Research On Multi Pedestrian Tracking Method Based On Sparse Representation

Multi object detection and tracking, as one of the key and difficult points in the study of computer vision, although it has achieved some results and applications, but it still has a long way to go, there are still some difficult problems not solved. The change of the scenes is one of the difficult points. Different light, occlusion, the position relationship between different targets and the observation point of view will affect the algorithm and have a great impact on the results of the calculation. Thus, there may be a matching failure and other errors. In recent years, the emergence of sparse representation theory can said to be a huge theoretical breakthrough in the field of signal processing.For the sparse signal collected, we can reconstruct the original data signal from the low dimension sampling value using the sparse representation method, which can resist the strong interference. In this paper, we mainly use the pedestrians in the dynamic scene as the object of tracking, and we use sparse representation algorithm to track multiple pedestrians, which can solve the problem of multi object matching, occlusion, etc.The main research contents and innovations of this paper are as follows:(1)Acording to the large proportion of background, this paper mainly uses a background modeling algorithm based on RGB color information, which can effectively distinguish the different parts of the scene, which can overcome the shortcomings of the traditional modeling methods to a certain extent.(2)In the detection and localization of multi pedestrian targets, the characteristics of the interference factors in the image are analyzed, such as the shadow, the light mutation and the target area division. The texture of the shadow and the light mutation region is similar to that of the background. In this paper, we use the normalized cross correlation method to detect and remove the shadow and the light mutation. By using the multiple features of the split connected domainto determine whether a target area can be split.The twi-difference algorithm is used to merge the split target, so that it can accurately detect and locate multi pedestrian targets.(3)In the aspect of multi pedestrian target tracking, the theory of sparse representation is mainly used. Under the guidance of this theory, the problem of multiple target tracking is transformed into sparse representation reconstruction problem of sparse signal. The modeling of sparse representation observation matrix is based on the color and texture features of objects. A twice reconstruction matching algorithm is proposed to track the trajectory of multiple targets. Thus, solve the problem of occlusion and splitting between multiple targets. A new method for the online update of the observation matrix is designed, which can not only adapt to the change of the target observation feature, but also improve the accuracy of the reconstruction of the target.