Vibration analysis techniques, such as modal analysis, are widely used to assess the dynamic properties of mechanical components and systems. Traditional methods rely on sensors (e.g., accelerometers) to measure vibrations; however, these sensors can alter the mass and damping properties of the system and require multiple tests to obtain spatially dense information. Cameras offer an alternative for vibration analysis, providing the usual advantages of contactless measurement, such as adding no mass or damping and offering high spatial resolution. Using a stereo camera setup enables 3D displacement measurements, which require accurate point matching between the cameras and consistent point tracking across video frames. Despite the potential of camera-based measurements, limitations in frame rate present a challenge. In this study, a previously developed random sampling framework is employed to capture high-frequency vibration signals at a low equivalent frame rate. To improve out-of-plane measurement accuracy, a large stereo angle (38◦ in this experiment) between cameras is preferred, although this angle introduces challenges in point matching. This study presents an approach for matching dense point pairs through image registration with projective transformation, based on initial point matches. Point tracking is accomplished using the Lucas-Kanade optical flow algorithm, a simplified approach to digital image correlation. Experimental testing on a car door, at an equivalent frame rate below 130 frames per second, identified 10 vibration modes up to 139 Hz using hammer excitations (i.e., approximately twice the Nyquist limit). Several excitation locations were tested, revealing that the excitation at the door corner should be avoided, as it affects wave transmission and deflection shape measurements.