Small Uav Target Detection And Tracking In Complex Weather Environment

Uav develops rapidly and is widely used.However,due to the lack of control means and the frequent occurrence of "black flight" incidents of small uav,the demand for anti-uav technology is increasingly urgent.Although airborne detection methods such as radar,passive and acoustic detection can obtain the location information of uav target,they cannot be effectively used in anti-small uav due to high cost.But the machine vision can detect and track the small uav target,through the camera real-time acquisition airspace video image,the acquisition of the video image corresponding processing to achieve the detection and tracking of the small uav.However,in the process of detection and tracking of anti-small uav system,the uav target size is small,the flight attitude is changeable,and the flight environment is complex.The video image quality collected is an important factor affecting the accuracy of visual target tracking.Therefore,it is particularly important to study the target detection and tracking technology of small uav in complex weather environment.Therefore,in order to improve the complex uav target detection and tracking accuracy of the weather environment,influence the quality of video acquisition of the largest fog and rain is captured by the two kinds of complex weather conditions of video image contrast and resolution are decreased,tracking target is easily lost,focuses on the two weather conditions of small uav target detection and tracking methods,the main research content is as follows:(1)Aiming at the problem that the image acquired by the dark channel prior defogging algorithm is easy to lose the clarity of the image at different depths of the scene,so that the target is difficult to be detected,an improved method for defogging haze is proposed based on the gaussian weighted fusion dark channel principle of different distant and close views.First based on dark channel prior gaussian weighting function,graph in different depth of field of dark channel gives different weights,using the boundary constraints on scene transmittance rough estimates,reuse context regularization on detail transmission rate to get more accurate transmission rate value,finally obtained the clarity and secure side degree good fog figure.(2)Aiming at the problem that the image information will be blocked and not clear in rainy weather,the image rain removal algorithm is studied.Based on the analysis and comparison of the principle and characteristics of the rain-removing algorithm,a more suitable rain-removing algorithm based on image stratification is selected.Firstly,the gaussian mixture model is used in the theory of image layering,then the image is constantly updated and the minimum error property is used,and finally the image after the rain is removed is obtained.(3)In view of the small size and changeable state of the moving target of uav,the video-based detection methods are mainly divided into two types: one is the uav detection based on single frame image features,and the other is the uav detection based on multi-frame motion information.Single frame image features detection mainly USES the gradient direction histogram(HOG)and support vector machine(SVM)method,but the HOG sliding window takes too long they cannot guarantee the real-time detection,thus combining primary Fourier descriptor(FD)simplifies the characteristics of the time,and the FD + HOG + improved SVM algorithm as a single frame detection algorithm and implement of unmanned aerial vehicle(uav).Based on motion information detection is mainly for unmanned aerial vehicle(uav)moving targets detection of frame difference method and the ViBe algorithm,combined with HOG + SVM algorithm for video successive frames of uav motion target detection test and experiment,the improved detection method has a better uav shape recognition ability,can carry on the valid inspection identification of unmanned aerial vehicle(uav),and the precision is high energy meet the demand of the accuracy of the system.(4)Aiming at the problem that Kernel correlation filtering(KCF)algorithm is fast to the moving target and difficult to track,an improved KCF algorithm based on kalman filter is proposed.By using the standard sample update mechanism,the image with high credibility at the previous moment is retained,and the tracking is resumed through the matching operation of feature points.Experimental results show that the improved KCF algorithm can track the moving target accurately,and can track the target even if the target size changes suddenly.Through the original pre-processing of the image in the environment of haze and rain and snow,and then the detection of the features of the small and medium-sized uav in the video image,and finally the tracking algorithm of the improved target sample update and feature point matching is used to track the uav,and the detection and tracking of the small uav target in the complex weather environment is completed.The results show that the proposed method can accurately detect and track small uav in complex environment,and has strong adaptability and robustness.