Video detection method for early smoke of field fire

Abstract

The invention relates to a video detection method for early smoke of field fire, and the method comprises the following steps: 1, carrying out the decompression of a video shot, and obtaining a framesequence; 2, carrying out the filtering of an original image through a 5*5 mean filter, and removing a noise point through a morphological method; 3, extracting fire smoke candidate regions through the movement characteristics of fire smoke, and extracting a movement foreground through a Gaussian hybrid model; 4, extracting fire smoke candidate regions for the second time through the static characteristics of the early smoke color of the field fire; 5, extracting fire smoke candidate regions for the third time through the direction characteristics of movement in a smoke diffusion process; 6, carrying out the recognition of the fire smoke candidate regions through a convolution neural network, wherein the convolution neural network is designed to have one input layer, three convolution layers, three pooling layers, one full-connection neural network layer and one output layer. The method is high in recognition rate, and facilitates the popularization and application.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for early smoke video detection of wild fires. Background technique [0002] Fire is one of the main dangerous disasters affecting people's production and life. Every year, a large number of wildfires burn down forests and grasslands, bringing huge ecological, economic and social disasters. Early detection and timely response are important ways to reduce losses. At present, the detection of wild fires in forests, grasslands and other fields is mainly through artificial timed observation of observation platforms, and some of the video cameras installed are only used for data collection, without intelligent detection functions. Therefore, when the fire develops to be detectable, the burning area is often large, and the best fighting period is missed, resulting in huge economic losses and even casualties. [0003] Traditional fire detection technology is not sui...

AI Technical Summary

Problems solved by technology

2. Fluid canonical swirling and helical properties of wildfire smoke are not detectable

Images