Fast Object Detection for Continuous Images in Line Inspection

Authors

  • Xiaoliang Zhang Lvliang power supply company of State Grid Shanxi Electric Power Company., Lvliang, Shanxi, 033000, People’s Republic of China
  • Jianjun Cheng Lvliang power supply company of State Grid Shanxi Electric Power Company., Lvliang, Shanxi, 033000, People’s Republic of China
  • Xuefeng Bai Lvliang power supply company of State Grid Shanxi Electric Power Company., Lvliang, Shanxi, 033000, People’s Republic of China
  • Hanyu Zhang Lvliang power supply company of State Grid Shanxi Electric Power Company., Lvliang, Shanxi, 033000, People’s Republic of China

Abstract


It is important to guarantee the stable and reliable operation of a power system to ensure power supply safety. The rapid expansion of emerging power systems has brought significant challenges relating to power line inspection, especially under hazardous conditions. The existing vision-based line inspection approach emerges as one promising solution. However, the required computation is prohibitive as it requires a convolutional neural network (CNN) inference for each image frame. In this work, we address this problem by investigating block matching and extrapolation algorithms. These two algorithms exploit the motion information in consecutive frames of real-time videos, thus avoiding the expensive CNN inference for every
image frame. According to the experiment evaluation, the processing rate is drastically increased by introducing a very limited amount of computation that leverages the temporal pixel motion. Moreover, the precision loss is negligible when the window size is small while the rate of improvement is significant.

Keywords: Power line inspection, continuous vision, block matching, neural network

Cite As

X. Zhang, J. Cheng, X. Bai, H. Zhang, "Fast Object Detection for Continuous Images in Line Inspection",
vol. 30 no. 4, pp. 279-284, 2022.

Published

2022-07-01