为了提升电力巡检机器人对输电线路障碍物识别的精度,对电力巡检障碍物定位识别模型进行了研究,并提出了一种结合半全局匹配(Semi-Global Matching,SGM)算法和单点多盒探测器(Single Shot Multibox Detector,SSD)的模型.采用加权最小二乘(Weighted Least Squares,WLS)滤波算法,对利用SGM算法所得视差图中的空洞和噪点进行了处理,并提出了一种轻量化单点多盒探测方法.实验测试结果显示,优化后的模型定位测量误差低于5.13％;其大小由300 MB下降到了2.03 MB;轻量化模型在89、250次迭代后的损失函数值分别降至0.23和0.33.验证了所提出的模型能够实现输电线路障碍物定位和识别,能够满足实际的应用需求.
Obstacle Positioning Model for Power Inspection Robots Based on SGM-SSD
To improve the accuracy of power inspection robots in identifying obstacles on transmission lines,the positioning and identification model of power inspection is studied,and a model combining semi-global matc-hing(SGM)algorithm and single shot multibox detector(SSD)is proposed.The weighted least squares(WLS)filtering algorithm is used to process the holes and noise in the disparity map obtained by SGM algorithm,and a lightweight SSD method is proposed.Experimental test results show that the optimized model has a positioning measurement error of less than 5.13％,its size has decreased from 300 MB to 2.03 MB,and the loss function values of the lightweight model decreased to 0.23 and 0.33 after 89 and 250 iterations,respectively.There-fore,the proposed model can achieve the positioning and identification of obstacles on transmission lines,which can meet practical application needs.
国家科技期刊平台
NSTL
万方数据
