摘要
以全部目标跟踪为计算量,无法支撑变电站实时数据变化,巡检机器人障碍检测结果存在偏差,只能预估出障碍的大体位置。该文采用双目深度估计的基本思路,生成变电站巡检机器人检测策略,匹配变电站的实时运行数据;按照决策边界定义数据最大处理边缘,划分巡检机器人行进路线中障碍特征;基于特征深度融合技术构建检测模型,对变电站巡检机器人的障碍目标进行定位;通过卡尔曼滤波计算障碍目标中心,完成巡检机器人的障碍自动化检测。以变电站内较狭窄空间为测试环境,在巡检机器人行进路线中,连续设定8个障碍点,进行测试,结果显示,该方法可以检测出全部障碍点,且位置参数匹配成功,横轴与纵轴位置误差在1 mm左右,但两组传统算法会存在障碍点未被检测到的情况,且位置匹配误差超过10 mm。
Taking all target tracking as the calculation quantity,it can not support the real-time data change of substation,and the obstacle detection results of inspection robot have deviation,so it can only predict the general position of the obstacle. The automatic obstacle detection algorithm of substation inspection robot based on feature depth fusion is studied. Using the basic idea of binocular depth estimation,the detection strategy of substation inspection robot is generated to match the real-time operation data of substation. The maximum processing edge of data is defined according to the decision boundary,and the obstacle features in the route of the inspection robot are divided.Based on the feature depth fusion technology,the detection model is constructed to locate the obstacle target of the substation inspection robot. The obstacle target center is calculated by Kalman filter,and the obstacle automatic detection of the inspection robot is completed. Taking the narrow space in the substation as the test environment,8 obstacle points are set continuously in the travel route of the inspection robot for testing. The results show that this method can detect all obstacle points,and the position parameters are matched successfully. The position error of the horizontal axis and the vertical axis is about 1 mm. However,the obstacle points are not detected in the two groups of traditional algorithms,and the position matching error exceeds 10 mm.
作者
朱合桥
ZHU He-qiao(Zhongwei Power Supply Company of State Grid Ningxia Electric Power Co.,Ltd.,Yinchuan 750000,China)
出处
《自动化与仪表》
2023年第1期43-47,共5页
Automation & Instrumentation
关键词
变电站
特征深度融合
巡检机器人
障碍检测
自动化
检测算法
substation
feature deep fusion
inspection robot
obstacle detection
automation
detection algorithm
