Multi-target Tracking Of AUV Based On Forward Looking Sonar

Due to the need of resources exploration and underwater operation,the technology of Autonomous Underwater Vehicle(AUV)is increasingly valued at home and abroad.The underwater acoustic detection has been the most effective way to detect underwater environment,and sonar can be used to detect and process the real-time ocean environmental information.Then target position,type,velocity and other information can be gotten.So the underwater target detection and tracking technology is mainly based on forward looking sonar.And the research is significant to target recognition and tracking,as well as obstacle avoidance and navigation.In this paper,a multi-target tracking problem of AUV is mainly studied.A single beam forward looking sonar is used as the acoustic vision sensor,to obtain the underwater multi-target information.And a multi-target tracking system is established based on AUV.Specific research contents are as follows:In this paper,the research on processing of forward looking sonar images is studied.This paper compares the different features of sonar images and optical images.Then median filtering is used to denoise,and an improvement of median algorithm greatly increases the filtering speed.The fuzzy image enhancement algorithm is also studied,by improving the fuzzy membership function to solve the low gray value loss problem in the classical Pal-King fuzzy image enhancement algorithm.Research on enhancement algorithm is conducted,which is based on Particle Swarm Optimization(PSO),by selecting the best optimal parameters to enhance the PSO effect.The improved adaptive dual-threshold region growing algorithm is used for image segmentation,which uses Otsu method to obtain a low threshold.Finally,image defect fitting is used for morphological processing.In this paper,the research on multi-feature fusion in underwater multi-target tracking is studied.Firstly,the feature description method of targets is studied,and 30 kinds of target features are extracted,then the dimension of features is reducted by sequential forward selection(SFS)and sequential forward selection(SBS)based on generalized regression neural network(GRNN),in order to select the optimal combination of features.The basic principles of particle filter and a variety of features fusion strategy are analyzed.An adaptive fusion scheme is put forward,and a fuzzy controller is used to improve the weighted sum fusion.The proposed fusion strategy can determine the particle weight,while the tracking condition changes.Finally,the effectiveness of the adaptive feature fusion is proved by compared with other fusion strategy.In this paper,the research on data association in underwear multi-target tracking is studied.Considering the targets into and out of the sonar scan range,the trajectory management and tracking method are combined,to establish a multi-target tracking file system,in order to start and end the trajectory at real time.Research on association algorithm using nearest neighbor data association(The Nearest Neighbor Data Association,NNDA)with particle filter(Particle Filter,PF)is under taken and the algorithm model of NNDA-PF is established.A method of joint probabilistic data association-particle filter(The Joint Probabilistic Data Association-Particle Filter,JPDA-PF)is proposed,and it is based on the feature matching.The feature matching and the relationship between the current measurement and the trajectory are introduced in to the weight calculation of particle filter.Comparison tests are carried out to compare the tracking results of particle filter,NNDA-PF and improved JPDA-PF,to verify the validity of improved JPDA-PF.In this paper,the architecture of multi-target tracking system is proposed.And simulation,tank tests and sea trials were carried out for verification of multi-target tracking algorithm.Results of underwater multi-target tracking tests verified the effectiveness,timeliness and reliability of multi-target tracking system presented in this paper.