Research On Human Detection And Tracking Technology Using Feature Fusion Based On UAV Vision

The development of UAV has benefited from the development of MEMS technology,the miniaturization of the control system,the improvement of power system capability,and the rise of smartphones and social networks.It can be said that the UAV is the product of the development of the Times,it also from the military field gradually expanded to people's lives.The four-rotor unmanned aerial vehicle is widely concerned with its simple operation,convenient take-off and landing,smooth flight and so on.To achieve human body target detection and tracking in the four-rotor flight platform,has entertainment,search and rescue,patrol reconnaissance and other application value.Human target detection and tracking technology has always been an important subject in the field of computer vision research,which covers the knowledge of image processing and machine learning.But the body shape of human changes easily,different styles of dress,but also the complex environment,the illumination change,appears the occlusion and so on.All of these factor is the research difficulties.In addition,the property control of space,load,cost and airborne computing resources in UAV flight platform is also the challenge in research.Aiming to solve the above problems,this thesis puts forward a method of detecting and tracking the human target by using the Fusion feature under UAV Vision,and the specific research work is as follows:1.Firstly,we research a method of human body detection using HOG features,the local edge information.Then,improve the algorithm,a supervised LDA linear discriminant algorithm is used to reduce the dimension of the extracted HOG features,and an improved fast HOG feature method is proposed.2.In order to solve the problem of insufficient performance of single HOG feature in complex environment,the LBP feature that characterizes the image texture is introduced.And improve the detection process,using two training classifiers for human detection.3.This thesis research a kind of mainstream Camshift tracking algorithm combined with EKF algorithm.The Camshift algorithm in the prediction area with the fusion of EKF algorithm is faster in iteration and more accurate in searching.It also has a good tracking effect in the case of partial occlusion of the tracking target.4.The tracking algorithm proposed in this thesis can achieve automatic detection and tracking of human targets,and in the event of loss of tracking,when the human target reappears in the field of vision,human detection can be used to reacquire the target area and the target can be continuously tracked.The improved algorithm is tested by the data set in different scenes,the result of the experiment shows that the detection and tracking effect of human is greatly improved.