Research On Underwater Multiple Target Detection And Data Association Algorithm

Along with the increasing demand for marine resource exploration and underwater operations,autonomous underwater vehicles(AUV)are receiving growing attention all over the world.Taking into account the complexity of underwater environment,underwater acoustic detection is currently an effective way of underwater exploration.The acoustic-visual tracking system based on forward-looking sonar is of wide application value and great strategic significance in the fields of coastal defense,operational surveillance,maritime safety operation and marine development.With the focus on multi-target tracking problem based on AUV front view sonar,studies are carried out on image processing,target detection,tracking and data association with sonar as the visual sensor.The main contents are as follows.First of all,image processing based on forward-looking sonar is studied according to the characteristics of underwater acoustic images.The echo data is obtained through software programming,and the original image is obtained by coordinate transformation and beam interpolation.Based on the characteristics of original image obtained by forward-looking sonar,a mean-accelerating median filtering algorithm and an improved Pal-King fuzzy image enhancement method are used in image preprocessing.The adaptive threshold segmentation algorithm based on fuzzy distance is used to segment the processed sonar image.Finally,the improved algorithm of connected region is used to merge the segmented image.Secondly,the research on underwater target feature extraction is carried out,and the underwater target detection method is researched based on principal component analysis.Considering the real-time requirement of sonar tracking system and the complexity of underwater environment,ten common features of sonar image are extracted,including area,length of major axis,length of minor axis and seven invariant features.The feature selection method based on principal component analysis is then used to reduce the dimensionality of the above features before a set of optimal feature combinations is chosen to cluster the objects.In the pool test,the data of ten characteristics of three different targets were extracted and processed to verify the effectiveness of the proposed method.Finally,the research on data association method in multi-target tracking is carried out.A multi-objective trajectory file system is set up for the acoustic-visual tracking system,and an underwater multi-target data association algorithm based on clustering of cloud-like models is proposed to improve the tracking performance.The proposed algorithm is compared with the traditional nearest neighbor data association and probability data association algorithms.The comparison results show that the algorithm has the characteristics of accurate clustering and fast convergence.Therefore,the method proposed in this paper has achieved better results in tracking accuracy and velocity.