Object Detection Based On Rotation-invarient Histogram Of Oriented Gradient In Aerial Images

The rapid development of remote sensing technologies has rendered many aerial sensors to provide optical imagery with high spatial resolution.Feature information of the objects in aerial remote sensing images is clear because of the advantages of high resolution,large-scale,high potential and so on,so it is easy to acquire objects' attribution by means of computer vision and facilitate the detection of objects.The environment,attitude and perspective that the aircraft is in are different during aerial photography,so the background and illumination of the image captured are different,at the same time there are differences in the rotation angle and size of the object in the image.In order to adapt to this scene,our paper makes a thorough study of object detection technology in the aerial image.Firstly,our paper introduces the background and significance of the thesis,the current domestic and foreign situation.Then,the machine vision technology is used to design an algorithm of object detection for the aerial image,the algorithm mainly includes feature extraction,classifier training and object detection.Finally,an experiment with the plane as the target is conducted and the experimental results are analyzed and evaluated.The design and implementation of the algorithm is the focus of our paper.On one hand,HOG feature,namely,Histogram of Oriented Gradient,is described.However,HOG feature can not perform well when facing objects with rotation variation,so our paper improves HOG descriptor to enhance its affine-invariant ability.Radial Gradient Transform(RGT)method is adopted to obtain the RGT gradient,which is affine-invarient,adopting the similar method of the HOG descriptor,RGT Histogram of Oriented Gradient in one Cell is counted,so a plurality of descriptors with rotation angle information are obtained.On the other hand,In view of the fact that the HOG descriptor can only detect the object the size of the training sample,our paper proposes a Multi-Scale Window detection algorithm.Finally,the combination of the rotation-invariant HOG feature and the Support Vector Machine(SVM)is used to detect planes with rotation angle,at the same time,the detection rate,false detection rate and time comsumption are computed.At the last of our paper,combinating the design and implementation of our algorithm,the shorage of the experiment and the future work are discussed.