Abstarct: Techniques of health monitoring and assessment of engineering structures by vibrational data are one of the most common methods for identifying structural damages. The basic accepted idea for damage detection baxsed on structural dynamic is variations in physical properties such as mass, damping, and stiffness due to damages that cause the modal properties (natural frequencies, modal ratios and mode shapes) would be detectable. Thus, changes in the modal parameters of a structure implies the changes of structural health condition, i.e., damage. So, experimental measurement and determination of dynamic behaviour of structure are very important. The central technique to describe the dynamic of structures is modal analysis, in which modal parameters controlling dynamic behaviour of the structure are identified. Digital video cameras are one of the measuring systems that are welcomed in this regard. Despite numerous advantages, modal analyses done baxsed on video measurements have created some challenges due to much dependence on high-contrast markers. On the other hand, wide acceptance of these techniques reveals the necessity of developing methods that utilize only the video measurements without additional structural surface preparation. Hence, providing an operational modal analysis algorithm is the goal of this dissertation. This algorithm doesn't require extra preparation of surface structures by only measured full-field responses taken from high-speed video camera. In order to achieve this goal, the phase of pixels of each frxame of video are extracted at different spatial scales by a multi-scale decomposition technique because the local phase indicates the local vibration of structure. baxsed on extracted phase for each pixel of each frxame, the full-field spatial-time matrix of motion can be formed in different spatial scales. Given that the calculated response has a much higher spatial dimension than the modal dimension of the structure, it is required to use dimension reduction technique. The output of dimension reduction technique is the linear combinations of the modal responses in which the number of these mixtures is as same as the number of active modes of structure. So, using output-only algorithms the modal responses can be separated baxsed on the observed mixtures. In the last step, the frequencies and damping ratios for each mode of structure are calculated. Also, by applying the magnification technique to each of these responses and performing the reverse process of video processing, the modes of the structure with high spatial resolution can be reconstructed. For verification, the free and random vibration video of two simple and cantilever beams with given modal parameters was reconstructed in MATLAB software environment. Then the performance of the proposed method was verified by applying the proposed algorithm to the prepared videos and comparing its output with the theoretical values. In addition, in order to evaluate the performance of the algorithm in experimental condition, a free and random vibration videos of an aluminum beam made in a laboratory environment was investigated using a high-speed camera. Finally, using the developed algorithm, the modal parameters of the aluminum structure were successfully identified.