Research On Insulator Tracking Method In Aerial Video Based On Recurrent Neural Network

In recent years,the economy of our country has been developing rapidly and all walks of life have a great demand for electricity.The insulator is an important part of the transmission line,which uses to support the wire and the earth to keep the insulation.The traditional insulator inspection is mainly done by hand,vulnerable to the terrain environment,inefficient and costly.In recent years,computer vision technology and unmanned aerial vehicles(UAV)technology made a major breakthrough,making the gradual replacement of manual inspection UAV inspection is the future development trend of intelligent inspection.UAV inspection is the main purpose of the transmission line,especially the insulator off,cracking,aging,fouling and other state implementation of efficient monitoring,which rely on the accurate tracking of insulators and rapid clear imaging.In this thesis,the insulators in aerial power lines are taken as the research object,and the method of insulator tracking in aerial video is designed.The main work of this thesis includes the following three aspects:(1)Aiming at the problem that the gray feature of the space-time context tracking algorithm can not accomplish the aerial image feature representation very well,a tracking method based on Local Binary Pattern(LBP)and spatio-temporal context is proposed.In order to solve the problem of large-scale insulator changes,a scale updating strategy is implemented.The scale updating strategy relies on weighting the image scale of each frame.The experimental results of CarScale and Lemming public datasets show that the proposed algorithm is better than the original algorithm in adapting to the change of target scale.Experimental results in the insulator dataset show that this thesis can deal with the tracking offset caused by large-scale insulator changes.(2)According to Sequentially Training Convolutional Networks for Visual Tracking(STCT)method,two different layers of Convolutional Neural Network(CNN)were used to locate the object extraction features and object respectively,and the structure was constructed based on the low-level artificial features.The scale prediction network adapts to the target's scale.The STCT method is introduced into the tracking of the insulator video,and in the Linux system platform and Caffe framework to complete the configuration and debugging,achieving good results,indicating that the tracking method can accurately track the insulators to meet the real-time requirements of electrical engineering applications.(3)For the recurrent neural network(RNN),it has a good application effect on image sequence.Based on Structure-Aware Network for Visual Tracking(SANet)method,this thesis uses RNN to model the target and combines the RNN feature of the target object with the CNN feature to improve the tracking discrimination of the device.The SANet tracking method is introduced into the tracking of the insulator video,and the configuration and compilation are completed on the Linux system platform and MatConvNet.The good results show that the tracking method can accurately tracking the insulator and solve the problem of insulator scale variation.