Research On Infrared Object Detection And Tracking Algorithm Based On Global And Local Feature Of The Image

The infrared image is the thermal image which transforms the infrared characteristics of the radiation into electrical signals and then displays in the form of visible light image.Compared with visible light,infrared has the characteristics of strong penetration,good stability,far ranging,night imaging,etc.So infrared image is widely used,and the detection and tracking of infrared target has attracted much attention.The infrared target in this study has three characteristics.The first is that the background of the target is complex.The clutter and interference in the complex background,such as cloud,sea clutter,Gobi desert,mountains and so on,bring great difficulties to target detection.The second feature is that most of the target areas are relatively small.Even for large targets,the area reaches several hundred pixels,and the area proportion of the whole image with an area of 256 * 256 pixels is generally about 1%.The third feature is that the resolution and SNR of the target are relatively low,there is no color information,and the details are not obvious.There are five main problems to be solved in this study.First,the background of the target in this study is very complex.How to eliminate the background interference? How to detect and track the target stably and accurately? Secondly,the target of this study is relatively small,and the resolution and SNR are relatively low.Besides,there is no color information,and the detail features are not obvious,especially for the "small and weak" infrared small and weak target.How to find effective feature highlighting target? Third,how to eliminate the false alarm point caused by clutter interference? Fourth,how to achieve long-term stable tracking? How to find and deal with unexpected situations(such as serious occlusion or distortion of the target)? Fifthly,most of the previous infrared image detection and tracking algorithms are based on the infrared short wave image.The blurred medium wave image is ignored.How to make good use of the medium wave image to enhance the detection and tracking effect?In this paper,we synthetically use image registration(background motion compensation),image background suppression,image significance detection,target similarity measurement,correlation filtering,image fusion,deep neural network and other methods to build a complete detection and tracking framework for infrared small targets.The framework is well adapted to the three characteristics of infrared targets in this study and solves the above five problems in the research.