Open AccessA method of farmland 3D terrain reconstruction based on UAV sequence images
Author(s) -
Cheng Xu,
Chenglong He,
Kun Liu
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1314/1/012102
Subject(s) - computer vision , artificial intelligence , terrain , computer science , feature (linguistics) , payload (computing) , centroid , matching (statistics) , point (geometry) , sequence (biology) , position (finance) , geography , mathematics , computer network , cartography , biology , genetics , finance , economics , philosophy , linguistics , statistics , geometry , network packet
When the plant protection UAV on farmland spraying, the UAV need to always maintain a relative height of 1-2 m away from the crops to ensure uniform and efficient spraying of pesticides. At present, the commonly used high-fixing schemes, such as GPS, barometer, ultrasound and so on, there are some deficiencies. Aiming at this problem, a method of reconstructing 3D terrain of farmland by UAV electro-optical payload taking sequence images is proposed, which can achieve the goal of UAV height-fixing. Firstly, the feature points selection module is established. The SILC image segmentation algorithm is adopted to segment the image and the centroid points of all the superpixel blocks are calculated. Then, the corner points are extracted by using Harris corner detector and these two parts of points are mixed together as the feature points. Secondly, the feature point matching module based on SAD algorithm is proposed. Finally, a three-dimensional reconstruction algorithm according to monocular sequence images, UAV position, attitude and camera pointing angle is derived. The results of the experiments demonstrated that this method can quickly and effectively reconstruct farmland terrain, which can provide a data base for UAV high-fixing and has important application value.