A foreground extraction method based on random clustering statistics of resolution reduction

Abstract

The invention discloses a foreground extraction method based on random clustering statistics of resolution reduction. The method comprises the following steps: step 1, respectively compressing video images of different scene data sets at multiple compression ratios; Step 2, respectively processing the obtained images with different compression ratios by using ViBe and GMM methods, and recording CPU processing time, precision rate, recall ratio information and F indexes under each compression ratio as selection reference data of the compression ratio; Step 3, selecting a corresponding compression ratio for compression based on the reference data according to the target monitoring scene, and then utilizing a ViBe method and a GMM method for processing; 4, performing AND operation on processing results of the ViBe and GMM methods, and complementing the incompleteness of the two foreground contours; And 5, carrying out contour search on the fused result, and drawing a contour to eliminatea cavity in the target. According to the method, the foreground extraction accuracy is improved, and meanwhile, the processing speed is greatly increased.

Description

technical field [0001] The invention relates to the technical field of video processing, in particular to a foreground extraction method based on random clustering statistics with reduced resolution. Background technique [0002] As an important application direction of computer vision, intelligent video surveillance has received extensive attention in recent years. Foreground extraction is one of the key steps in intelligent video surveillance, and its performance is crucial for later applications such as object classification, object tracking, and behavior understanding. At present, there are many common foreground extraction methods. Considering the complexity of the method, the amount of calculation, the performance effect and other factors, the background subtraction method is selected for foreground extraction, mainly including the GMM method and the ViBe method. [0003] The GMM method uses multiple Gaussian distributions to describe the background model. Use a mixe...

AI Technical Summary

Problems solved by technology

[0005] However, the speed of the foreground extraction method is directly related to the resolution of the video or image acquisition equipment. When more and more high-definition probes are used in the monitoring scene and information of interest is extracted from a large number of monitoring videos, the speed problem is A major obstacle to application

In the prior art, most background subtraction methods are based on grayscale. When the foreground and background grayscales are close, the extracted target has some phenomena such as missing contours or internal holes. This is an inherent defect of the background subtraction method.

Images