Research On Image Registration Method For Unmanned Aerial Vehicle Remote Sensing Images

Image registration is a technique which deals with images consisted of somesame areas in part. This kind of images can be obtained from different camerasworking towards the same target. Image registration technique has been widely usedin the remote sensing areas for years. Recently, unmanned aerial vehicle(UAV) hasattracted many attentions because of its flexibility and low cost, as a result of which,UAV remote sensing system becomes a new and exciting way to get aerial images.However, aerial images got by UAV remote sensing system can only cover a smallpart of the interested areas because UAV tends to fly much lower than the traditionalsatellite or planes controlled by pilots. So it is necessary to apply image registrationtechnique upon UAV remote sensing aerial images.Image registration technique has developed for years, and many different kindsof ways have been proposed. Feature-based method of registration is very popularnow, because this kind of registration technique can deal with images of differenttransforms. But feature-based method is computational expensive. When dealing withUAV remote sensing aerial images, feature-based method needs huge amount ofresources and it works very slowly. What’s more, UAV remote sensing aerial imagesusually have huge pixels, which make the situation worse.In this paper, we analyze many kinds of feature-based registration methods, andon the basis of it, we propose an improved similar-curve based registration methodfor UAV remote sensing system. This method achieves a good balance betweentechnique’s computational cost and result’s precision. The main contents of this paperare as follows:1. This paper first introduces some basic geometric knowledge of cameraimaging system and analyzes in detail the UAV remote sensing aerial images’characteristics. Besides, this paper describes two popular feature-based registrationtechniques’ principles, on the purpose of preparing a comparative experiment with theproposed method. 2. On the consideration of the large amounts of UAV remote sensing aerialimage files and the high resolution of the single image, an improved similar-curvebased image registration method is proposed, which first estimates the approximaterotation angle between two images using edge feature and then registers the images.This paper takes image rotation into consideration and uses Grey Relational Analysismethod to analyze the similar-curves, thus making the proposed method insensitive toimage rotation and scale. So the proposed method can be used on UAV remotesensing aerial images efficiently. In order to get more precise results, this paper usesHarris-based method as the last step of refined search.3. The image sets got by UAV remote sensing system can easily be disorderlyranked. On the consideration of this problem, this paper uses image color features andtexture features to analyze the images’ similarities. In this way the imagerearrangement into right orders can be achieved. This method is computational cheapand fits to deal with UAV remote sensing aerial images.The experimental results show that the proposed registration method cansuccessfully be used on UAV remote sensing aerial images, which has lower timecomplexity and space complexity than the traditional ones. When dealing withdisorderly arranged aerial images, the proposed method of image rearrangement canbe used as a preprocessing part, achieving full-automatic image registration.