UAV Autonomous Positioning Technology For Feature Map

Because the large error will happen in doors or between the buildings,localization based on vision provides a cheap solution for UAV.Therefore,localization based on vision for UAV has been gotten a lot of attention.Different from the localization for ground robot,the localization of UAV in the air is limited from many factors.In the first place,the images obtained by UAV flying in high altitude are different from the images obtained by ground view from scale,affine,light and space-time.Secondly,due to the high speed when UAV flying in the air,the localization algorithm based on vision needs high requirements on time.Finally,the localization algorithm based on vision relies heavily on surrounding environment images.However,the flight control system in UAV always cannot handle so many images in a short time.And it is important for UAV to extra the key features from the environment.This paper focus on the difficulties in the global localization of UAV mentioned above and the following is research contents:(1)Proposed the FA-ORB feature matching algorithm.This algorithm solves the problems of images matching of different source,different time and different views.The FA-ORB based on the ORB feature extraction algorithm,using the affine camera model,setting the parameters of virtual view,complete the image real-time matching between the ground images and air images.(2)Build an affine image fast matching structure based on vocabulary tree.First,paper construct the vocabulary tree by extracting the FA-ORB features and clustering hierarchically.Then we assign the TF-IDF value to vision vocabulary and compute the image similarities.In order to meet rapidly speed in image retrieval in large-scale data set,this paper index the vocabulary with reverse order list model and use heap sort algorithm returning the most similar images of n according to the image similarity score.(3)Localize the UAV based on vocabulary tree and support vector machine.Paper explains the autonomous localization algorithm framework.Through the sparse the images,we construct the local feature map database.Localize the UAV in the local scenes within vocabulary tree and support vector machine.The experimental results showed that(1)The speed of FA-ORB feature matching algorithm matching exceeds the ASIFT about 32 times,A-ORB about 5 times,meeting the needs of real-time.(2)The frame based on visual vocabulary tree can realize the second level image matching in large-scale data set.(3)The UAV localization algorithm based on vocabulary tree and support vector machine(SVM)is more consistent with the human cognitive environment process and can provide reliably global localization results.