Device Target Checking for Power Patrol Robot Based on Objectness Estimation

Abstract

In order to free people from daily patrol tasks in power substations, patrol robots are designed to check power equipment status and read various electric meters. According to the requirements of patrol tasks, a hierarchical, coarse-to-fine, fast device detection and recognition method is proposed. The coarse detection is based on objectness, using 8 × 8 Binarized Normed Gradients (BING) feature to generate proposals and filtering them using Support Vector Machine(SVM) trained by Local Binary Pattern(LBP) feature combining with histogram matching. The fine detection utilizes the color and contour feature of the basic elements of a certain device to obtain the status information or meter data. In this paper, instances of the device status checking and the pointer meter data reading are developed. The experiments validate the effectiveness, accuracy and real-time of this method.