Structural health monitoring is increasingly attracting research interest, especially in view of all the societal issues linked to the ageing of existing civil structures and infrastructures. By handling datasets collected through a network of sensors deployed over the monitored structures, (big) data analytics can be carried out. Traditional inertial sensors, like accelerometers or strain gauges, require complex cable arrangements and also display high maintenance costs. Recently, there has been a growing interest in non-contact, vision-based methods to address the aforementioned problems, still with a noteworthy capability to assess in real time, or close to it, the structural health. Among such methods, Digital Image Correlation (DIC) can provide a map of tracked displacements at various points on a structure, especially if physically-attached targets are exploited by the tracking algorithm. In this study, a video of a vibrating structure is considered, to focus on markers placed at specific points like e.g. structural nodes where damage can be initiated, or whose response turns out to be affected by the said damage to be sensed. Displacement time histories are obtained, and a blind source identification technique is adopted to dig into the data and assess the structural health. More specifically, the proposed methodology is shown to accurately extract the vibration frequencies and the mode shapes of the structure, even when they change in time due to damage inception or growth.