Traditional obstacle avoidance method of UAV is to establish a global map of the environment in the form of an Octomap, and then perform autonomous obstacle avoidance. The autonomous perception obstacle avoidance algorithm based on 3DVFH (three-dimensional vector histogram) does not require the construction of a global map, only real-time construction of a local map of the current position of the LiDAR sensor, improving computational efficiency. However, it does not consider any data or operations in previous time steps, which usually leads to unstable behavior and local minima. This article proposes an autonomous local obstacle avoidance algorithm based on enhanced 3DVFH, which directly replaces the global map with a 3D point cloud provided by LiDAR, and designs a low-cost computational memory strategy to alleviate the inherent problems of local methods. Through experimental verification, it has effectively improved the efficiency and storage capacity of environmental perception, which is conducive to better autonomous obstacle avoidance for UAV.