This paper proposes an effective approach to the detection of small objects by employing watershed-based transformation. In our work, moving objects with small size and low contrast are first detected from an image sequence which was captured from a video