Research On Target Tracking Algorithm Based On Sonar Image

Sonar is the "eye" to detect the ocean,and the target detection and tracking algorithm is an intelligent means to make the "eye" clearer.With the development of high resolution sonar image,it is a challenging task to study the image processing algorithm based on imaging sonar to solve the problem of target tracking.Compared with optical image,the development of sonar image target tracking is relatively slow because of the low resolution of sonar image,different imaging principles and many other factors.The research on target tracking technology suitable for sonar images is of far-reaching significance in the strategic plan of ocean power in the 21 st century.In this paper,the target tracking algorithm based on sonar image is studied,and the target tracking algorithm with the main process of "detection and location-feature matching-filter tracking" is designed and verified in the sonar image data set.Firstly,this paper studies and designs the target detection and location algorithm and process based on sonar image.Based On the characteristics of the target imaging in the image data set of the DIDSON forward-looking sonar,a set of target detection and location algorithm is designed in this paper,which aims at improving the quality of the sonar image and facilitating the accurate detection and location of the target.Underwater target detection and positioning is realized through a series of processes such as filtering and denoising,morphological operation,threshold segmentation,target screening,maximum bounding rectangle selection and coordinate output,and its performance is verified through sonar data sets.Secondly,based on the target feature matching algorithm,the target feature set establishment and matching method based on sonar image are studied.The ability of gray feature,point feature and edge feature to represent the target is analyzed,especially the ability of these features to distinguish the aliasing of the target.Different feature extraction methods are used to establish the feature set of the target,and the subsequent inter-frame matching experiments are carried out to analyze the matching effect.After the sonar image sequence verification,the gray feature is selected to extract the target feature,and the normalized correlation coefficient matching is combined with the HU invariant moment to match and identify the target between frames.Finally,the extended Kalman filter algorithm based on interacting multiple model is studied,and the motion state prediction and estimation of the target are established.The purpose of the algorithm is to solve the uncertainty of the target motion state and the instability of feature matching due to aliasing,noise and other factors in the tracking problem.The feasibility of the designed algorithm is verified by simulation data experiments and sonar image data experiments.By improving the function of the overall multi-target tracking algorithm,the algorithm realizes the main functions of automatic detection and positioning of the target in the sonar image sequence,extracting and matching the target features by using the detection and positioning region,and filtering and tracking by using the matching results.The algorithm is implemented in Python language to process the sonar image sequence.The algorithm designed in this paper can get excellent tracking results in complex scenes with new targets entering and targets aliasing.