A lightweight binocular vision‐supported framework for 3D structural dynamic response monitoring

Current three-dimensional (3D) displacement measurement algorithms exhibit practical limitations, such as computational inefficiency, redundant point cloud data storage, reliance on preset targets, and restrictions to unidirectional measurements. This research aims to address computation efficiency and accuracy issues in binocular camera-based 3D structural displacement measurement by proposing a lightweight binocular vision-supported framework for structural 3D dynamic response monitoring. Through the optimization of sub-algorithms and code structures, this framework enhances both measurement accuracy and computational efficiency. The research incorporates a hybrid feature point processing algorithm and a lightweight tracking algorithm, which improve the accuracy of feature point recognition and tracking, enhance the adaptability and flexibility of the monitoring process, and increase tracking efficiency and overall system performance. These improvements make the framework more applicable to various civil engineering scenarios. Experimental validation on a full-scale three-story structure shows that the framework enables effective, target-free, 3D dynamic monitoring. Compared with reference displacement sensors, the framework achieves a relative root mean squared error of 14.6%, closely matching the accuracy of traditional methods that utilize accelerometers. The framework processes 1000 frames at 9.2 frames per second, offering a novel solution for contactless structural dynamic response monitoring in civil engineering applications, such as residential buildings and bridges, within a reasonable distance.