Open Collections

Autoencoder-based key-frame identification for water pipeline CCTV inspection

University of British Columbia

Closed-circuit television (CCTV) is being widely adopted in water pipeline inspection. The inspector needs to spend a long time to watch the recorded video during the office-based survey and can get fatigue easily. An automated process can release the inspector's workload and ensure the consistent quality of the survey. However, a fully automated survey of varied structural discontinuities still remains as a challenge. This study aims to first identify the key frames of the CCTV video, which contain the major anomalies captured from the internal surface of the pipe. Thus, the inspector can focus more on these key frames. First, the key-frame identification task is performed by using an unsupervised learning-based framework with state-of-the-art autoencoders. Three popular autoencoders were implemented in this framework, and results were collected accordingly. The experimental results illustrate that this framework achieves up to 0.940 accuracy with a relatively lower precision metric of 0.901. Secondly, a key-frame identification framework based on the steerable pyramid autoencoder (SPAE) is proposed in order to improve the accuracy of this task. Both the parameter optimization and comparative studies for the proposed SPAE were carried out in this research. Explicitly, the SPAE develops good capability of representation learning and extraction. The experimental results demonstrate that the SPAE-based approach can achieve 0.984 accuracy, which outperforms the other experimental methods selected for comparison. Thirdly, a log-Gabor autoencoder (LGAE) based framework is proposed to further enhance the performance on key-frame identification in terms of representation learning. It can be regarded as an improved version of the SPAE-based approach. Comparative studies were conducted, and the results show that this novel LGAE takes advantage of the feature detection over the other methods with the accuracy of 0.988 and the recall metric of 0.996. Additionally, the LGAE has a superior performance for the generalization on varied datasets. Hence, LGAE-based framework will significantly assist the office-based survey, which will highly facilitate the pipeline condition assessment through the CCTV inspection.

Text

2021-02-24

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/77378

Electrical Engineering

University of British Columbia

UBCO

http://creativecommons.org/licenses/by-nc-nd/4.0/