No abstract

No abstract

No abstract

Detection and tracking of moving objects is an important topic of research in the field IntoductionTracking of moving objects in a sight is surely a significant subject matter in the field of computer vision. It is one of the fundamental steps in so many video-based systems. For instance, applications such as intelligent surveillance systems, robot visions and human computer interface (HCI), [1] require detection and tracking of moving objects. Numerous techniques have been proposed and are verified to be victorious for detection of moving objects in case of nonmoving cameras [2,3], but in case of pan-tilt-zoom (PTZ) or mobile cameras the existing methods do not work well because of many factors that take place when using moving cameras. Several methods have been proposed for tracking of moving object. When applied to moving cameras, the motion compensation procedure for the algorithm involves more number of computations comparative to the total number of samples employed for a pixel. This may be time-consuming for the method in significant amounts unless there is some sort of hardware support. Apart from the computations involved, when modeling the scene, it is also important that the model considers not only the errors and noises that arise in stationary cameras, but also the errors that arise when compensating for the motion of the camera. This is a serious reason that we cannot simply just adopt background subtraction algorithms for nonmoving cameras with simple motion compensation algorithms. Algorithms based on moving camera background modeling usually focus on constructing accurate model for every pixel. But for moving case, we cannot assure that the model used to estimate a pixel is in fact related to that pixel. Even the smallest amount of erroneousness in motion compensation could result in building the technique use wrong algorithms for few pixels. To account for such motion compensation errors, small nearby neighborhoods are considered [4]. However, considering neighborhoods increases the necessary computation, slowing down the whole algorithm. In order to take in to consideration the motion compensation errors, the nearby neighborhood pixels in small amounts are considered. But, considering these pixels increases the computations and as a result shows down the whole algorithm. Our model which is based on background subtraction is intended in a way that reduces the computational requirements and demonstrates accurate tracking of path in moving cameras. The pseudo path can be eliminated by using this method. The proposed algorithm gives an accurate trajectory of the object of interest in the case of moving camera. The algorithm assumes that the motion of the camera is either horizontal or vertical and the motion is linear. Section II gives the related work in this area. Section III describes the Gaussian Mixture Model. Section IV explains the Mean Shift Algorithm. Experimental results and conclusions are given in Section V and VI. II. Related WorkRoughly all existing methods for static suspi...