基于双目视觉的无人机室内定位研究

Research on indoor positioning of UAV based on binocular vision

基于双目视觉测距和RSS自主定位算法,设计了一种利用地标进行辅助定位的无人机室内定位方法,并对定位系统的有效性和定位精度进行了试验仿真。首先,采用K-means聚类分析和图像的腐蚀与膨胀运算,对RGB格式图像中的地标进行识别与分割得到了地标二值图像;其次,利用统计二值图像连通域的方法求解了地标的形心坐标和最小外接矩形位置,并对地标形心的定位做了误差分析。针对SURF算法匹配误差和计算速度问题,引入图像立体校正和地标最小外接矩形确定的区域匹配约束进行改进,成功得到了地标区域的最佳匹配特征点坐标;最后,设计实验验证了测距精度,并通过对垂直距离和视差倒数的一次多项式拟合,得到了更精确的垂直坐标测量模型。仿真结果表明在模组和算法可正确识别地标的空间范围内,方法的定位误差在3 cm以内,对无人机的室内定位应用有实际意义。

An indoor positioning method of UAV was designed based on binocular vision ranging and RSS autonomous positioning algorithm,which couldrealize assistant position by landmarks.Effectiveness and positioning accuracy of the positioning system were also simulated andtested.At first,the landmarks in RGB image wereidentified and segmentedby K-means clustering analysis and image corrosion and expansion calculationto obtain the landmarks in binary image.Centroid coordinates and minimum external rectangle positions of the landmarks were further calculated by the connected domain method of binary images.Errors of the centroid coordinates were also analyzed.In order to solve matching error and computing speed problem,SURF algorithm was improved to obtain the coordinates of optimal matching feature pointsby introducingthe correction of 3D image and the region matching constraints caused by the minimum external rectangle of landmarks.Finally,the range measuring accuracy was verified by designing experiments,and a more accurate vertical coordinate measuring model was established by fitting the linear polynomial aboutthe vertical distance and the reciprocal of parallax.The results show that positioning error of the method is less than 3 cm within the space range where the module and algorithm can correctly identify landmarks,which has a practical significance for the indoor positioning of UAV.