Research On Extended Target Detection Based On Sky Scene Perception

In recent years,with the rapid development of detection capability,measurement distance and artificial intelligence in photoelectric tracking system,which plays an increasingly important role in the military mission,such as spacecraft launch,weapon test,flying target monitoring.Target detection is one of the key technologies of target tracking system in the range,its accuracy determines the overall performance of the tracking system and directly affects the quality of measured target data.However,target imaging presents unclear and low contrast on overcast or fog weather conditions,resulting in target edge details unrecognized.In addition,clouds are mistakenly detected as flying targets or foreground objects are blocked by mass clouds under cloud condition.Therefore,target detection performance greatly depends on the weather condition.To solve the difficulty of target detection in complex dynamic sky scene,this paper proposes an extended target precision detection approach based on sky scene perception,intelligently selecting the applicable algorithm for target detection according to the specific sky background types.The method mainly includes two research focuses on scene perception and target precision detection,as follows:(1)Scene perception is the key and innovation of this subject,which is to identify background weather types based on image visual feature information,and it can intelligently connect image multi-mode technology and accurate detection for moving target.To break through the existing approach which usually regard the target as interference,a sky background perception method for dynamic scene is proposed based on multiple kernel learning,making use of the fusion features of foreground and background.More precisely,firstly,due to foreground is the flying target as rigid body which has structural characteristics and corner properties,but background is the sky image including clouds or pure sky which has only irregular natural texture features.Hence,target regions and background regions for dynamic sky scene can be determined by corner detection algorithm.Then,texture and power spectrum slope feature are extracted from the background regions,as well as contrast and sharpness features of the foreground regions.Finally,the fusion features of target and background are used to identify dynamic sky scene conditions by multiple kernel learning.The results show that target properties are salient features to recognize sky background conditions,especially it can obviously improve the accuracies when background regions are uniform or have similar features.Moreover,the proposed method can improve the performance and enhance the adaptability of sky scene condition recognition.(2)Target precision detection is to extract target outline and position information from complex background area.In this paper,image multi-mode detection technology is selected to adapt to different sky scene.The combination method of edge detection and mathematical morphology is used to obtain target outline in sunny scene.The approach based on image enhanced technique is taken to accurate detection extended target in overcast scene.The method based on image defogging technique is chose to obtain target outline in fog scene.The target moving state is automatically judged by the number of corner points in cloud background.The method based on foreground region is used to accurately detect the common flying target in cloud scene,and the template matching approach is selected to extract target outline when the flying target is partially obscured.What's more,the method based on adjacent multi-frame position prediction is chosen to obtain target precise outline when the flying target is overall obscured.The results show that the proposed method can extract the complete contour and position information of flying target in complex dynamic sky scene.In this paper,the proposed approach overcomes the shortcomings of single target detection algorithm which is only applicable to local scene,it also has the advantages of high accuracies and strong adaptabilities.Therefore,target detection for complex sky scenes is of great significance to achieve the precise recognition and stable tracking,and it has a solid theoretical foundation to realize high precision and intelligent for photoelectric tracking system.