Research On Reducing Complexity Of SURF Feature Detection Algorithm For Video Stitching

With the development of science and technology,security monitoring,video conferencing and many other fields are increasingly used in video splicing,using video splicing technology,people can obtain large-format,wide-angle video images,making up for the lack of a single camera.At the same time,it is easier to find the target information in the spliced screen.Therefore,this article uses video splicing to get panoramic video.SURF algorithm is currently a widely used video splicing algorithm,but SURF algorithm will detect a large number of feature points in the region with rich texture,resulting in longer feature extraction and matching time and slower image splicing speed.This affects the effect of video stitching.In view of the deficiency of SURF feature detection algorithm,this paper makes the following improvements to SURF algorithm: First,it limits the range of feature extraction,and only extracts features from overlapping regions,which reduces the number of extracted feature points;The second method is to set thresholds for the number of detected feature points and the distance between feature points so that the detected feature points are evenly distributed and meet the requirements for feature matching;Third,the dimensionality of feature descriptors is reduced.,the dimension of the descriptor is reduced from 64 to 16 dimensions,which not only reduces the feature points but also speeds up the splicing of images.The improved SURF algorithm is used to extract feature points,and rough matching of feature points is performed.Hessian matrix traces are used for rapid matching in rough matching;feature purification is performed using RANSAC to reduce false matching pairs;common image fusion algorithms are compared and the weighted average is finally selected;the last generation of the mosaic effect results in better splicing of the image.This experiment verifies that the improved SURF feature detection algorithm compared to the original SURF algorithm detects image feature points,the number of improved feature points is reduced from 314 to 73,and the feature extraction time is reduced from 3222 ms to 1138 ms.This article also uses a binocular camera to complete the video splicing system.Through real-time analysis,we can see that in addition to the first frame splicing takes a long time,the remaining frames of the splicing time has reached 28 ms,to meet the video splicing real-time requirements.Finally,this paper also applies the system to traffic monitoring systems.