Small Object Tracking And Recognition Under Sky Background

With the development of Modern high-tech, human being hasn’tactivated only at earth surface or Low-altitude sky，but also at outer spacewhich has become the place no country can ignore. And whether the outerspace is well controlled has been a matter of national security. It is one ofthe areas which every country is devoting their high technology to andproceeding military struggles at now. Space Intelligent VideoSurveillance system has been playing a pivotal role in this kind ofmilitary struggle. The task of Space Intelligent Video Surveillance systemis to achieve target capturing, tracking, matching and finding the targetproperty of spacecraft which is a threat to aerospace system，andclassifying the data of target property. But it is difficult to achieve targettracking and recognition when the weather is complex and changeable,especially at the background of night sky. The technology of targettracking and recognizing has combined the knowledge of imageprocessing, pattern recognition, Artificial Intelligence with automaticcontrol etc.It is a advanced topic of multi-disciplinary and challengence.And it has become a hot subject in computer vision field.This paper mainly studies the tracking and recognizing algorithm ofsmall moving object under the sky background. In this paper, the trackingand recognizing algorithms of small moving object under the skybackground are improved as compared with the results of previouspeople’s studies. Main tasks are as follows:Firstly, we introduced the image registration based on starspositioning，after which we used the object detection method based onGaussian mixture model to detect the satellite. Then we estimated track ofobject based on kalman filtering. Secondly, according to the problem of object tracking, in this paperwe put forward the methods based on particle filter and a mixed modelwhich combined affine transformation with particle filter. By using thisalgorithm of particle filter，we try to achieve object tracking. Then wealso studied the method based on affine transformation and particle filter.The data of Experiments shows that object tracking can be well achievedwith this method. Comparing with single model, mixed model hasimproved the accuracy of tracking. This technology also effectivelycombine the color of object with moving information. It eliminates thedisturbing of the background with the similar color as the object.Meanwhile, we find out this kind of method is insensitive to objectbarrier.Finally, according to the problem of object recognition under skybackground, we have used Fourier descriptor and Shape Contextdescriptor to get the feature of object outline. We find the similar outlineto achieve object recognition from those descriptors.From the twoexperiments in chapter Four, we achieve object recognition successfulbased on the two descriptors.