Research On Underwater Multitarget Tracking Algorithm Based On Sonar Image

Forward-looking sonar is widely used in sea protection,fishery and deep-sea exploration,and is often used for underwater target detection,identification and tracking in military and civil fields.To address the problems of missed detection,missed tracking,wrong tracking and low tracking accuracy in practical applications,the paper investigates underwater multitarget detection and tracking technology based on sonar images,focusing on four specific parts:image preprocessing,multitarget detection,two-dimensional(2D)and three-dimensional(3D)multi-target tracking,to complete 2D and 3D stable tracking of multitarget.First,interpolation algorithms are used to downsample the sonar images.Improve the contrast of the sonar image and reduce missed detection through spatial and frequency domain enhancement.Due to the effect of underwater noise and side lobe,which lead to high false alarm rate after detection,noise reduction is achieved using filtering algorithms,and the beam is normalized to reduce side lobes.After the sonar image preprocessing is completed,K-means clustering and Cell average constant false alarm rate(CA-CFAR)algorithms are chosen for multitarget detection in water tank and field experiments.To address the problem of too many false targets after detection,the water tank walls are removed for each frame of detection results based on the rotation angle,and clutter is removed by area features.In addition,the detection effect of CA-CFAR relies on tuning parameters,and the K-mean clustering algorithm has the lower false alarm rate and missing alarm rate for targets.Then,based on the detection results,the PHD filtering algorithm is used for multitarget tracking,and three tracking indexes,the average OSPA distance,the number of trajectory fracture,and the missed tracking rate,are proposed to judge tracking effect.The2D-SMC-PHD has a wider application range and more stable tracking results than2D-GM-PHD algorithm.To address the problem of trajectory fracture and low tracking accuracy caused by missed detection of consecutive frames,a thresholded 2D-SMC-PHD(2D-TH-SMC-PHD)algorithm is proposed based on the 2D-SMC-PHD algorithm.By setting the threshold for the number of consecutive lost frames,the missing tracking rate is reduced and the number of trajectory breaks is reduced.By using the minimum variance(MSV)resampling method,the OSPA distance is reduced and tracking accuracy is improved.The2D-TH-SMC-PHD algorithm has been validated in simulation experiment with different detection probabilities,water tank and field experiments with different detection effects respectively.It has achieved tracking results for multitarget with non-fracture,high accuracy,and non-missing tracking.Next,2D multitarget tracking algorithms are improved to obtain 3D multitarget tracking algorithms.For 3D multi-target tracking,it is necessary to calculate the 3D measurement information based on the 2D information and the pitch angle of the target in the detection results.Through simulation experiment with different detection probabilities,the3D-TH-SMC-PHD algorithm also achieves tracking results with non-fracture,high accuracy,and non-missing tracking,but with lower tracking accuracy than the 2D-TH-SMC-PHD algorithm.Finally,the paper is summarized and an outlook is given for future practical 3D underwater multitarget tracking applications,where innovations can be made in the direction of 3D detection to reduce the error of 3D underwater measurement information.