Obstacle detection, representation and avoidance are crucial tasks for indoor and outdoor mobile robot navigation. A suitable representation of obstacles can improve the performance of obstacle avoidance. The objective of this paper is to develop a smooth representation of obstacles sensed by 2-D laser range finder in real time in unknown environments. The proposed method relies on line fitting methods while introduces a criterion to extract convex chain of an obstacle boundary. A cubic B-spline curve is then used to approximate the convexity of the obstacle boundary. We apply the proposed method to build a reactive collision avoidance task. The proposed method is validated with both simulation and experimental results of a mobile robot navigation task.