Reliable Extended Target Detection And Tracking In Complicated Environments

Perception technology has extremely important application value in transportation,military,agriculture and other fields.In the context of low latency,large capacity,and high speed,more and more application scenarios place higher and higher requirements on environmental perception.At the perceptual level,target detection and tracking have a pivotal position,and its related technical achievements have penetrated into all aspects of people's lives.When the detection environment is complex,it is more complicated to use traditional sensing means and methods to complete the detection and tracking of the target.And because the target has a certain spatial structure,the previous processing methods often ignore the target's multi-scattering point model.Therefore,it is challenging to perform accurate and reliable target detection,tracking and recognition in a complex environment.This article uses Linear Frequency Modulated Continuous Wave(LFMCW)signals on the radar transceiver,which can obtain the relative distance,angle and speed of the target on each individual radar.In order to distinguish the target from the interfering target,this article applies the micro-Doppler effect and combines the Empirical Mode Decomposition(EMD)method to effectively identify typical targets.In addition,in order to reduce the impact of scattered points in a complex environment,this article try to use a probabilistic data fusion framework to appropriately "pair" the observations from multiple radars.After the "pair" results are processed by the clustering method,the use of the minimum mean square error estimator in each type of data set can significantly improve the positioning accuracy of the target.On the other hand,for the extended target multi-scattering point model,this article analyzes the extended target joint tracking and recognition algorithm based on random matrix under non-linear measurement.For the non-linear characteristics of the measured value,this article draws on the idea of extended Kalman filter to make it linear.In order to effectively identify the target while tracking,it is necessary to fully model the prior size and shape information of the target in the form of pseudo-measurement into the target tracking framework based on a random matrix.Then,we can simultaneously obtain the probability density function of the extended target state and the probability mass function of the object category within the framework.After obtaining the orientation of the target through maximum likelihood estimation,the closed form of the estimator and classification probability can be derived.This article focuses on the above two parts of the research content,and verifies the feasibility of the proposed algorithm through system modeling,theoretical analysis,mathematical derivation and numerical simulation,and uses the software-defined radar platform to test and verify the algorithm,paving the way for the effective application of the algorithm in practice.