Smoke detection method

Abstract

The invention discloses a smoke detection method. The smoke detection method comprises the following steps of: video stream access: reading a video stream from a front-end camera device or a local video file, and decoding the accessed video stream to obtain a frame video image; video image preprocessing: preprocessing the acquired video image by using an image preprocessing technology to improve the accuracy of later-stage smoke detection; detection of a moving object based on background modeling; target classification based on a deep convolution neural network; and video smoke target calibration: recording the position of an image block determined to be smoke, and marking the area of the smoke in the video. The method integrates the advantages of a deep convolution neural network and a background difference method, so that the smoke is accurately and real-timely detected and positioned, the false detection rate is effectively reduced, the smoke in the video can be early warned in realtime, and the method has an important significance and an application value.

Description

technical field [0001] The invention relates to the technical field of visual detection, in particular to a smoke detection method. Background technique [0002] Human beings have struggled with fire for a long time, accumulated a lot of experience in fire prevention and fire extinguishing, gradually mastered the combustion mechanism, combustion conditions and development process of fire, and created various fire prevention and fire extinguishing methods. According to changes in physical quantities such as smoke, gas, and temperature, various smoke detectors are designed using temperature sensing, smoke sensing, light sensing, gas detection, and composite detection. However, the smoke detector is limited by the space due to its own characteristics, and has the following disadvantages: 1) it needs to be detected after the smoke arrives; 2) in a large indoor environment, the smoke will be cooled by the surrounding air after rising to a certain height, and even Staying in the ...

AI Technical Summary

Problems solved by technology

[0004] However, in the prior art, the frame difference method is used to detect moving objects on video images, and feature detection or feature fusion detection is performed on features such as the area, color, direction or shape of the moving object. This feature-based detection method is limited The shape of the smoke, and the color, shape, area, and direction of the smoke are very unstable, and it is difficult to extract its features, resulting in low detection accuracy

Images