计算机应用与软件
計算機應用與軟件
계산궤응용여연건
COMPUTER APPLICATIONS AND SOFTWARE
2015年
4期
165-169,221
,共6页
图像定位%改进的SIFT算法%无人机%K-means聚类算法
圖像定位%改進的SIFT算法%無人機%K-means聚類算法
도상정위%개진적SIFT산법%무인궤%K-means취류산법
Image localisation%Improved SIFT algorithm%Unmanned aerial vehicle (UAV)%K-means clustering algorithm
针对在GPS(全球定位系统)失效的环境中无人旋翼直升机定位问题,提出基于视觉的迭代定位算法。采用改进的SIFT (尺度不变特征转换)图像匹配算法得到匹配点坐标,再经过坐标变换解算得到无人机三维坐标信息。改进SIFT算法中的关键点特征描述符由128维降低到64维,K-means聚类算法的应用减少了关键点的数目,从而减少了匹配时间而且去除了不必要的点,提高了匹配精度。算法实现阶段匹配点的选择利用了一幅图像匹配两次的思想,前后两次匹配结果中找到相同的特征点用于下一步的位置解算,使得算法能自动从大量特征点中找到所需要的点。仿真实验计算出无人机位置相对误差不超过4%,验证了定位算法的有效性和可靠性。
針對在GPS(全毬定位繫統)失效的環境中無人鏇翼直升機定位問題,提齣基于視覺的迭代定位算法。採用改進的SIFT (呎度不變特徵轉換)圖像匹配算法得到匹配點坐標,再經過坐標變換解算得到無人機三維坐標信息。改進SIFT算法中的關鍵點特徵描述符由128維降低到64維,K-means聚類算法的應用減少瞭關鍵點的數目,從而減少瞭匹配時間而且去除瞭不必要的點,提高瞭匹配精度。算法實現階段匹配點的選擇利用瞭一幅圖像匹配兩次的思想,前後兩次匹配結果中找到相同的特徵點用于下一步的位置解算,使得算法能自動從大量特徵點中找到所需要的點。倣真實驗計算齣無人機位置相對誤差不超過4%,驗證瞭定位算法的有效性和可靠性。
침대재GPS(전구정위계통)실효적배경중무인선익직승궤정위문제,제출기우시각적질대정위산법。채용개진적SIFT (척도불변특정전환)도상필배산법득도필배점좌표,재경과좌표변환해산득도무인궤삼유좌표신식。개진SIFT산법중적관건점특정묘술부유128유강저도64유,K-means취류산법적응용감소료관건점적수목,종이감소료필배시간이차거제료불필요적점,제고료필배정도。산법실현계단필배점적선택이용료일폭도상필배량차적사상,전후량차필배결과중조도상동적특정점용우하일보적위치해산,사득산법능자동종대량특정점중조도소수요적점。방진실험계산출무인궤위치상대오차불초과4%,험증료정위산법적유효성화가고성。
To solve the localisation problem of unmanned rotorcraft in GPS (globe positioning system)signals failure environment,we propose a vision-based iterative localisation algorithm.It finds the coordinates of matching points by using the improved SIFT (scale invariant feature transform)image matching algorithm,and then calculates UAV’s 3D coordinates information by solving coordinates transformation.In the improved SIFT the dimensions of key feature points descriptor are reduced from 1 28 to 64,and the application of k-means clustering algorithm decreases the amount of key features so that the matching time is reduced as well while those unnecessary points are also eliminated, thus the matching accuracy is improved.In the phase of algorithm implementation,the selection of matching points makes use of the idea of matching one image twice,from two matching results one after the other the same feature points are found for using in the next position solution,this enables the algorithm to find the required points from a great deal of feature points.Simulation experiment calculates that the relative error of UAV position is less than 4%,which validates the effectiveness and reliability of this localisation algorithm.