无人机侦察视频超分辨率重建方法
发布时间：2018-07-20 10:44
【摘要】：目的无人机摄像资料的分辨率直接影响目标识别与信息获取,所以摄像分辨率的提高具有重大意义。为了改善无人机侦察视频质量,针对目前无人机摄像、照相数据的特点,提出一种无人机侦察视频超分辨率重建方法。方法首先提出基于AGAST-Difference与Fast Retina Keypoint(FREAK)的特征匹配算法对视频目标帧与相邻帧之间配准,然后提出匹配区域搜索方法找到目标帧与航片的对应关系,利用航片对视频帧进行高频补偿,最后采用凸集投影方法对补偿后视频帧进行迭代优化。结果基于AGAST-Difference与FREAK的特征匹配算法在尺度、旋转、视点等变化及运行速度上存在很大优势,匹配区域搜索方法使无人机视频的高频补偿连续性更好,凸集投影迭代优化提高了重建的边缘保持能力,与一种简单有效的视频序列超分辨率复原算法相比,本文算法重建质量提高约4 d B,运行速度提高约5倍。结论提出了一种针对无人机的视频超分辨率重建方法,分析了无人机视频超分辨率问题的核心所在,并且提出基于AGAST-Difference与FREAK的特征匹配算法与匹配区域搜索方法来解决图像配准与高频补偿问题。实验结果表明,本文算法强化了重建图像的一致性与保真度,特别是对图像边缘细节部分等效果极为明显,且处理速度更快。
[Abstract]:Aim the resolution of UAV camera data directly affects target recognition and information acquisition, so the improvement of camera resolution is of great significance. In order to improve the video quality of UAV reconnaissance, a super-resolution reconstruction method for UAV reconnaissance video is proposed according to the characteristics of UAV camera and photographic data. Methods first, a feature matching algorithm based on AGAST-Difference and Fast Retina Keypoint (FREAK) is proposed, and then the matching region search method is proposed to find the corresponding relationship between the target frame and the aerial picture. Finally, the convex set projection method is used to optimize the compensated video frame iteratively. Results the feature matching algorithm based on AGAST-Difference and FREAK has great advantages in the changes of scale, rotation, viewpoint and running speed. The matching region search method makes the UAV video frequency compensation continuity better. Compared with a simple and effective super-resolution video sequence restoration algorithm, the reconstruction quality of this algorithm is improved by about 4 dB, and the running speed is about 5 times higher than that of a simple super-resolution video sequence restoration algorithm. Conclusion A video super-resolution reconstruction method for UAV is proposed, and the core of the video super-resolution problem of UAV is analyzed. A feature matching algorithm and a matching region search method based on AGAST-Difference and FREAK are proposed to solve the problem of image registration and high-frequency compensation. The experimental results show that the proposed algorithm enhances the consistency and fidelity of the reconstructed image, especially for the details of the image edge, and the processing speed is faster.
【作者单位】： 军械工程学院;
【基金】：国家自然科学基金项目(51307183) 军内科研项目(ZS201507132A1208)~~
【分类号】：V279.3
本文编号：2133273
[Abstract]:Aim the resolution of UAV camera data directly affects target recognition and information acquisition, so the improvement of camera resolution is of great significance. In order to improve the video quality of UAV reconnaissance, a super-resolution reconstruction method for UAV reconnaissance video is proposed according to the characteristics of UAV camera and photographic data. Methods first, a feature matching algorithm based on AGAST-Difference and Fast Retina Keypoint (FREAK) is proposed, and then the matching region search method is proposed to find the corresponding relationship between the target frame and the aerial picture. Finally, the convex set projection method is used to optimize the compensated video frame iteratively. Results the feature matching algorithm based on AGAST-Difference and FREAK has great advantages in the changes of scale, rotation, viewpoint and running speed. The matching region search method makes the UAV video frequency compensation continuity better. Compared with a simple and effective super-resolution video sequence restoration algorithm, the reconstruction quality of this algorithm is improved by about 4 dB, and the running speed is about 5 times higher than that of a simple super-resolution video sequence restoration algorithm. Conclusion A video super-resolution reconstruction method for UAV is proposed, and the core of the video super-resolution problem of UAV is analyzed. A feature matching algorithm and a matching region search method based on AGAST-Difference and FREAK are proposed to solve the problem of image registration and high-frequency compensation. The experimental results show that the proposed algorithm enhances the consistency and fidelity of the reconstructed image, especially for the details of the image edge, and the processing speed is faster.
【作者单位】： 军械工程学院;
【基金】：国家自然科学基金项目(51307183) 军内科研项目(ZS201507132A1208)~~
【分类号】：V279.3
【相似文献】
相关期刊论文 前3条
1 杜小平,赵继广,崔占忠,曾德贤;基于超分辨率重构的航天器位置姿态测量方法[J];北京理工大学学报;2005年03期
2 顾聚兴;超分辨率三维成像传感器[J];红外;2001年10期
3 ;[J];;年期
，本文编号：2133273
本文链接：https://www.wllwen.com/kejilunwen/hangkongsky/2133273.html
