Vision-based UAV Flight Environment Perception Measurement System

In recent years,UAV(unmanned navigate aircraft)technology has developed rapidly and is widely used in the fields of reconnaissance,mapping,patrol and agricultural plant protection.There are more and more scenes for UAV,its flight environment is becoming more and more complex,and the protection of UAV flight safety has become a hot issue of current research.Especially in the space covered area and indoor area,UAV's GPS navigation signal is not available,the drone's flight safety problem is more prominent when the GPS navigation signal is poor or even completely signal areas,and the development of visual technology has brought about the development of flight safety of the drone indoors or in poor signal areas.Based on the research of vision technology and UAV flight safety,this paper studies a general vision-based flight environment perception measurement system for small UAV in the indoor environment,identifies the indoor scene to find the path target and the obstacles in the perceptual environment,and finally provides the corresponding flight strategy.Firstly,the general principle and process of vision algorithm are analyzed,and the flight environment characteristics of indoor UAV are summarized.On this basis,the appropriate hardware is selected,and the calibration of binocular camera are carried out.Then,according to the filtering processing in the calibration work,an improved idea is proposed to reduce the processing calculation.Secondly,aiming at the selection of matching window and the aggregation of matching cost in stereo matching algorithm,three improve ideas are proposed.One is the color threshold when the matching window is established,and the adaptive color threshold method is added to make use of the image's own regional features to improve the quality of window selection.The other is the noise elimination principle in the matching window based on the statistical principle.The third is to use fixed weights for the original algorithm matching cost,which leads to the problem that the parameters are not adaptable to the image.The adaptive weight method is adopted according to the image features.Finally,experiments are carried out to verify the improvement effect.Then use neural network to classify and identify common indoor scenes,then use Hough transform to detect the linear elements in the scene,use its distribution characteristics to find the path target,and judge the relative position of the drone,then study how to get the double from the match.The obstacles are extracted and measured in the visual difference diagram.The reason why the original method is fast and the effect is not ideal is analyzed.The efficiency of a certain time is sacrificed without affecting the overall performance.The iterative idea is added to the position of the obstacle.Further accurate,and experimentally verified its feasibility and effect,and finally used the distance and size information of the combined obstacles to give the corresponding flight strategy to ensure the safe flight of the drone.Finally,experiments were carried out to verify the classification effect and the perceived accuracy of the path target.