Recently, in order to decrease the number of traffic accident, various automotive safety technologies have been studied. As one of those technologies, the authors have proposed the autonomous frontal obstacle avoidance system, which recognizes a frontal obstacle such as a pedestrian, predicts the location of collision based on the obstacle motion and avoids it by autonomous evasive steering. However, the proposed system calculated avoidance trajectory based on the predicted obstacle position by first-order prediction. Since pedestrians' motion in the real world is not always constant, change of motion of the obstacle must be considered. Therefore, in this paper, the authors solves this problem by consecutive prediction of obstacle position and revision of avoidance trajectory during avoidance action. In addition to, we also proposes the method to judge whether or not it is necessary to update the avoidance trajectory and a method to choose the appropriate avoidance trajectory in response to the change of the motion of the obstacle. Then, the newly proposed system is verified by numerical simulations. The results show that the newly proposed system is effective.