基于神经网络的无人机视觉定位方法

A neural network-based visual positioning method of UAV

先进的导航系统在无人机系统中具有重要作用。无人机高精度自主定位导航技术是决定无人机完成作战任务、提高生存力的关键。基于神经网络设计了仅依赖视觉的无人机定位技术。首先,利用SURF算法对无人机拍摄的图像与卫星地图进行匹配,得到二者之间的相似度参数;其次,提取无人机拍摄的图像与卫星地图的不变矩并求其欧氏距离,作为样本集的输入与输出;然后,建立基于BP神经网络模型;最后,在测试过程中,采取两步策略,先利用已训练好的神经网络进行粗匹配,再利用SURF算法检测出的无人机图像和瓦片图像的特征点进行精匹配。结合瓦片的地理位置信息,计算转换得到该景象区域的地理位置信息,从而实现对无人机的定位。

Advanced navigation systems play an important role in drone systems.UAV’s high-precision autonomous positioning and navigation technology is the key to determining the completion of combat tasks and improving its survivability.This paper mainly studies the design of drone positioning technology that only depends on vision based on neural network.First,the SURF algorithm is used to match the image taken by the drone with the satellite map to obtain the similarity parameters between the two.Second,the invariant moment of the image taken by the drone and the satellite map was extracted and its Euclidean distance as the input and output of the sample set was found.Then,a model based on BP neural network was built.Finally,during the test,a two-step strategy is adopted.In this strategy,the trained neural network for rough matching was applied,and then the SURF algorithm to detect the feature points of the UAV image and the tile image for precise matching was used.Combined with the geographic location information of the tiles,the geographic location information of the scene area is calculated and converted to achieve the positioning of the drone.