Research On Target Detection And Tracking Technology Using Imaging Sonar

Visual information processing technology of imaging sonar has very important meaning and application in both military and civilian.Imaging sonar visual processing can not only assist obstacle avoidance for underwater vehicle,and navigation,but also helpful to underwater environment mapping,appointed target detection,and moving target tracking.The signal of imaging sonar is easy to be understood,at the same time,it is robust to turbidity of the water.However,low resolution,information missing,unsteadiness between frames,multi-path disturbance and side-lobe disturbance exist in imaging sonar frames.Besides,2D sonar image is lack of recognition capability for multi-targets at the same horizontal angle of sonar,the same distance between target and sonar,but with different depth underwater within the vertical angle of sonar.In this way,the object with lower reflection energy will be easily covered by the object with higher reflection energy.This paper focuses on the forward looking sonar target detection and tracking technology that can go against the character of forward-looking sonar images.As regards underwater target detection.Firstly,underwater linear target detection is discussed which is mainly against disturbance.By associating the good spatial mapping of the Hough transform with the low-cost performance of threshold segmentation,a Hough transform combined threshold segmentation is proposed for underwater target detection.The linear relation of the Hough transform constraints the threshold evolvement.The effectiveness of the proposed method is proved by experiments,and the method is capable of underwater linear target detection.Secondly,in order to extract the moving object from sonar images,instead of conventional method,a target detection flow is proposed.Gaussian background reduction model is improved by spatio-temporal combined filtering.The proposed flow improves detection effect by combining morphologic processing,connecting region labeling,and trace difference based foreground building.A series of experiments on different targets is implemented and the results show the proposed flow has robustness.The output of detection benefits the judgment of human being or robot,which is dedicated to the following target tracking.As regards underwater target tracking,particle filtering that is based on Monte Carlo and recursive Bayesian estimation is the main clue in this paper.Multi-feature fused PSOPF algorithm,LBFPF algorithm,and anti-aliasing algorithm are proposed for single target tracking,double target tracking and anti-aliasing target tracking respectively.Firstly,it could not satisfy the real-time requirement if particle number was too big in the process of system statement prediction.However,robustness and precision might fall down if particle number was too small.At the same time,weight of most particles falls down after updating which might easily decrease multiformity if likelihood probability was at the end of prior probability distribution.Multi-feature fusion is used to construct observation model of particle filtering,and the discrete random estimate is used to approximate the correlation probability distribution.Considering the similarity of motion mechanism,the optimal solution,and the updating mechanism between PSO and PF,multi-feature fused PSOPF tracking algorithm is proposed which drives particles to high likelihood probability area by PSO optimal process.The proposed algorithm uses redundancy of multiple features,and optimizes the fitness value using adaptive weighted eigenvalue that is calculated according to each contribution.Experiments on different targets proof that the proposed algorithm can improve the using efficiency of particles,instead of wasting large amount of particles for accurate statement prediction.By this way,particle meagerness and dispersing can be solved,and tracking accuracy and robustness are high enough for underwater single target tracking.Secondly,due to the unsteadiness exist in forward-looking sonar images,not only the computational cost will be high,but also fuzziness of statement space exists in the region of superposition tracks in case of multiple target tracking.LBF model is inducted into particle filtering,and a LBF-based contour tracking algorithm is proposed.Level-set method transits curve evolvement into searching the solution of partial differential equation,which avoid parameterization and supervision on evolvement.At the same time,level-set method is insensitive to topology change of close curve which could contribute contour tracking better than conventional methods.A space prior is created using underwater target contour information which is used to restrict the boundary particles in re-sampling process.Two groups of experiments using different targets are set up and their results show that LBF-based contour tracking algorithm has better expression on target contour comparing with conventional contour tracking methods.The proposed algorithm has robustness on local information unsteadiness between frames and statement space fuzziness.Finally,as for aliasing,both of situations,looking forward and looking down,are analyzed.Centricity is not enough to describe target when echo signal was submerged.The Euclidean distance between targets is calculated and used to aliasing detection and aliasing accomplished detection when target is going into and going out of the aliasing area respectively.The proposed method maintains target statement estimate and transmit the tracking model by modeling the aliasing area.A new model for the obstacle will be automatically created when aliasing happened,and the tracking target turns to be the obstacle.At the same time,the created new model will be joined to the joint model,but the statement transition keeps the same.Tracking target turns to be the prior one after the aliasing factor judging that the aliasing is accomplished,and the unnecessary model will be ignored.Experiments of both looking forward situation and looking downward situation are designed,and the results show that compare with conventional methods,the proposed method can deal with side-lobe disturbance and aliasing felicitously and it has robustness in simple obstacle or simple underwater background.