低空无人机航空摄影高度自动测量方法研究

为提高产品外观质量的检测精度和实时性,提出一种基于特征融合的多尺度滑动窗口机器视觉检测方法。在训练阶段,首先提取图像的HOG特征和Lab颜色特征,并采用典型相关分析法(CCA)进行特征融合。接下来,采用支持向量机(SVM)对融合的特征进行训练,生成分类器。在检测阶段,产品外观不同区域对精度的要求不同,为提高检测效率,生成不同尺度的滑动窗口,在每个窗口中都进行图像的特征提取与特征融合。最后,对采集的图像序列进行匹配,实现产品外观划痕的实时检测。实验中,选取不同的特征提取方法进行对比,并分别生成大小不同的滑动窗口,通过分析实验结果,结合检测时间与精度,确定各个区域的窗口尺度。实验表明,与传统的检测方法相比,所提方法在检测精度和实时性上具有显著提高。

Research on Automatic Measurement Method of Aerial Photogrammetry in Low Altitude UAV

In order to control the altitude of low altitude unmanned aerial vehicle (UAV) and obtain clearer geographical information images, it is necessary to optimize the height measurement method of low altitude unmanned aerial vehicle (UAV). The current method mainly uses the automatic calibration method of projective geometry knowledge to realize automatic aerial photogrammetry height measurement of low altitude unmanned aerial vehicle (UAV). The method has the problems of serious noise influence and large measurement error. Therefore, a highly automatic aerial photogrammetry method for low altitude unmanned aerial vehicle (UAV) based on multisensor and Calman filtering is proposed. Firstly, through the analysis of aerial photography has no effect on the height of the low pressure air pressure measurement method for calculation of various factors, and then push the troposphere with low pressure induced no change in aerial photography height; and then the double GPS system work at the same time, for the low altitude GPS, altimeter and IMU measurements obtained the UAV aerial photography altitude signal redundant backup; using two order polynomial correction method based on low altitude UAV aerial photography sensor output value compensation and correction; according to the kinetic equations of dynamics equation of low altitude UAV aerial photography obtained by measuring the height of the state equation; linear minimum variance estimation criterion Calman filtering of low altitude UAV aerial photography were highly variance estimation and realization of low altitude automatic measurement and calibration. Experimental results show that the proposed method has the advantages of high precision, good convergence and ideal filtering effect.