Techniques of tracking multiple human targets using ultra-wideband radar are important for security applications. Tracking techniques that use the texture information of the received signal have been proposed. However, these methods require off-line processing or are limited in terms of the number of targets that can exist in the same range bin at one time. In this study, we propose a technique that does not have this limitation. First, we estimate the Doppler spectrum employing the Capon method, which is used in adaptive signal processing. We determine the target velocities by detecting peaks in the estimated Doppler spectrogram. We next track targets in the range-velocity plane using density-based spatial clustering because the targets should have different velocities or exist in different range bins. Finally, we track targets using the estimated clusters. In an experiment, the proposed method was used to successfully track three targets intersecting in almost the same range bin at the same time.