红外与激光工程
紅外與激光工程
홍외여격광공정
INFRARED AND LASER ENGINEERING
2014年
8期
2709-2714
,共6页
三维目标识别%局部形状描述符%测地自旋图
三維目標識彆%跼部形狀描述符%測地自鏇圖
삼유목표식별%국부형상묘술부%측지자선도
3D object recognition%local shape descriptor%geodesic-spin-image
针对自旋图中欧氏测量存在歧义、形状信息存在丢失的现象,提出一种新的三维物体表面局部形状描述符:测地自旋图(Geodesic-spin-image, GSI)。GSI 采用测地距离取代欧氏距离限定局部支持区域;GSI 在自旋图的基础上,还引入了3个新的特征,以补偿自旋图在方位角上的信息丢失。这3个特征为:局部支持区域在描述点切平面上投影的长轴长度、短轴长度,以及局部支持区域形心到描述点切平面的距离。基于 GSI 描述改进了现有自旋图匹配算法,首先基于3个新特征进行粗略匹配,而后基于测地支持区域内的自旋图进行精确匹配,以提高识别效率。仿真实验结果表明:相比于标准自旋图方法,该方法提供的物体局部形状信息更丰富,具有更高的目标分辨能力和匹配识别效率。
針對自鏇圖中歐氏測量存在歧義、形狀信息存在丟失的現象,提齣一種新的三維物體錶麵跼部形狀描述符:測地自鏇圖(Geodesic-spin-image, GSI)。GSI 採用測地距離取代歐氏距離限定跼部支持區域;GSI 在自鏇圖的基礎上,還引入瞭3箇新的特徵,以補償自鏇圖在方位角上的信息丟失。這3箇特徵為:跼部支持區域在描述點切平麵上投影的長軸長度、短軸長度,以及跼部支持區域形心到描述點切平麵的距離。基于 GSI 描述改進瞭現有自鏇圖匹配算法,首先基于3箇新特徵進行粗略匹配,而後基于測地支持區域內的自鏇圖進行精確匹配,以提高識彆效率。倣真實驗結果錶明:相比于標準自鏇圖方法,該方法提供的物體跼部形狀信息更豐富,具有更高的目標分辨能力和匹配識彆效率。
침대자선도중구씨측량존재기의、형상신식존재주실적현상,제출일충신적삼유물체표면국부형상묘술부:측지자선도(Geodesic-spin-image, GSI)。GSI 채용측지거리취대구씨거리한정국부지지구역;GSI 재자선도적기출상,환인입료3개신적특정,이보상자선도재방위각상적신식주실。저3개특정위:국부지지구역재묘술점절평면상투영적장축장도、단축장도,이급국부지지구역형심도묘술점절평면적거리。기우 GSI 묘술개진료현유자선도필배산법,수선기우3개신특정진행조략필배,이후기우측지지지구역내적자선도진행정학필배,이제고식별효솔。방진실험결과표명:상비우표준자선도방법,해방법제공적물체국부형상신식경봉부,구유경고적목표분변능력화필배식별효솔。
Geodesic-spin-image (GSI), a new three-dimensional(3D) local shape descriptor, was proposed to solve the problems of ambiguous Euclidean measure and insufficient shape information description when utilizing the Spin Image (SI) descriptor. The local support region of GSI was defined by Geodesic measure instead of Euclidean measure. Besides, three new features were also introduced to compensate the azimuth information loss while these features were long axis length and short axis length of the support region′s projection onto the tangent plane of the given point, and the distance from the centroid of the support region to the tangent plane. Based on the GSI, matching strategy of SI was modified through a combination of coarse matching utilizing the three new features and fine matching utilizing SI defined by geodesic measure, so as to improve the matching efficiency. Simulation experiment results show that compared with the original SI method, the proposed GSI descriptor is proved to provide more shape information and be more discriminative, and its matching strategy is more efficient.