display control function.The proposed algorithms are validated through simulation and real-world testing,and the results show that the algorithms can effectively improve the detection capability of millimeter-wave radar for faint targets.The developed system is validated and analysed through real-world measurements,and the results show that the system is able to perform the target detection function and meet the real-time requirements.