基于视觉的无人机自主着陆地标识别方法

在无人机自主着陆过程中,传统地标识别方法的相似阈值确定需大量实验估计。为解决此问题,采用一种基于仿射不变矩和支持向量机的识别方法,首先设计了六圆组合的图标作为无人机自主着陆地标;由于无人机会拍摄到发生扭曲的地标图像,因此提取地标的仿射不变矩作为输入特征;最后将其输入支持向量机分类模型,完成地标的识别。与传统的几何不变矩和BP神经网络相比较,该方法提高了地标的识别精度并降低了识别测试时间,因此对地标识别具有一定的实用性。

Landmark recognition for UAV autonomous landing based on vision

In the traditional method of recognizing the landmark during the process of UAV autonomous landing, conformed the threshold through lots of experiments. In order to solve this problem, this paper studied a kind of method based on affine invariant moments and SVM classifier. First of all, it designed a new landmark combined with 6 circles. Second, considering the fact that the UAV in flight could take distorted landmark images, it extracted the affine invariant moments as features. Finally, it put affine invariant moments into SVM classifier to complete landmark recognition. It compared the proposed method with Hu invariant moment and BP neural network. The experimental results show that combination of affine invariant moment and SVM classifier improve the accuracy and decrease test time of UAV landing landmark classification. Therefore, the classification method based on affine invariant moments and SVM classifier has a certain degree of practicality in the landmark recognition.