Marker pixel detection is a low-level task used by video-based motion capture systems for motion analysis. This is a highly time-consuming task because there are usually a large number of frames to be processed in motion analysis applications. In this paper a novel method for determining marker pixels is proposed. The basic idea behind the proposed method is to reduce the repeated observations of pixels that have already been examined. The method is compared to two classic fill algorithms that are used for this purpose. The results indicate that the proposed method performs better for small circular markers used in this type of applications.