Power Line Extraction From Aerial Images Using Object-Based Markov Random Field With Anisotropic Weighted Penalty

Le Zhao,Meng Tian,Zini Jian,Hongtai Yao,Xianpei Wang

doi:10.1109/access.2019.2939025

Le Zhao, Meng Tian + Show 3 more

Open Access

https://doi.org/10.1109/access.2019.2939025

Copy DOI

Abstract

The extraction of power line plays a key role in power line inspection by Unmanned Aerial Vehicles (UAVs). While it is challenging to extract power lines in aerial images because of the weak targets and the complex background. In this paper, a novel power line extraction method is proposed. First of all, we create a line segment candidate pool which contains power line segments and large amount of other line segments. Secondly, we construct the irregular graph model with these line segments as nodes. Then a novel object-based Markov random field with anisotropic weighted penalty (OMRF-AWP) method is proposed. It defines a new neighborhood system based on the irregular graph model and builds a new potential function by considering the region angle information. With the OMRF-AWP method, we can distinguish between the power line segments and other line segments. Finally, an envelope-based piecewise fitting (EPF) method is proposed to fit the power lines. Experimental results show that the proposed method has good performance in multiple scenes with complex background.

Highlights

INTRODUCTIONBased on the irregular graph model and neighborhood system, we do not have to consider the pixels in background region because they are not used in classification process
Power line inspection is a basic and extremely important task for department of power grid [1], [2]
We proposed an object-based Markov random field (MRF) with anisotropic weighted penalty (OMRF-AWP) method to extract power line segments from line segment candidate pool

Summary

INTRODUCTION

Based on the irregular graph model and neighborhood system, we do not have to consider the pixels in background region because they are not used in classification process It greatly improves the accuracy of power line extraction. We assume that the L2 and L6 are divided into class 1 and the other line segment regions in the neighbor set of L1 are divided into class 2 In this condition, the local probability that L1 is classified into the class 2 is significantly higher than class 1 based on the original MLL model because the number of non-power line regions in the neighborhood set of L1 is higher than the number of power line regions.

GROUP AND CONNECT POWER LINES

EXPERIMENTAL RESULT

PARAMETERS DISCUSSION

CONCLUSION