The structural safety of highway bridges plays a crucial role in traffic safety operations and regional economic development.Due to the rapid increase and complex changes in traffic loads，the demand for real-time monitoring of bridge ’s structural deformation and safety conditions has been increasingly prominent.However，the structural health monitoring system，dominated by contact sensors，has limitations such as difficult installation and construction，complicated maintenance and replacement，and the need to interrupt traffic.To this end，a non-contact deformation measurement system based on machine vision was proposed.Template matching and feature point recognition methods were utilized to achieve recognition of structural surface markings and displacement extraction of key positions.The accuracy and stability of visual methods were tested through laboratory and on-site experiments.In addition，a corresponding finite element model was built for the actual operation status of the bridge structure，and the multi-level warning threshold was set through the dynamic and static mechanical experiments to further determine the safety status of the structure.The experimental results show as follows.① The displacement measurement error of the established visual deformation measurement system is within 5%，and the error of vibration frequency is within 1%；② Measurement techniques based on template matching and feature point recognition can recognize the inherent features of structures，and meet the needs of long-term deformation monitoring；③ Through finite element analysis under multiple working conditions，warning thresholds can be set for long-term displacement monitoring of structures.