摘要
高精度快速准稠密匹配是计算机视觉和摄影测量等领域的急切而富有挑战性的目标,核线约束下的一维匹配将大幅度提高匹配的速度和准确率。在核线重排影像的基础上,设计了轻量级的线状多尺度三角塔结构,并提出通过多尺度下的广义亮度Sinc序列、广义梯度序列、广义θ序列和广义中心偏离序列的极值检测,得到大量稳定特征点并进行描述,然后采用属性特征和数值描述特征在预测区间进行快速匹配。大量试验表明该方法能够稳健地快速得到分布均匀、密度较大、亚像素级精度的稳定同名点。
Quasi-dense matching with high quality and speed is an urgent and challenging goal within the search scope of computer vision and photogrammetry. One dimensional matching assisted by epipolar line would largely improve the speed and precision. The linear triangle multi-scale tower structure with light computation and high efficiency is proposed based on the epipolar resample images, and series of distinguishable sequences such as extended lumirtartce Sinc sequence, extended gradients sequence, extended θ sequence and extended bias sequence are constructed for detecting the key points on the continuous scale space. And then, the key points are described both by attribute characters and numerical label. And based on it, the quick matching is carded out during the predicted scope. Experiments show that the algorithm can reach quasi-dense matching robustly in the whole continuous scale even on distorted images in defect of features, with more couples in good topology and precision.
出处
《计算机技术与发展》
2013年第4期111-114,138,共5页
Computer Technology and Development
基金
国防预研基金(20060826(重大专项))
关键词
准稠密匹配
核线
特征匹配
计算机视觉
quasi-dense matching
epipolar line
feature matching
computer vision