The Substation Switch Detection And State Recogniton Based On Hough Forests

With the scale of China’s electric power network continues to expand, it’s essential that the substations operate securely and reliably, and the operation state of switches is one of the key factors that determine its stable and secure operation. With the development of video monitor technology, unmanned substation has become the development trend of the power grid dispatching automation. However, the existing video surveillance system is limited to monitor the substation, lack of automatic recognition of the state of switches. Therefore, with the aid of remote video surveillance system, we can realize the automatic recognition of the state of switches, which can contribute to realize the unmanned substation.This paper mainly research on the problem of substation switch state recognition, proposing a switch state recognition algorithm based on Hough Forests. Firstly, make use of Random Forests to learn the "off’ and "on" switch models, which generate the corresponding voting codebook. Secondly, utilize the voting codebooks to vote for the center position of the switch, completing the detection and localization of the switch. Finally, recognize the state of the switch according to the detection results of the two models.Due to Random Forests can fuse a variety of characteristics as the discrimant information, This paper make a research on the edge feature, gradient feature and the color feature of the switch. As a result, we use different methods to extract the features of the switch, which make up for the flaws of the single feature description and improve the accuracy of the switch detection.The target detection based on the generalized hough transform need to use nonmaxima suppression to locate and distinguish peaks in Hough images. But the threshold setting of the nonmaximum suppression easily lead to the target undetected or false detected. So a consistent probabilistic modeld has been constructed to explain the relationship between the voting elements and the hypotheses. It can reduce the false voting and avoid the step of nonmaxima suppression which solves the problem of multi-switch detection and localization.The experimental results show that the algorithm possesses strong stability, high robustness and strong versatility. Even if the switch is partial occluded, the algorithm can also ensure the accuracy of switch detection and localization. So it can be applied to robot inspection system or substation intelligent video surveillance system.