中文摘要：

针对变电站的开关设备，提出了一种基于霍夫森林（HF）的开关设备检测及状态识别方法。采用多种特征提取方法，利用HF构建分类器，通过霍夫变换框架对目标物体中心位置投票，并结合一种无需非极大值抑制的启发性算法，解决多个开关设备同时检测与定位问题，同时提高开关检测准确率；训练“分”和“合”两种开关模型，分别检测与定位开关设备，以此判别开关状态。实验结果表明，该方法识别准确率高，鲁棒性好，可应用于机器人巡检系统或者智能视频监控系统。

英文摘要：

Aiming at substation switches, a switch detection and state recognition algorithm based on improved Hough forests is presented. The multi-feature extraction method and the classifier formed by Hough forests are used to construct a Hough transform framework to vote for the central position of target. A heuristic algorithm without non maxima suppression is used to solve the problem of multi-switch localization and improve the accuracy of target detection and localization. The ＂off＂ and ＂on＂ switch models are trained to detect and recognize the switches, respectively. Experimental results show that the method with its high recognition accuracy and high robustness can be applied to robot inspection system or intelligent video surveillance system.