This paper presents a full research and development cycle of a vision-based system for long-term remote monitoring of civil engineering constructions. To measure displacements, the system analyzes reference mark position using a developed image processing algorithm. All computational operations are performed by FPGA to ensure high performance. The developed monitoring system has an accuracy of 0.45 mm (3 standard deviations) at the distance up to 50 m. The measurement error remains within this value even in harsh environments. The value of this paper is formed by the possibility of stable 24/7 monitoring in low-temperature conditions (up to C).