An Illumination Sensitive Video Moving Object Segmentation Method

Abstract

The invention provides a photosensitive video moving object segmentation method. The photosensitive video moving object segmentation method comprises steps that a novel eight-direction sensitive filter (OSF) is provided, and a contribution value of a neighborhood pixel is calculated from the eight directions of the pixel, and can be used as a local descriptor to measure interblock similarity; for an input original video, a plurality of continuous frames of the video are processed in a segmented manner; the OSF is introduced in a variation model, and a foreground object area is calculated, and therefore a foreground seed point and a background seed point are acquired; the OSF descriptor is used for tracking the object, and is used to update the positions of the seed points, and finally, the OSF is introduced in image cutting, and a foreground object is acquired by adopting the seed point segmentation. The OSF has an illumination constant characteristic quantity, which is used as the local descriptor to be introduced in the segmentation method provided by the invention, and robustness is still provided for a condition, the illumination of which is changed, in a video scene.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to an illumination-sensitive video moving target segmentation method. Background technique [0002] As one of the most active research topics in image and video processing, video object extraction has been concerned by many related researchers. Video object segmentation aims to extract the foreground moving object segmentation in the image sequence. Reasonable video moving target segmentation results are the basis for many other tasks, including target detection and recognition, video compression, and robot vision, and further video editing and processing applications. It can use computer resources more efficiently and further promotes these researches. develop. [0003] For the video foreground moving target extraction algorithm, the existing popular technologies and algorithms are generally based on the motion estimation method, that is, the position of the m...

AI Technical Summary

Problems solved by technology

However, most of these methods are not sensitive enough to sharp lighting changes or smooth surfaces due to shadow changes, and the robustness of the algorithm is low.

At the same time, the large displacement caused by the fast movement of the object or the fast operation of the camera is also a huge challenge to the extraction of moving objects.

In addition, the algorithm based on model optimization often takes about ten seconds to process a frame, while the execution time of the algorithm based on geometric model or space-time consistency optimization will be longer, and the time efficiency is not high enough.

Images