Temperature-Induced Error Compensation in Computer Vision-Based Displacement Measurement Using Deep Learning-Based Time Series Forecasting Model

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Computer vision technology has emerged as a promising approach for multipoint displacement monitoring of civil infrastructure, owing to its inherent noncontact operation and remote measurement capabilities. However, its measurement accuracy is greatly affected by ambient temperature variations during long-term monitoring applications. Therefore, this study investigates temperature-induced displacement measurement errors in vision-based measurement techniques and proposes an error compensation method based on a time series forecasting model. First, a deep learning-based feature descriptor is employed and enhanced for image feature extraction and displacement extraction, followed by a quantitative analysis of the impact of temperature variations on vision-based displacement measurement errors. Second, the ModernTCN deep learning model is used to establish the nonlinear relationship between temperature differences and temperature-induced measurement errors. Finally, the established model is employed to predict and compensate for the temperature-induced measurement errors. The effectiveness of the proposed method is validated via indoor laboratory experiments, outdoor temperature tests, and field tests on a long-span bridge. The results reveal a nonlinear relationship between temperature differences and vision-based displacement measurement errors, and this relationship can be effectively mitigated using the proposed time series forecasting model. Experimental results indicate that correlation coefficients and cosine similarities obtained by the proposed method outperform conventional Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and state-of-the-art architectures (i.e., Transformer, Informer, PatchTST, TimesNet, and N-BEATS). In field tests, the proposed method achieved a Mean Absolute Error (MAE) of approximately 0.002 mm, representing an error reduction of over 90% compared to the Transformer model. The proposed method enables robust multipoint displacement monitoring with integrated temperature self-compensation, providing a reliable data foundation for safety evaluation and health monitoring of engineering structures.
Funding Information
  • National Natural Science Foundation of China (52308330)

This publication has 41 references indexed in Scilit: