Region Matching Algorithm Based On Local Features For Cloud Motion Wind

Numerical forecasting is an area closely related to the science and livelihood.With the progressive development of meteorological satellite technology,the information we can get is increasingly rich,and computer vision technology is used for numerical forecasting instead of experience of professionals which makes the results more objective and more accurate.Cloud motion wind technology is a process of tracking cloud and inversing cloud motion vector within a sequence of satellite images,which play an important role in numerical forecasting,especially for places where the location is remote and the weather monitoring sites are scarce.Currently,the cross-correlation based approach is the main method of cloud motion wind technology.However,for the complex semi-fluid motion of cloud,there is no mature and widely accepted model for tracking cloud motion.As for the traditional method,which is mainly based on temple matching,and hard to detect the rotation of clouds caused by cyclone,in addition,the searching process leads to a large amount of calculation and time-consuming and there is still a great dependence on artificial.So there is big space in the studying range.In order to tracking cloud motion better,and improve the computational efficiency,the proposed approach use the distribution of feature points tracked by SIFT algorithm as the cloud feature,and then combine with template matching,which is a new attempt of cloud motion wind technology.The SIFT algorithm can effectively overcome the influence of scale change,rotation and brightness change,so the key points generated by the SIFT algorithm can be used to describe the characteristics of clouds.In order to avoid the disturbance of complex motion of cloud,the proposed approach use the distribution of key points instead of the single key point as the representation of the clouds feature,and then combine the template matching and local features region,which avoids the difficulties of detecting the feature points in the smooth region of the clouds,and greatly reduces computational complexity of the template matching.Finally,the experimental results show that the region matching algorithm based on local features can not only adapt to the variety of cloud motion,but also get a certain rotate information,and obtain more wind vector compared with tradition algorithm.Especially for high-resolution satellite images,the proposed method can significantly improve the efficiency of calculation,while ensuring the accuracy at the same time.