Control barrier functions (CBFs) have been successfully implemented in various control strategies; one of those is a human assist control for moving obstacle avoidance by using time-varying CBFs. However, the human assist control contains the complete information on the motion of environments; in general, a derivative of moving obstacle states needs estimating. In this paper, we apply an exact differentiator to estimate derivatives of obstacle state signals in real time. Then, we propose a human assist control by using both a time-varying CBF and an exact differentiator. Moreover, the effectiveness of the proposed method is confirmed by computer simulation and experiments of an electric wheelchair.