The Research On Improved Optical Flow Algorithms And Their Applications In UAV Positioning

As people’s life is becoming more and more intelligent,the demand for location services becomes more and more urgent,and the demand for precise positioning in the field of drone applications is becoming higher and higher.At present,the widely used GPS positioning technology,in addition to being susceptible to electrical electromagnetic interference,will also be reduced in sheltered spaces(such as buildings,vehicles,etc.).Vision has become a hot field due to its rich image information,low hardware cost and fast access to information.In visual positioning,the improvement of optical flow algorithm has become a research hotspot in recent years.After years of research,optical flow positioning has achieved good results in simple scenarios.However,in some complex light source scenes,the effect of optical flow positioning is not ideal.For example,in low-light scenes,optical flow positioning is unstable;in stroboscopic scenes,optical flow positioning is not accurate.Therefore,in response to these problems,based on the existing research results of our research group,this paper firstly studies and analyzes the research background,significance and the domestic and foreign research status of optical flow algorithm and UAV visual positioning.Secondly,the article carefully studies the basic principles of optical flow algorithms,commonly used optical flow estimation algorithms,their advantages and disadvantages,and the flow of optical flow algorithms in visual positioning applications.This thesis completed the following research content:(1)Study an improved ORB feature extraction algorithm.Aiming at the scenes of low light,multi-scale and multi-region that often appear in the UAV visual positioning,an improved feature extraction algorithm is proposed.This paper mainly improves the ORB feature extraction algorithm from feature points and descriptors by combining SURF algorithm,Shi-Tomasi evaluation criterion and retina-like model.Experimental results show that the improved algorithm has obvious quality and speed improvement for feature point extraction in low light,multi-scale and multi-region scenarios.(2)Study an improved Lucas-Kanade sparse optical flow algorithm based on feature matching.Aiming at the scenes of low light and strobe that often appear in the UAV visual positioning,an improved LK optical flow algorithm is proposed.First,pre-process the image,secondly use the proportional progressive method to weaken the effect of stroboscopic light,then use the improved feature matching rules to filter the improved ORB feature points,and finally use the improved LK golden tower optical flow model for calculation.Experimental results show that the improved algorithm has obvious accuracy improvement for the UAV visual positioning in low light,strobe and other scenes.