Dynamic window approaches(DWAs) are widely used in local obstacle avoidance of mobile robots. A self-adaptive DWA algorithm is proposed for the problem that the classical DWA exists unreasonable path in dense obstacles and it cannot ensure both speed and security. In order to adapt to the dynamic environment, the weight of speed in the target function is adjusted automatically based on the distance and orientation between the robot and obstacles. The optimal velocity and reasonable planned path of the mobile robot are obtained using the self-adaptive DWA. Thus the performance of the robot crossing an area of dense obstacle is significantly improved. The problems that the robot might move around the dense obstacles and result in an unsmooth path are solved. Our experiments show that the proposed algorithm can make running track more reasonable through gradual optimization in the complex environment than old one, and it ensures the smoothness and security of robot route simultaneously. When the robot is far from the dense obstacles, the robot keeps high speed. When the robot passes through a narrow passage or a dense obstacle area, the speed appropriately reduces and the safety grows high. The total number of iterations and the run-time can be reduced more than 20% in our experiments.