The Small Moving Target Detection Methods In Heavy Fog Weather

Moving target detection is the first step in intelligent analysis.In practical applications,in addition to some common moving targets such as pedestrians and vehicles,there are some small moving targets that need to be detected.In addition,the collection of information can be affected by the weather.With the increase of foggy weather,the study of small moving target detection technology under heavy fog weather has a strong practical value in the transportation system,customs and space sectors.Robust principal component analysis is a hot research direction for moving target detection.This thesis is based on robust principal component analysis.By the study and analysis on the characteristics of fog images and small moving targets,robust principal component analysis method is used to realize the small moving target detection in a single fog image.The main research content of this thesis is as follows:(1)The principle of the robust principal component analysis method is expounded.And the matrix low-rank sparse decomposition model and related solving algorithms are emphasized.Because the local sub-images of the fog image are usually related to each other,an attempt is made to use robust principal component analysis method to solve the problem of small moving target detection in a single fog image.Firstly,the original single fog image is preprocessed to obtain the fog patch image,which satisfy the low-rank sparse decomposition condition.The problem of small moving target detection in a single fog image is transformed into an optimization problem of recovering low-rank and sparse matrices,which is solved by robust principal component analysis.(2)Robust principal component analysis method is used to detect small moving target in a single fog image.In matrix decomposition of fog path-image,three solution algorithms are introduced in this thesis: Iteration Threshold algorithm(IT),Accelerate Proximal Gradient algorithm(APG)and Inexact Augmented Lagrange Multipliers algorithm(IALM).In order to analyze the performance of these three algorithms,the simulation about detecting small moving target in a single fog image with the matrix low-rank sparse decomposition model has been done.The result of the simulation demonstrates that: the IALM algorithm has the highest detection accuracy compared with the IT algorithm and the APG algorithm.(3)To solve the problem that the matrix low-rank sparse decomposition model ignores the relationship between the target pixels.The structured sparsity-inducing norm has been introduced into the matrix low-rank sparse decomposition model.A novel model named structured matrix decomposition is proposed.About the solving algorithm,the solution of the structured sparsity-inducing norm has been combined with the IALM algorithm.In order to verify the effectiveness of the proposed method,the simulation about detecting small moving target in a single fog image has been done.The result of the simulation demonstrates that: the detection precision is higher than the previous IALM algorithm after introducing the structured sparsity-inducing norm.