Identification And Tracking Method For Insulator Targets In Transmission Line Inspection Video

With the problems caused by the traditional patrol mode of artificial transmission lines becoming more and more prominent,the intelligent patrol mode of transmission lines using unmanned aerial vehicles has gradually become the mainstream.The research of insulator target recognition and tracking in patrol video is the key to realize the intelligent patrol inspection.At the present stage,the method of insulator target recognition and tracking can’t solve the problems caused by complex background and scale changes,and can’t meet the requirements of real-time,accuracy and robustness.This paper studies from the following two aspects: Firstly,according to the scene of insulator target recognition,insulator target is divided into moving insulator target and static insulator target.Insulator target recognition based on motion information is researched by combining traditional moving target detection method.To study static insulator target recognition,SVM and Faster-RCNN models in machine learning algorithm are used,and Hog feature and convolution feature with strong representation ability are introduced respectively.Secondly,generative Camshift algorithm,discriminant KCF algorithm and its improved algorithm are used to study insulator target tracking.The main work of this paper is summarized as follows:(1)The current methods of insulator target recognition and tracking are studied and analyzed,and to identify the insulator target based on the insulator moving information in the video.Experiments show that the traditional method of moving object recognition is not suitable for patrol video with background updates and can’t accurately identify the insulator target.(2)Several machine learning models commonly used in machine learning methods are analyzed.According to engineering requirements,insulator recognition models based on SVM and Faster-RCNN are designed and built respectively.The models are trained with samples collected from transmission line inspection.Finally,the trained models are used to achieve accurate insulator target recognition.(3)An insulator target tracking algorithm based on Camshift is adopted.The hue H component of insulator image is extracted.On this basis,the back projection is calculated to obtain the appearance model of insulator target and the algorithm is combined to achieve insulator target tracking.Finally,through experiments and comparative analysis,it is found that the algorithm only considers the insulator target itself,ignoring the background around positive sample and the local patterns between positive and negative samples,which lead to the insufficient robustness to complex factors in the practical application of insulator tracking.(4)A multi-scale insulator target tracking algorithm based on KCF is proposed.Firstly,insulator fHog features are extracted,then a classifier is trained by minimizing the decision function,and the locations of insulator targets are obtained.Finally,a multi-scale estimation method is introduced to determine the size of insulator targets,which improves the accuracy of insulator tracking.At the same time,aiming at the problem of multiple insulator targets in patrol video,the idea of extending single-target tracker to multi-target tracking is proposed.By encapsulating multiple single-target trackers,a multi-target tracker with multiple insulator targets and counting function is realized.