基于SGM-SSD的电力巡检机器人障碍物定位模型

Obstacle Positioning Model for Power Inspection Robots Based on SGM-SSD

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。

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.