Generally, image motion is induced all over the retina when an observer moves around. In spite of this, we can immediately detect objects moving in our surroundings. In this paper, a computational model is proposed to explain such detection using self-motion signals. First, we analyze the general characteristics of retinal image motion caused by both object motion and observer motion. Then a computational model is constructed in which true object motion is detected using signals of self-motion and depth values under the supposition that the surroundings are always stationary. It is suggested that the nervous system uses similar mechanisms.