In this study, internal model that estimates visually induced roll-tilt perception was pro-posed. In the model, we used Kalman filter to reproduce the role of the cerebellum which is known to integrate multiple sensory information, such as vision, vestibular and somatosensory inputs, and estimate the current human body motion. To verify the model, we conducted experiments to measure the motion perception induced by trapezoidal (Amy. 2012) and sinusoidal visual stimuli. The result showed that subjects perceived environment tilt in the opposite direction and roll-tilt perception in the same direction. The model showed the consistent estimation to the experiment for both of visual stimuli. Additionally, this result was consistent under visual stimuli with various frequency and constraints. This study has meaning that we applied the sensory integration internal model which has been applied to step or ramp stimulus to the sinusoidal visual stimulus. Also, it can be used as a theoretical framework for estimating human motion perception in 3D animation.