实时鲁棒的特征点匹配算法

发布时间：2018-06-04 05:29

  本文选题：特征点匹配 + 实时　； 参考：《中国图象图形学报》2016年09期

【摘要】：目的针对传统的图像特征点匹配算法数据量大,计算耗时长的特点,提出一种实时鲁棒的特征点匹配算法(RRM)。方法通过微分操作确定图像的边缘区域,找出边缘区域中很有可能成为特征点的锚点,即梯度局部最大的点。对于每个检测出来的特征点,通过计算Intensity Centroid来确定特征点的方向,并且使用改进的Brief来对特征点进行描述,使之具有旋转不变性。最后,结合Hamming距离和对称匹配检验对特征点进行匹配。结果本文算法与多种算法进行对比,在光照发生变化的情况下,RRM表现出明显的优越性和稳定性,正确匹配率达到83%左右,而其他算法的准确匹配率随着光照的变暗明显下降;在视角、尺度和旋转变化条件下,RRM也具有较高的准确匹配率。结论实验结果表明,RRM在保证匹配精度的前提下,有效地解决了传统特征点匹配方法中的缺点。因此,本文算法能更好地应用于图像拼接、目标跟踪和对象识别等领域。
[Abstract]:Aim in view of the large amount of data and long computation time of traditional image feature point matching algorithm, a real-time robust feature point matching algorithm is proposed. Methods the edge region of the image is determined by differential operation, and the anchor point of the edge region which is likely to be the feature point is found, that is, the local maximum point of the gradient. For each detected feature point, the direction of the feature point is determined by calculating the Intensity Centroid, and the improved Brief is used to describe the feature point to make it rotation-invariant. Finally, the feature points are matched with Hamming distance and symmetry matching test. Results compared with other algorithms, RRM showed obvious superiority and stability in the case of light change, the correct matching rate was about 83%, while the accurate matching rate of other algorithms decreased obviously with the darkening of illumination. RRM also has a high accuracy matching rate under the condition of angle of view, scale and rotation. Conclusion the experimental results show that RRM can effectively solve the shortcomings of the traditional feature point matching method on the premise of ensuring the matching accuracy. Therefore, this algorithm can be applied to image mosaic, target tracking and object recognition.
【作者单位】： 广东工业大学应用数学学院;
【基金】：广州市科学研究专项基金项目(201510010059)~~
【分类号】：TP391.41
，

本文编号：1976172