Research On Multi-Frame Track-Before-Detect For The Extended Target Based On Millimeter Wave Radar

Target detection and tracking are crucial functions of modern military and civilian sensory information systems.The advent of high-resolution sensors such as millimeterwave radar has made target information more abundant,and targets usually occupy multiple resolution cells.Currently,the mainstream extended target detection and tracking algorithms employ the single-frame detect-before-track processing framework.In this framework,the detected point traces past the threshold are fed into the tracking filter to estimate the target state.However,due to the complexity of the extended target and its environment,the processing framework of single-frame detect-before-track is susceptible to target miss detection and false alarms in strong clutter or low signal-tonoise ratio scenarios.On the other hand,the multi-frame track-before-detect technique directly processes multiple frames of raw echo data,which significantly enhances the detection performance of faint targets with strong clutter background.Nevertheless,the multi-frame track-before-detect technique primarily targets the point target model,and limited research has extended the technique to the field of extended target tracking.Therefore,it is essential to establish a suitable extended target model and accurately estimate the target motion state and shapes parameters to provide support for subsequent processing,such as classification and identification.In address to the above problems,this thesis focuses on the research related to the millimeter wave radar extended target multi-frame track-before-detect algorithm and engineering applications,the main research work and contributions are as follows:1.To address the issue of missed target detection or high false alarm rate in singleframe DBT algorithms,we propose a multi-frame joint likelihood ratio-based multi-frame track-before-detect algorithm.This algorithm provides a mathematical model for extended target tracking at the echo level within the framework of Bayesian estimation theory.It derives a closed expression of the multi-frame joint motion state-size parameter likelihood ratio and utilizes the shape characteristics of extended targets to accumulate target energy within frames based on real or complex measurement values.This approach effectively improves the utilization of extended target information.The proposed method achieves higher target detection probability,tracking accuracy,and shape estimation accuracy compared to traditional detect-before-track algorithms.2.In order to track extended targets in a low signal-to-noise ratio environment without background noise and target a priori information,we propose a track-beforedetection algorithm based on multi-scale kernel correlation filtering.Our approach uses the kernel correlation filtered correlation response map as a novel detection statistic and combines it with a scale adaptive estimation strategy.This enables us to estimate the scalable extended target center-of-mass motion state and shape parameters,leading to robust extended target tracking.Our proposed algorithm outperforms existing methods in terms of its ability to achieve accurate and reliable tracking,even in challenging conditions.3.Based on the above research,an engineering implementation scheme based on GPU parallel processing platform is proposed for the poor real-time performance of the extended target MF-TBD algorithm,and a parallel acceleration scheme for the MF-TBD algorithm with iterative processing is designed,which effectively improves the real-time performance of the proposed algorithm in millimeter wave radar system processing.All the above works are verified by simulation experiments and millimeter wave radar real measurement data,which prove the effectiveness and stability of the proposed algorithm in the field of extended target tracking.