Study On The Robust Visual Tracking Algorithm In Complex Scenes

Visual tracking technology in complex scenes is one of the important research directions in computer vision.It has many fields such as image processing,pattern recognition and artificial intelligence.It has a wide range of applications in academic research and practical application.So it has important research value.In this paper,we have studied and analyzed the robust visual tracking technology in the case of the still camera based on the horizontal project in the indoor office scene.The main work has the following aspects:(1)For the indoor office environment in this complex scene under the motion target detection problem,in this paper,the ViBE algorithm,the dense optical flow method and the background subtraction method based on the Gaussian mixture model are studied and analyzed in detail.The three methods are experimentally compared with each other.The Gaussian mixture model with equal performance is used as the final motion target detection algorithm.At the same time,in order to overcome the shadows caused by indoor lighting,the shadow detection algorithm is integrated into the moving target detection module,which effectively solves the foreground noise problem caused by shadows.But also for the follow-up tracking can be sustained and sustained,the target object will be tracked back to the motion target detection in the background update module,effectively avoids the problem that the tracked target is incorporated into the background model due to stillness.(2)Fixed camera with a 45-degree angle overlooking the shelves in the height of 3-4 meters of indoor office environment,is used to real-time monitoring.Indoor office staff close to each other,in the video image,the head will not be blocked or only a slight occlusion.Therefore,the Adaboost head classifier is incorporated into the visual tracking framework,which effectively avoids the problem of human occlusion.In order to make the head classifier be able to work steadily,the characteristic optical flow tracking and Hough circle detection module are introduced to visual tracking framework,and they are complementary to the head classifier.(3)Aiming at the problem of visual tracking in complex scene,the visual tracking algorithm based on adaptive color attribute is adopted,which makes the system can perform real-time processing on the monitoring video,and has certain robustness to the indoor illumination change.In order to solve the problem of mutual occlusion between targets,a new mechanism for judging occlusion and solutions is proposed,so that the tracking can be carried out steadily and continuously.