The Research On Track-Before-Detect Approach For Tracking Weak Targets

The emergence of weak targets,such as stealth fighters and small unmanned aerial vehicles,poses a serious threat to the air defense system.Therefore,research on joint detection and tracking of weak targets has become the forefront and hot spot of current theoretical research.Track-Before-Detect(TBD)approach is different from the traditional method of tracking after detection.It can utilize multiple frames of observation data to continuously accumulate target information,thereby achieving the purpose of enhancing the intensity of the target.This thesis has done a lot of research on the application of the TBD approach in various scenarios.This thesis mainly includes the following parts:The second chapter uses the Bayesian theory to construct a unified target tracking reasoning framework,and provides a theoretical foundation for the derivation of tracking algorithms.First,the recursive Kalman filter and non-recursive Gaussian Newton filter are derived using Bayesian theory,and then the equivalence proof of the two algorithms is given without considering the process noise and observation noise being uncorrelated.Then,with the help of the relevant knowledge of random finite sets,Bayesian reasoning was extended to the field of multi-target tracking,and the reasoning framework in the case of multi-target was given.So far,the unity of single-target and multi-target tracking reasoning was completed.The study provides a theoretical foundation for subsequent research.In the third chapter,for joint detection and tracking of a single weak and maneuvering target,the two algorithms,particle filter based TBD(PF-TBD)and efficient particle filter based TBD(EPF-TBD),are compared and studied.The research finds that EPF-TBD algorithm is superior to PF-TBD algorithm in both estimation accuracy and efficiency.Then,for each key step of the EPF-TBD operation,the impacts of the ceil number influenced by target,resampling algorithm,number of particles,target signal-to-noise ratio(SNR),and particle prior distribution on the performance of the filter were analyzed by simulations from two perspectives of accuracy and computing efficiency.These studies provide valuable reference and basis for rationally setting filter parameters,and also provide guidance for further improving the performance of filter.In Chapter 4,for joint detection and tracking a single maneuvering weak target,a multi-model(MM)based EPF-TBD(MM-EPF-TBD)algorithm is proposed,which can solve the divergence of filter due to model mismatch.Then two simulation experiments were designed,and the target motion model set consisted of two and three motion models,respectively.The proposed algorithms can complete the detection and tracking of the target well in the two scenarios.Therefore,we can increase the number of models in the model set when facing strong maneuvering weak targets.In Chapter 5,aiming at the problem of overestimation of target number of general probability hypothesis density based TBD(PHD-TBD)algorithm,a PHD-TBD algorithm combined with clustering algorithm is proposed.Simulation results show that the proposed algorithm can effectively solve the problem.Then,combining the Multi-Bernoulli(MB)filter and the TBD algorithm,the properties of the MB-TBD algorithm were studied,and an adaptive prior information MB-TBD algorithm was proposed.Simulation experiments show that the proposed algorithm can also work well even without the prior information of the target location.In Chapter 6,for the joint detection and tracking of multiple maneuvering weak targets,a multi-model method based labeled Bernoulli TBD filter(MM-LMB-TBD)algorithm is proposed.In simulation,four targets with different motion models were designed,which will finish four differet trajectories.The results show that the proposed algorithm can extract the track of each target and provide the estimation of the probability of motion models of each target,which verifies the label performance and motion model matching performance of the proposed algorithm.