Research On UAV Autonomous Positioning Technology Based On Visual Information

Nowadays,due to the advantages of UAV positioning and navigation technologies based on computer vision information with small collaborative interference and strong real-time performance,they have quickly become research focuses in the current UAV field.Computer vision technologies are utilized to process the images acquired by UAV imaging sensor and satellite map.And then the location information of UAV will be obtained to achieve autonomous navigation.In this paper,UAV visual localization and navigation methods will be researched based on image interest point extraction and image matching.First,a visual positioning method for UAV based on neural network is proposed.This method adopts a two-step strategy with coarse matching and fine matching.The coarse matching uses the SURF algorithm to obtain the similarity parameters between the ground image collected by the UAV and the satellite map,and extract the image invariant moments.The similarity between the invariant moments of the two images is used as the input of the network,and the matching degree of the two images is used as the output of the network,and then the BP neural network model will be established.The trained neural network is used for coarse matching between the UAV images and the tile images,and then the feature points of the UAV image and the tile image detected by the SURF algorithm are used for fine matching.The information of the optimal tile selected by the fine matching is converted into the UAV latitude and longitude information,in order to realize the positioning of the UAV.Then,a visual tracking algorithm based on SURF and Kanade-Lucas-Tomasi method for UAV is proposed.This method constructs an image pyramid of two adjacent frames and extracts the SURF corner points of each layer of this image.It calculates the offset of each feature point from the top of the pyramid,and lets the the sum of the starting position and the offset be the starting position of next tracking point.The offsets for all the feature points will be obtained,also the results of the proposed tracking algorithm.In order to verify the feasibility,efficacy and accuracy of the proposed methods,images collected by UAV in Shenbei and Hunnan areas of Shenyang City are used to simulate.The first algorithm can accurately obtain the location information of UAV.And the visual tracking algorithm is verified on the continuous frame images of the Hunnan Changbai.Compared with Harris corner-based and SIFT-based visual tracking algorithms,the results show that the tracking results obtained by the proposed method in this paper are more accurate and faster.Meanwhile,a GUI interface is designed,and visual results such as the location informations and the flight trajectories of UAV can be easily obtained through this interface.This research results of this paper will effectively solve the current application problems of UAVs in civil and military fields,and have important engineering application value.