| Moving targets segmentation in video sequence is the key technology of computervision and it is the basis of moving targets tracking, recognition and analysis. Therefore, ithas important research significance and application value. Common methods for movingtargets segmentation, such as background subtraction, optical flow method and inter-framedifference method are aimless and non-selective. Therefore, some non-significant movingtargets are extracted.Thus the task size is increased and the following processing becomesmore difficult. How to segment the ROI completely needs to be deeply investigated. What’smore, this is the important research significance where my subject lays on.The research of this thesis focuses on extracting moving target of interest from videoscaptured by static camera. A new method of ROI segmentation with spatial-temporalsaliency was proposed, which can screen the interested moving target automatically andextract it. Main contribution and innovative points are as following:1. A new method was proposed which combines spatial salient maps based on graphtheory and Markov chain with temporal salient maps based on motion vector and entropy.Static information in the spatial domain and dynamic information in the time domain aredrawn into video object segmentation.2. Post-processing algorithm was proposed to solve the problem that there are falsepoints belonging to background after segmentation. This algorithm uses backgroundsubtraction and morphology dilation operation.The proposed model and method were tested on various kinds of videos collected indifferent scenarios and the results were compared with that of other4popular salientmodels. The result of the experiment validates that the proposed algorithm which bringsspatial and temporal saliency in moving target segmentation can extract the interestedobject effectively while increasing precision ratio and performing well in segmentation. Italso shows that the post-processing algorithm is effective. |
PDF Full Text Request