摘要
结合SIFT特征和正交DLT算法,给出一种较为精确的物体识别方法。该方法首先采用SIFT特征描述子进行测试图像关键点的提取,然后将测试图像和模型数据库中模型的特征点逐一匹配,设定匹配阈值,若匹配达到该阈值,则认为匹配成功,最后通过正交DLT算法在测试图像中识别该模型的位置。SIFT特征与正交DLT算法的结合有效地提高物体的识别精度。实验结果表明,该方法具有较强的鲁棒性,不仅可以在复杂背景下,较好地识别模型在测试图像中的位置,而且还可以在物体被部分遮挡的情况下,较为准确地对物体进行定位。
A method for accurate object recognition is presented using the Scale -Invariant Feature Transform (SIFT) and normalized Direct Linear Transform (DLT) algorithm. The recognition proceeded by extracting SIFT keypoints from the sample image and matching individual features to a database of features from model images using a fast nearest - neighbor algorithm,followed by a threshold estimation to identify whether the recognition is finished or not, and finally using normalized DLT algorithm to accomplish the approximation to the object location. Moreover,the precision of object recognition can be improved effectively by the combination of SIFT features and normalized DLT algorithm. Experimental results show that the robust object recognition can be achieved in cluttered background and partially - occluded image.
出处
《现代电子技术》
2009年第14期142-145,共4页
Modern Electronics Technique
基金
山西省自然科学基金资助项目(2007012003)
山西省重点实验室基金资助项目(9140C1204020608)
关键词
SIFT特征
正交DLT算法
物体识别
模型匹配
图像分割
SIFT features
normalized direct linear transform algorithm
object recognition
model matching
image segmentation