| The unmanned surface vessel(USV)is a special high-agent.The visual system is one of the key equipment to realize independent operation,automatic obstacle avoidance,scene surveillance,target detection,target tracking and object recognition,et al.However,the marine environment is very complexity and often accompanied with fog and other inclement weather.The scattering of atmosphere particles under sea fog will cause the video image collected by USV vision system degradation and fuzzy,and it will seriously affect the independent operation,target detection,tracking and recognition of USV.In order to solve this problem and improve maritime adaptability of USV,the scene has fog or not is needed to be classified and then the video sea fog of video is needed to be removed.In this paper,firstly,a classification algorithm is proposed to distinguish the scene whether has fog or not.Secondly,a fast defogging algorithm for single sea fog image is proposed.Thirdly,we present a novel video defogging algorithm for USV.The experiments show that our algorithms are effective and useful for improving the adaptability of USV.Firstly,in order to improve the intelligent defogging ability of the USV visual system,a sea fog image classification method is proposed.On the basis of a surface image sample library which include a lot of fog images and clear images is established,8 features of which have significant difference between fog image and clarity image are extracted,including image information entropy,improved mean,standard deviation,average gradient,contrast,visual contrast,visibility and intensity of dark channel image.Then the BP neural network and SVM classification method with different combinations of above 8 features are used for classification of surface fog and clarity image.Experimental results show that the proposed algorithm has good performance on the classification of whether the image has fog or not.And when BP neural network is used for fog image classification,we just need to extract 3features of improved mean,visibility and image intensity,then the average recognition rate can reach 98.75%.Secondly,a single sea image defogging algorithm based on physical model is proposed.We first illustrate the image mechanism of atmospheric physics model under the sea fog in detail,and then a defogging algorithm based on dark channel theory is presented.The dark channel theory,estimation of the air value and optimization of the transmission image are introduced in detail.At last,the defects and shortcomings of the dark channel defogging algorithm are analyzed.Thirdly,an improved single sea image defogging algorithm is proposed.For sea fog image scene,we proposed an improvement algorithm to estimate the value of air.For optimization of the transmission image,an improved optimization algorithm by using the guided image filtering is presented.With compared through other effect image defogging algorithms and our algorithm,from two aspects of subjective and objective for processing results of image quality evaluation.The simulation experimental results show that our algorithm can realize fast single image sea fog defogging,the effect is obvious.Lastly,for sea fog video defogging of USV,in order to improve the defogging efficiency,combined with the method of background frame difference,a video image defogging algorithm of based on fog mask theory and guided image filtering is proposed.The scene whether have big change is determined by using the method of background frame difference.For the background of consecutive video frames has little change,firstly,the method of our improved single image defogging algorithm is performed on the first frame,and then the fog mask of the current background is obtained.The fog mask of current fog frame can be obtained by guided image filtering.In this way,the fast video image defogging is realized.Simulation experiments show that our video defogging algorithm has high efficiency.And the phenomenon of target “smear” is not appearing in our defogging results,the defogging effect is obvious. |
PDF Full Text Request