It is sometimes difficult to apply water replacement to measure soil mass density of geomaterials like crushed stones, because particles disrupt sheets owing to their angular shapes. It is also difficult to conduct this method at steep slope ground. To overcome these difficulties, in this study, computer vision technique is newly introduced. Model grounds mainly composed of railway ballasts were prepared in the laboratory. Stereo pair images of the excavated holes were obtained with using a 3D digital camera. Then image analyses were conducted on the stereo pair images and three dimensional shapes of the excavated holes were estimated to evaluate their volumes. Water replacement technique was also carefully attempted on the same holes. The results suggest that the computer vision technique could give soil mass densities which were comparable to those obtained by the water replacement.