Fast recognition of human climbing fences in transformer substations
Tianzheng Wang, Kangning Wang, Jie Li, Hua Dong Yu, Shuai Wang, Jiang Bian, Xiaoguang Zhao · 2017
Recognition of human climbing fences in trans-former substations is very essential in a power substation. This paper proposed an innovative and practical human climbing fences detection method based on image processing. At first, the Gaussian Mixture Model background modelling algorithm is exploited to detect motion objects under a view of fix surveillant camera in a power substation. After obtaining the motion regions of interest, the Histogram of Oriented Gradient (HOG) feature is extracted to describe inner human. And then, based on the result of HOG feature extraction, the Support Vector Machine (SVM) is trained to classify pedestrians. Next, an improved Hough Transform is implemented to detect fences. Finally a Sparse Optical Flow method is applied to track the motion of human. Compelling experimental results demonstrated the correctness and effectiveness of our proposed method.