面向森林防火的全景拼接系统
|
摘要
为了解决直接用模板匹配直接拼接效果较差并且无法将该方法应用于森林大场景的问题,该文使用了了一种面向森林防火复杂大场景的全景拼接方法,对于输入当前检测范围的图像序列,通过使用特征点从相邻多幅图像中提取sift特征,使用k-d树算法为每个特征点找到最邻近的匹配特征点,然后使用RANSAC算法找到几何一致的特征匹配来得到图像间的对应矩阵,之后使用捆绑调整算法来调整对应矩阵,最后使用多波段融合算法得到最后的拼接全景图。实验结果表明,使用该方法应用到复杂的森林大场景中得到的拼接效果较好。
For solving the problem of directly using template matching and its inability to apply the method to large forest scenes,this paper uses a panoramic mosaic method for forest fire complex scenes. For inputting the sequence of the current detection range, by using feature points to extract sift features from adjacent images, the kd-tree algorithm is used to find the nearest matching feature points for each feature point, and then the RANSAC algorithm is used to find the geometric matching feature to obtain the corresponding matrix between images. After the use of binding adjustment algorithm to adjust the corresponding matrix, and finally use the multi-band fusion algorithm to get the final mosaic panorama. The experimental results show that the splicing effect obtained by using this method in complex forest scenes is better.
For solving the problem of directly using template matching and its inability to apply the method to large forest scenes,this paper uses a panoramic mosaic method for forest fire complex scenes. For inputting the sequence of the current detection range, by using feature points to extract sift features from adjacent images, the kd-tree algorithm is used to find the nearest matching feature points for each feature point, and then the RANSAC algorithm is used to find the geometric matching feature to obtain the corresponding matrix between images. After the use of binding adjustment algorithm to adjust the corresponding matrix, and finally use the multi-band fusion algorithm to get the final mosaic panorama. The experimental results show that the splicing effect obtained by using this method in complex forest scenes is better.
引文
[1]Brown M,Lowe D G.Automatic Panoramic Image Stitching us-ing Invariant Features[J].International Journal of Computer Vi-sion,2007,74(1):59-73.
[2]Lowe D G.Object Recognition from Local Scale-InvariantFeatures(SIFT)[J].1999.
[3]Lowe D G.Distincitive Image Features from Scale-InvariantKeypoints[J].Ijcv,2004,60(2):91-110.
[4]方贤勇.图像拼接技术研究[D].杭州:浙江大学,2005.
[5]Lindeberg T.Scale-Space Theory in Computer Vision[M].Klu-wer Academic,1994.
[6]Triggs B,Mclauchlan P F,Hartley R I,et al.Bundle Adjust-ment—A Modern Synthesis[C]//International Workshop onVision Algorithms:Theory and Practice.Springer-Verlag,1999:298-372.
[2]Lowe D G.Object Recognition from Local Scale-InvariantFeatures(SIFT)[J].1999.
[3]Lowe D G.Distincitive Image Features from Scale-InvariantKeypoints[J].Ijcv,2004,60(2):91-110.
[4]方贤勇.图像拼接技术研究[D].杭州:浙江大学,2005.
[5]Lindeberg T.Scale-Space Theory in Computer Vision[M].Klu-wer Academic,1994.
[6]Triggs B,Mclauchlan P F,Hartley R I,et al.Bundle Adjust-ment—A Modern Synthesis[C]//International Workshop onVision Algorithms:Theory and Practice.Springer-Verlag,1999:298-372.