To improve the flight ability of multi-rotor unmanned aerial vehicle (UAV) in unknown environment, a vision-based UAV path planning and collision avoidance algorithm is proposed. Firstly, an octree map is constructed based on a visual simultaneous localization and mapping algorithm. Then, on the basis of the obstacle information in the map, a three-dimensional path planning algorithm is designed using the artificial water flow method. On the foundation of path planning, the collision avoidance control law for multi-rotor UAV guidance is designed. Furthermore, the integrated software implementation of visual localization, mapping, path planning and collision avoidance guidance algorithm is established. Finally, the multi-rotor UAV platform is constructed, flight control laws and task procedures are designed according to requirements. Flight test results demonstrate the effectiveness and reliability of the designed algorithms, and the capability of the proposed system which safely navigate and perform collision avoidance tasks in global positioning system denied (GPS-denied) environments.