The Research On Moving Target Detection And Tracking Algorithm Based On Omnidirectional Vision

The omnidirectional vision is a widely concerned research field in recent years.And it is also a research hotspot in the different countries.Compared with the ordinary camera,its advantage is that it can capture the information of 360 degrees around itself.Therefore, it is necessary to explore the target detection and tracking method that is suitable for the omnidirectional vision.This paper use the omnidirectional vision to investigate the unwrapping of omnidirectional image.In combination with characteristics of the omnidirectional vision,this paper proposed a series of improved methods after the investigation of the methods of moving target detection and tracking.In this papaer,there are the main works and contributions finished as follows:Firstly, omnidirectional image unwrapping.Although the omnidirectional vision has a large visual field,its images exist serious distortion and distortion.So the images need be unwrapped.This paper discussed the cylinder unwrapping algorithm through analyzing the process of imaging of the omnidirectional vision.And on this basis,the paper also studied the fast unwrapping algorithm.It also applied the interpolation algorithm the process of unwrapping to improve image quality.Secondly,the moving target detection.Under the condition of fixed camera,this paper applied the idea of frame differential method to background modeling to propose a background modeling method based on frame difference.The algorithm take advantage of the continuous frame difference information to build gradually background model.What’s more,a new background updating strategies are put forward.In the case of the camera movement,this paper proposed a moving target detection algorithm based on background compensation.It uses the affine transformation to estimate the background motion.Thirdly,target tracking.As traditional Mean- Shift algorithm exists some problems in target tracking and omni-directional image has the problem of nonlinear distortion,this paper put forward the improved algorithm.It uses the adaptive scalingmethod to revise the window bandwidth to meet the target scale adaptive change.And it also combining with Kalman filter to estimate target motion.Finally,the experiment prove that:the improved algorithm applied to the omni-directional vision achieve good effect in tracking the target in the complex environment.