In response to challenges such as illumination interference, motion blur, and target deformation in structural displacement detection under real engineering environments, this paper proposes a comprehensive technical solution integrating "hardware filtering – software recognition – super-resolution enhancement – sub-pixel localization." The method suppresses ambient stray light at the source by combining an IR850 infrared filter with a 940 nm target. Additionally, the YOLOSR algorithm is employed to significantly enhance adaptability to projective distortions of elliptical targets caused by vibration and viewpoint variations. Through comparisons with mainstream object detection models, the superiority of the YOLOSR cascaded framework in both detection accuracy and efficiency is validated, ensuring algorithmic reliability for the system. Experimental validation, including calibration tests, translation stage verification, vibration tests, and on-site photovoltaic panel evaluations, demonstrates that the proposed method achieves high detection accuracy and robustness under various working conditions. The system requires only an industrial camera, an infrared filter, and an infrared target, eliminating the need for contact with the measured structure. This significantly reduces equipment costs and maintenance complexity, making it suitable for long-term health monitoring of large-scale engineering structures.