Research On Multi-target Tracking Algorithm And Software Module Design Based On Forward-looking Sonar

In tasks such as marine resource exploration,territorial waters security maintenance,and routine underwater operations,a certain level of perception ability of the underwater environment is required.Multibeam forward-looking sonar,as a commonly used underwater acoustic sensor,and can provide two-dimensional image information of the surrounding underwater environment.By detecting and tracking targets based on the two-dimensional sonar image,the position,velocity,category,and size information of the targets can be obtained,which is of great significance for improving the perception ability of the underwater environment.This paper focuses on the problem of multi-object tracking based on forward-looking sonar,and from the perspective of real-time and practicality,it deeply studies the algorithms of multi-object tracking based on forward-looking sonar to improve the robustness of target detection and tracking algorithms.Furthermore,this paper expands the target tracking task to enrich the usage scenarios of forward-looking sonar.The main work contents are as follows:1.The sonar image preprocessing methods are studied.To solve the problem of forward looking sonar image restoration,a forward looking sonar image restoration method based on look-up table method is proposed,which reduces the time consumption of the image restoration process to 28% of the traditional bilinear interpolation algorithm.After that,the performance of the three edge preserving algorithms was verified and analyzed using actual sonar data,and the results showed that the non local mean filter had the best denoising and edge preserving performance.2.The target detection algorithm based on sonar images is studied.The YOLOv5 series of target detection models are trained and tested using the dataset constructed in this article,focusing on the impact of model size and data enhancement methods on model performance.The results show that the YOLOv5 s model has the best detection accuracy compared to the other three size models.At the same time,this paper designs a data enhancement method based on Mosaic enhancement and Copy-Paste enhancement methods.The model trained using this data enhancement method has the highest detection accuracy.3.The target tracking algorithm based on sonar images is studied.According to the characteristics of the target tracking problem in forward looking sonar images,the target matching strategy in the Deep SORT target tracking algorithm is improved by integrating the motion,appearance,and size information of the target to complete the matching between the target and the trajectory.The algorithm was validated through multiple sonar image sequences,and the results showed that the probability of ID switching in the improved target tracking algorithm was reduced by 60.7% compared to the original Deep SORT algorithm.4.A multi target tracking software module based on forward looking sonar is designed and implemented.The software module integrates the above target detection and tracking algorithms,and this article also designs corresponding process communication,target size,and speed estimation functions for it.