针对四轴飞行器平台上单目视觉避障算法实时性不够，准确率不高的问题，提出了一种新的基于金字塔LK(Lucas-Kanade)光流与平移光流融合的单目视觉避障算法并给出了这一方法的数学推导过程。在配有320 pixel×240 pixel摄像头的ArduCopter 四轴飞行器上开发和实现了该算法，并进行了多次飞行测试验证了算法的可行性与有效性。在飞行器自主飞行过程中，采用消失点进行导航，并结合神经网络PID对其进行飞行自适应控制。与单目即时定位与地图构建(SLAM)避障算法、Horn-Schunck光流避障算法以及图像分割避障算法比较了算法的实时性和准确性， 结果显示：本算法具有实时性好和准确率高的优点，可以满足实际工程中四轴飞行器的实时避障要求。

For bad real-time ability and poor accuracy rate of the monocular obstacle avoidance algorithm for quad-rotor helicopters in practical engineering applications, a new monocular obstacle avoidance algorithm based on Pyramid Lucas-Kanade optical flow and moving optical flow was proposed and the mathematical derivation process of the algorithm was given. The algorithm was established and implemented on a ArduCopter quad-rotor helicopter equipped with a 320 pixel×240 pixel camera and many experiments was done to verify the feasibility and effectiveness of the algorithm. In the process of vehicle autonomous flight, the vanishing points were used for navigation and a neural net PID was used to control flighting adaptively. As compared with the monocular Synchronous Location Mapping(SLAM) obstacle avoidance algorithm, Horn-Schunck optical flow obstacle avoidance algorithm and the image segmentation obstacle avoidance algorithm, the proposed algorithm shows better real-time ability and accuracy rate, and can meet the requirements of obstacle avoidance in real time for quad-rotor helicopters.