Remnant detection method

Abstract

The invention relates to a remnant detection method. The remnant detection method is characterized by comprising the following steps: (1), obtaining an image acquired at the front end according to an image photographed by a camera; (2), extracting foreground targets of the obtained image; (3), extracting suspicious remnant targets according to the characteristics of the foreground targets; (4), performing characteristic analysis of the extracted suspicious targets, and removing non-remnant targets; and (5), performing further characteristic analysis of the screened targets, and determining the remnant targets. The remnant detection method disclosed by the invention has the benefits that the problems of being slow in detection speed, high in complexity and the like due to a double-background model can be avoided; the remnant detection method can deal with a certain degree of target occlusion; therefore, leak detection of remnants can be reduced; false detection due to the problems, such as shadow and illumination, can be reduced; false detection due to the fact that an object in a scene moves can be eliminated; the problem that the false report and leak report rate is high in detection of the remnants can be solved; and the remnant detection accuracy is improved.

Description

(1) Technical field [0001] The invention belongs to the technical fields of video monitoring, video image processing and analysis, and machine vision, and in particular relates to a method for detecting remnants. (2) Background technology [0002] The anti-terrorism situation in the world today is severe. In order to ensure public safety, it is an imminent task to prevent terrorist attacks in public places such as banks, airports, subways, exhibition halls, gymnasiums, railway stations, shopping malls, etc., and the terrorist attacks in these places are mainly based on There are many ways to package bombs, so the detection of leftover packages in these places is an indispensable function of the intelligent video surveillance system. [0003] There are two main types of existing remnant detection methods: methods based on object tracking and methods based on object detection. [0004] The method based on target tracking: first detect the foreground target, track each target ...

AI Technical Summary

Problems solved by technology

Method 1: Based on the dual-background modeling method, this method establishes two background models with different update rates, and obtains two types of foregrounds through the background difference method, including fast-moving targets and slow or temporarily stationary targets, and then screens out the remaining target; the disadvantage of this method is that the establishment of the double-background model has high computational complexity and slow detection speed, which is not suitable for real-time legacy detection

Method 2: Perform background modeling to determine the target area, and generate suspicious targets according to the existence time of the foreground pixels in the target area, extract the object mask according to the binary image of the suspicious target, and then analyze the characteristics of the image of the current scene, using the current scene image The degree of correlation with the characteristics of the extracted mask image determines whether it is a remnant; the problem with this method is that it does not further distinguish whether the target of the final alarm is the object being left behind or the object originally in the scene being moved

Method 3: Comparing the background modeling method based on local update with the foreground object obtained by the improved three-frame difference method, segmenting and obtaining temporary static object clusters in the scene, and using the centroid distance method to count the static time of each cluster , to further judge and screen out the remnants of the stationary clusters that reach the time threshold; the disadvantage of this method is that it is easily affected by pedestrian occlusion when counting the static time of the clusters, so that the statistical static time does not reach the set time threshold. After a certain period of time, the residues are integrated into the background model and will no longer be detected, resulting in missed detection

Images