A Novel Video Object Segmentation Method And Its Application On Intelligent Person Surveillance For Transformer Substation

Intelligent maintenance and inspection of substations are urgent needs for the development of smart grids,and are also effective ways to reduce labor costs and improve inspection and maintenance efficiency.Considering the engineering needs of substations,this article solves the problem of substation person localization and tracking based on the video object segmentation algorithm and conducts an in-depth analysis from the theoretical research and engineering application levels.A new video object segmentation algorithm is proposed,and has been put into practical use in the intelligent person-monitoring system of substation.The contribution of this paper could be summarized as follows:(1)Consider that substation person localization and tracking tasks raises requirements for both accuracy and speed,starting from theoretical research,a new video object segmentation algorithm--"State-Aware Tracker" is proposed.The algorithm innovatively proposes an adaptive adjustment mechanism based on state evaluation and feedback.It adaptively adjusts the tracking and segmentation strategy according to the specific state of the current frame,and at the same time uses the information of precedent frames to dynamically enhance the model.State-Aware Tracker can track targets of unlimited categories in fast speed and obtain accurate position masks.It shows state-of-the-art performance for accuracy and speed on several open-source datasets.(2)Considering the problem that the algorithm is not robust to persons large-scale actions,State-Aware Tracker was enhanced by using the key information of the human body and the prior information of the joint structure of the human body.This paper proposes a keypoint-guided human segmentation algorithm.Through attention guidance and feature fusion,the keypoint information is used to improve the performance of the model,and more accurate masks are obtained while the precise positioning information of human keypoints is obtained.(3)In terms of engineering application,according to the actual situation like the distance of monitoring cameras in the substation,low frame rate,and occlusion due to dense equipment,the algorithm was adjusted to make it meet the actual needs of substation video surveillance.In addition,based on State-Aware Tracker and keypointguided human segmentation algorithm,the basic framework of the substation monitoring system is established.With the help of the results of tracking and segmentation and human keypoint information,a series of tasks in substation monitoring could be solved,such as intrusion detection,illegal path detection,dangerous action alarm,illegal wearing alarm,etc.