Research On Video Object Segmentation With Pixel-based Background Modeling

Video object segmentation,with which object regions can be extracted from video sequences,is widely used in various kinds of video processing application.It's an essential and basic technology in computer vision.Video object segmentation usually acts as pre-processing step for technologies such as video object recognition,object tracking,content-based humancomputer interaction,video conferencing,intelligent traffic system,intelligent security and so on.The research on video object segmentation is of great significance.In recent years,video object segmentation is a hot topic in the field of computer vision,for which many scholars put forward numerous ideas.The present methods,such as PBAS,have achieved some results,but there are still some problems which are difficult to be solved.When the background has a strong dynamic change,the accuracy of the segmentation will decline sharply.The ghost artifacts caused by intermittently moving objects will affect the subsequent segmentation results and accuracy.When there is shadow,the foreground segmentation will be affected by the shadow which results in great error.The real-time performance is difficult to meet the needs of practical application.In order to tackle the above problems,this paper puts forward an improved pixel-based background modeling method for video object segmentation.The paper's main work are as follows.a)In this paper,we have made a comprehensive and deep research on the existing methods of image segmentation and video object segmentation,and analyzed the advantages and disadvantages of these existing methods.First of all,the image segmentation algorithm is introduced,which is the basis of the research of video object segmentation.Then,the general methods of video object segmentation are introduced.The pixel-based video object segmentation methods are mainly researched.b)In order to tackle the problem in threshold response and adjustment a nonlinear threshold adjustment method is applied to accelerate process,which improves the accuracy in complex scenes.c)Aiming at removing the ghost artifacts,a foreground region life cycle supervise method is proposed to remove the ghost area which is difficult to be eliminated,a re-identification process is applied to accelerate the process.d)In the light of the effect of shadow,a method based on invariant color features is introduced.The foreground,background and shadow regions are distinguished with normal color feature and invariant color features.e)To tackle the real-time performance problem,the super-pixel segmentation is fused into the video segmentation process to improve the efficiency.The PBAS method is improved in four aspects.The experiment results on Change detection net 2014 shows that the improved method is more robust to complex scenes,ghost artifacts and shadow with higher accuracy in foreground segmentation and overall segmentation.