Research On Visual Image Fusion Technology For Unmanned Ship Navigating Obstacles

In the future,unmanned ships will be an important tool for China to defend its territorial sea sovereignty and safeguard maritime safety.Meanwhile,unmanned ships will also play an increasingly important role in areas such as overwater monitoring,commercial aquaculture,scientific investigation,and overwater equipment monitoring.Visual perception is fast,intuitive and informative,and has gradually become an irreplaceable means of perception for unmanned ships.It plays an irreplaceable role in the collision avoidance of unmanned vessels,the operation and maintenance of Marine structures,and the inspection of target objects.At present,the research on the visual perception technology of unmanned ships mainly focuses on the identification of obstacles on the overwater,while there is little research on the identification of obstacles on the overwater,especially those with both surface and underwater structures.This kind of obstacle has both the overwater part and the underwater part.Due to the refractive influence caused by the different overwater and air media,it is often necessary to arrange the overwater camera and the underwater camera to obtain the overall view of the obstacle on the overwater and underwater,and the information of the two is stitching and fusion.However,there are few reports on the research work in this field.This study carries out the research on the fusion technology of visual images of unmanned ship navigation obstacles.According to the characteristics of obstacles,algorithms and experiments such as preprocessing,edge detection,splicing and fusion are carried out on the abovewater and underwater images,so as to reconstruct the overall situation of the above and underwater obstacles.In this way,the all-round perception performance of the unmanned vessel can be further improved,and the autonomous decision-making ability and intelligent level of the unmanned vessel can be improved.The relevant results can also be applied to the offshore operation and maintenance of the unmanned vessel,target recognition,target detection and other application scenarios.The specific work of the paper includes:(1)Based on the different characteristics of the overwater and underwater image of the unmanned ship navigating obstacles,the mosaic and fusion characteristics of the overwater and underwater image of the obstacles are analyzed,and on this basis,a visual image fusion system for the navigating obstacles of the unmanned ship is carried out.The software and hardware framework is designed and built,and the key sensors are selected and jointly calibrated.(2)For overwater image defogging,based on the traditional dark channel prior technology architecture,the estimation of transmittance is optimized.Through the rapid estimation of atmospheric light value,the time consumption of the algorithm is reduced and the quality of defogging is improved.Aiming at the traditional dark channel prior technology,RGHS and adaptive parameters are introduced to carry out RGHS underwater image sharpening processing based on adaptive parameters.Symmetric features were extracted from the actual images of obstacles and their reflections,and the symmetry axis was detected by using the multi-example learning method.(3)After image preprocessing,the actual image of obstacles is segmented based on automatic multi-seed points.An improved Canny operator edge detection algorithm was proposed.Based on the calculation of gradient intensity statistics,the adaptive threshold value was realized,the image preprocessing method was changed,and the histogram of the Mosaic area was normalized.Finally,the experimental results before and after improvement were evaluated.(4)An extended neighborhood based obstacle Mosaic model is constructed,the features in the obstacle images are defined and expressed,and the normalized difference calculation of coding points is carried out for classification.By obtaining the source code and sub-code of the obstacle,the matching calculation and stitching of the coding points of the image on the edge of the obstacle are realized,and the effectiveness of the algorithm is evaluated.The attention module is introduced,and the Frustum-Pointnet fusion detection algorithm based on image drive is used to achieve the fusion of obstacle image and point cloud,so the comprehensive information of obstacle collision avoidance is obtained.