Method and device for maintaining image background utilizing multiple Gaussian models

Abstract

The invention discloses a method and a device for maintaining image background utilizing multiple Gaussian models. The method comprises the following steps: a. capturing image picture containing a plurality of pixels to acquire background information; b. calculating the background information to build a main Gaussian model; c. capturing a plurality of continuous image pictures at a fixed time interval to acquire image information, and calculating image information to build a sub-Gaussian model; d. repeating step c to build a plurality of sub-Gaussian models; and e. comparing two sub-Gaussian models and judging the image information corresponding to the two sub-Gaussian models, if the judged image information all belong to background information, then updating the learning of the main Gaussian model by the sub-Gaussian model; if one of the judged image information is not at least the background information, then not updating the learning of the main Gaussian model but maintaining the background information of the main Gaussian model.

Description

technical field [0001] The present invention relates to a method and device for maintaining image background, in particular to a method and device for maintaining image background by using multiple Gaussian models for comparison. Background technique [0002] In today's moving object detection technology, in order to correctly identify moving objects even when the background changes, an adaptive background technology is usually used to learn from the background change. Since it is impossible to directly determine whether the input pixels belong to the foreground or the background, generally all the obtained pixels are added to the background model, and then given a corresponding weight value. For example: when the probability of occurrence of the pixel or pixels close to the pixel is relatively frequent, the corresponding weight value will also be relatively large. The weight value can be judged according to the critical value to divide the pixels into two categories, the l...

AI Technical Summary

Problems solved by technology

[0006] In summary, using the above two methods, it is still impossible to accurately and effectively solve the misjudgment of background learning caused by the foreground being too long or the foreground edge and background color being too close to maintaining the background image.

Because the foreground exists for too long, the weight will increase accordingly, and it may be misjudged as the background

However, if the edge color close to the background is judged to belong to the Gaussian distribution of the background, the average value of the Gaussian distribution will be changed, so that the background color is similar to the foreground, resulting in misjudgment.

Images