Research On Detection Algorithm Of Aerial Rotor Uav Based On Region Of Interest

In recent years,with the development of UAV maneuverability and target detection technology,rotary-wing UAV has been widely used in civil,security,police and military fields.Because the rotorwing UAV is not easy to be discovered in the execution of its mission,it is easy to be used by criminals to harm the society,steal state secrets and affect social security.In the paper,the real-time detection method of aerial rotor UAV is studied,and how to extract the region of interest of rotor UAV in suspended state and cruising state from video data is mainly studied.The main work of the paper is as follows:1.Such as the video target detection algorithms including inter-frame difference method,Gaussian background model and optical flow method an so on,the image detection algorithm including sliding window detection,image segmentation combined with classifier and so on,and the feature extraction algorithm such as HOG,LBP et al and the classifier model of SVM,Random forest had already be experimented,and the experimental results from the aspects of detection time,the detection results and the requirement of computer hardware are compared with each other,and the system scheme of detecting rotor UAV is obtained.2.The flying characteristics of the rotor UAV in motion are summarized.Aiming at the limitation of the traditional algorithm for extracting the region of interest in the scenario that the two moving states of the UAV are suspended and cruising is also discussed,the algorithm of extracting region of interest is studied and compared by experiments,and the conclusion is made.3.Given the distance from the camera to the non-cooperative UAV,the inconsistency of the angle and the position presented in the camera,the complex background and the fact that it may be in motion or hovering,it is easy to make rotor-wing UAV in the video deform slightly in the sizeand shape and other aspects.Using classifier classification method which can extract the robust features of the rotor-wing UAV could avoid the situation that the detection results are affected by the slight deformation of the UAV.4.In order to solve the problem of extracting the region of interest of non-cooperative rotor UAV in both suspended and cruising motion,the method combining inter-frame difference and improved selective search algorithm is adopted to segment the region of interest.Firstly,the region of interest of gray-scale video is segmented by morphological-processed inter-frame differential image combined with connected area marking,and then the region of interest of frame image is segmented by improved selective search method.Finally,the regions of interest extracted by the two methods are combined to form the regions of interest Set.5.In the detection algorithm of the paper,firstly the image is processed with median filtering and adaptive histogram equalization,then the positive and negative sample sets are described by the methods of HOG-FLD features,and then the two-classification model based on SVM is trained,which can recognize the rotary-wing UAV.Finally,the acquired region of interest set is sent to the two-classification model for feature matching.The experimental results show that rotor UAV with static or dynamic background video could be detected quickly and accurately by the proposed detection algorithm that could meet the requirement of real-time and accuracy rate in the evaluation index of target detection algorithm.