Study On The Moving Object Detection And Tracking Based On Background Subraction And ORB Algorithm

In recent years, moving object detection and tracking technology havebeen concerned by more and more scholars. They have very importantsignificance on intelligent monitoring, urban transportation and many otherapplications. Therefore, the research of object detection and tracking has quitehigh scientific and economic value. This paper focuses on how to improve thealgorithm speed and real-time, as well as how to overcome common problemsin practice.In connection with the slow speed and poor robustness of shelter anddeformation in the current moving object tracking algorithm, this paperproposed a moving object tracking method based on the ORB. Firstly, we usedthe Gaussian mixture model to estimate the background. Secondly, thedifference of frame and background was treated as the foreground, then weobtained the objects in the foreground processed by morphological algorithm.Thirdly, the ORB features were extracted in all object areas and were used toestablish feature set for every object. Finally, we found the locations of allobjects and updated feature sets in the current frame. However, when we doORB features match in the object area, there may be many wrong matches.In order to solve this problem, we proposed an improved LMedSalgorithm-PROLMS. The proposed algorithm can exclude the wrong matchesand easily be used. For multi-object tracking, we used the mean-shift algorithmto predict the object position, which can avoid the time-consuming problem ofglobal feature matching. Compared by the similar tracking algorithm CamShift,the ORB-based tracking algorithm can overcome the shortcoming of theCamShift when the object color is similar to the background. It also wascompared with tracking algorithm based on SIFT and SURF. The results showthat the ORB-based algorithm is significantly faster than the above twoalgorithms. Finally, the algorithms were achieved and tested in Visual Studio2010.The results show that the tracking algorithm proposed in this paper canaccurately track the moving target and has a good real-time.