This paper aims to experimentally evaluate a simplified vision-based method for structural health monitoring (SHM). Contrary to conventional solutions that rely on extracting motions through image processing, this paper proposes to conduct the SHM analysis by the direct processing of pixel brightness without extracting the motion signals beforehand. After some pre-processing steps, it is shown that the brightness data reveal essential information about the dynamic characteristics of the monitored vibrating structure. Furthermore, the low-level information of the pixel is compensated by an efficient selection of the so called "active pixels" throughout the image time flow. Finally, a subspace system identification-based method is applied to the brightness data, so that the modal parameters with uncertainty bounds are estimated within a large range of model orders displayed in a stabilization diagram. The experiment database consists of image time flows of a cantilever beam excited by a shake table driven by a band limited random noise. Modal vibrations range from 1 to 173 Hz.