Three-dimensional (3-D) motion analysis is widely used in numerous fields of physical education, including biomechanics, coaching and test and measurement. Among the available 3-D motion analysis methods, the nonlinear transformation (NLT) method is frequently used. Many automated motion capture systems currently utilize this method. However, as this method is based on the perspective projection of ordinary cameras, its use is limited in various situations. Because ordinary cameras have a relatively limited angle of view, they cannot cover a wide target area or be set up near the target area. In this study, we propose a method for 3-D data collection using omnidirectional cameras that can capture images of the entire celestial sphere. For the points of interest, the longitude and latitude of the celestial sphere were obtained from the image of an omnidirectional camera. The direction of the camera ray was determined by considering the polar coordinate system with the longitude and latitude. During the calibration process, the relative position and orientation of the omnidirectional cameras were determined from the images of the control points with unknown 3-D coordinates. We verified that the accuracy of the proposed method (resultant error: 0.34%) was comparable to the accuracy reported in preceding studies regarding traditional NLT and omnidirectional Direct Linear Transformation-inspired methods.