无人机双目视觉鲁棒定位方法

Robust localization method for unmanned aerial vehicle binocular vision

无人机(unmanned aerial vehicle,UAV)在全球定位系统(global positioning system,GPS)信号拒止环境中的应用受到限制,传统视觉同步定位与建图(simultaneous localization and mapping,SLAM)技术一定程度上解决了该问题,但在动态场景和弱纹理场景中定位精度较差。针对该问题提出一种基于双目视觉的多场景鲁棒SLAM方法,重点考虑了真实环境中的动态和弱纹理2类具有挑战性的场景,利用双目相机为UAV在动态和弱纹理场景中提供位姿信息。针对动态场景利用掩膜基于区域的卷积神经网络(mask region-based convolutional neural network,Mask R-CNN)分割潜在动态内容并剔除动态特征,通过计算稠密光流同步相邻帧的掩膜,减小了掩膜的计算成本。对于弱纹理场景,在传统SLAM算法使用的点特征基础上融合了线特征,充分利用了环境中的结构特征。数值模拟和仿真实验证明了本文算法具有更高的鲁棒性和精确性。

The application of unmanned aerial vehicle(UAV)in a global positioning system(GPS)signal rejection environment is limited,traditional visual simultaneous localization and mapping(SLAM)technology has solved this problem to a certain extent,but the positioning accuracy is poor in dynamic scenes and low texture scenes.To solve this problem,a multi-scene robust SLAM method based on stereo vision is proposed.This method focuses on two types of challenging scenes—dynamic and low texture in a real environment,and uses binocular cameras to provide pose information for UAVs in dynamic and low texture scenes.For dynamic scenes,the mask region-based convolutional neural network(Mask R-CNN)is used to segment potential dynamic content and remove dynamic features,and reduce the computational cost of the mask by calculating the mask of dense optical flow to synchronize adjacent frames.For low texture scenes,based on the point features used in traditional SLAM algorithms,line features are fused to make full use of the structural features in the environment.Numerical simulation and experimental verification have demonstrated that the algorithm presented in this paper possesses higher robustness and accuracy.