Research On Moving Object Detection And Tracking Algorithm Based Multi-Spectral

With the development of spectral imaging technology,spectral equipment will be miniaturized,intelligentized and integrated by traditionally bulky machines.The application field is also naturally transferred from the aerospace industry to the civilian market.Snapshot Spectroscopic Imager is a new portable product in this situation.With the development of science and computer technology,many achievements have been made in the detection and tracking of moving objects,many excellent algorithms emerged during the period,Practical applications are also becoming wider.However,there are still many problems in the complex and disguised scenarios.To solve this problem,multispectral data acquired by multi-spectral imaging technology is used for moving target detection and tracking.In general,the main research work and innovation of this article are as follows:1.The pretreatment of multispectral images.First,describe the principle and data characteristics of two snapshot spectral imagers;then study the calibration and color correction in the process of spectral data acquisition and give the corresponding solutions.In the process of preprocessing the raw data of the 9 spectral multispectral imager,Problems involving spatial location registration,Therefore,this paper describes in detail the block matching method based on gray information and the SIFT matching method based on features.Comparing the results of the two by simulation results,finally,the block matching method based on gray information is applied to the data preprocessing of the 9 spectral multispectral imager.2.Research on Multispectral Moving Object Detection Algorithm.During the study of moving object detection algorithms,Focuses on the detection of camouflage targets.This article introduced the principles of the three methods: interframe difference method,background subtraction method and optical flow method.Combining the characteristics of the collected data,select the optical flow method to detect the camouflaged target.Then,the flow of multi-spectral optical flow calculation is elaborated in detail,and the feasibility of the spectroscopic optical flow method is verified by simulation results.3.Spectral feature extraction.Spectral feature extraction is the basis for detection and tracking of camouflage targets.In the process of feature extraction of spectral information involves the selection of spectral bands and the parametric quantification of spectral features.Spectral band selection basis: First,how much information content of band or band combination;Second,the strength of correlation between bands;finally,the spectral response characteristics of the target to be identified in the study area.The greater the amount of information,the smaller the correlation,and the greater the difference between the target and the background spectrum,the more suitable it is for target tracking.Parametric description of spectral features mainly include spectral slope,spectral binary coding,spectral absorption index,spectral derivative,and spectral integration.4.Research on Multispectral Moving Object Tracking Algorithm.This chapter compares the target tracking strategy based on Bayesian posterior probability estimation theory,selecting particle filter as multi-spectral moving target tracking algorithm.During the tracking algorithm implementation,Choose a band with a significant difference between the target and background spectrum as the feature band,Feature extraction of spectral information,Selecting the correlation of spectral features as a measure of similarity,Select the state of the particle with the highest weight among all particles as the state estimation of the moving target in the next frame of image.Target tracking simulation of RGB images and multispectral images with the same algorithm,the results show that multi-spectral images have good accuracy and robustness for tracking camouflage targets.