Study On Visual Landing Techniques Of Unmanned Helicopter Under Unknown Environment

Unmanned Aerial Vehicle(UAV)’s autonomous landing is a key technology to achieve UAV’s autonomous flight,of which UAV visual landing technology is an emerging research branch in recent years.The traditional method of UAV visual landing mainly depends on the pre-setting ground cooperation target.Through the detection,recognition and tracking of the cooperation target,UAV’s positioning and autonomous navigation can be realized effectively.However,these methods require manmade pre-setting cooperation target,which means a kind of limitation of the application of UAV.Therefore,in view of the limitations of the above method,this thesis did some researches on the visual landing techniques,which do not depend on any man-made cooperation target,of unmanned helicopters.The main work of this paper is as follow:1.A vision-based unmanned helicopter landing approach is designed.Firstly,navigation regions are detected from ground image and the spatial relationship between landing point and navigation regions is estimated before taking off.Then,the navigation regions are matched and tracked during the process of UAV’s landing.Lastly,the landing point is found according to the spatial relationship above.2.A navigation region detection algorithm based on the saliency of superpixels and region match probability estimation is researched.At first,the salient regions are extracted utilizing superpixel segmentation.After that,regions with high match probability are chosen as final navigation regions through region match probability estimation,which is designed by us.3.The classical kernel correlation filter tracking algorithm is improved.Based on the filter’s response which reflects object’s position in the next frame,a method to adjust the size of the object box is introduced.Therefore,the scale adaptive of kernel correlation filter tracking algorithm is realized.4.A feature classification based safe landing area detection algorithm is researched.First of all,regions that are inappropriate for UAV’s landing are removed according to the appearance.Then,several features are introduced to represent the rest regions.At last,a classifier is trained to classify the feature vectors of candidate areas so as to extract safe landing areas from ground image.