基于改进Harris算法的快速视频拼接技术研究

发布时间：2018-05-14 12:00

本文选题：快速视频拼接 + 各向同性Sobel算子　；参考：《内蒙古农业大学》2017年硕士论文

【摘要】：由于受自身广角限制,单一摄像头无法得到一系列清晰度较高的视频图像。为了更好地去解决单一摄像头的缺陷,视频拼接技术因此得到发展。伴跟着计算机技术深入研究,尤其是利用计算机视觉的行车智能系统、智能车载技术和主动行驶的研究,实时捕获到目标物体360°范围内的视觉图像,也正被各个领域的研究学者不断的关注。本文针对快速视频拼接算法的拼接性能以及拼接所呈现出的效果进行了研究,重点在一下几个方面展开工作:(1)本文采用普通的CCD摄像头组成视频采集单元,多路摄像头采集通道的同步问题进行了研究。同时,本文对输入视频图像进行预先处理,消除因为外部光线以及视角问题产生的图像质量不高以及摄像头本身产生畸变等问题,有效的优化了拼接质量。(2)本文分析了基于传统的Harris角点算法以及SIFT特征算法的配准方法。基于SIFT特征的算法运行准确度高,但实验流程较为复杂,无法满足快速视频拼接;基于Harris角点的配准算法,复杂度较低,实验运行时间短。最后本文决定采用基于Harris角点的拼接算法,并对原始Harris存在特定点检测准确率低、运行过程复杂等问题,进行了改进和优化,在不降低拼接质量的同时,加快了速度。(3)此外,在视频融合方面,本文采用两步处理的方法,通过分布处理,结合线性融合的方法,得到了可以应用于本文实验的融合方法,实现了快速视频拼接。作为视频拼接的重点,视频帧图像的配对以及融合起到十分核心的地位,本文针对原始Harris角点算法进行改进,采用更合理的各向同性Sobel算子对角点进行处理,同时,利用四邻域判别法,快速选择特征点,加快了算法运行速率。针对算法改进,本文进行了多组实验,最终结果表明本文提出的改进算法是切实有效的。
[Abstract]:Because of its wide-angle limitation, a single camera can not get a series of high-definition video images. In order to solve the defect of single camera, video stitching technology has been developed. Following the in-depth study of computer technology, especially the intelligent driving system using computer vision, intelligent vehicle technology and active driving research, the visual images of the target object are captured in real time in the range of 360 掳. It is also being paid more and more attention by researchers in various fields. In this paper, the performance of fast video stitching algorithm and the effect of stitching are studied. The emphasis is on the following several aspects: 1) this paper uses the common CCD camera to form the video capture unit. The synchronization of multi-channel camera acquisition channels is studied. At the same time, the input video image is pre-processed in this paper to eliminate the problems of poor image quality caused by external light and angle of view, and distortion of the camera itself. This paper analyzes the registration method based on traditional Harris corner algorithm and SIFT feature algorithm. The algorithm based on SIFT features has high running accuracy, but the experimental flow is more complex, which can not meet the fast video stitching. The registration algorithm based on Harris corner has low complexity and short running time. Finally, this paper adopts the algorithm based on Harris corner, and improves and optimizes the original Harris, which has some problems, such as low detection accuracy of specific points and complex running process, which accelerates the speed of stitching without reducing the quality of stitching. In the aspect of video fusion, this paper adopts the two-step processing method, through the distributed processing, combining the linear fusion method, obtains the fusion method which can be applied to the experiment in this paper, and realizes the fast video stitching. As the focus of video mosaic, the pairing and fusion of video frame images play a very important role. This paper improves the original Harris corner algorithm, uses more reasonable isotropic Sobel operator to deal with the corner points, at the same time, The fast selection of feature points by using the four neighborhood discriminant method speeds up the running speed of the algorithm. Several experiments are carried out to improve the algorithm. The final results show that the improved algorithm is effective and effective.
【学位授予单位】：内蒙古农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】