To overcome the defects of traditional bridge health monitoring methods, visual technology is employed for non-contact bridge vibration monitoring. The method consists of video data acquisition, video data analysis and results visualization. And the viedo data analysis contains three steps. First, the stable pixels of a video are located. Then, a video thermal noise model is built via video data mining by tracking the variations of all stable pixels of the video. At last, the vibration frequency of each local position of the bridge in the video is calculated by the thermal noise model. The experimental results show that the method can effectively detect the imperceptible tiny vibrations in visible structure components of bridges, and it can measure the frequencies of vibrations in different local positions of bridges quantitatively and synchronously. Therefore, the method can provide a novel bridge structure health monitoring approach that is convenient, low-cost and universal.