Research On Smoke Detection Algorithm Based On Low Illumination Indoor Video Image

Smoke detection based on video image is a new type of fire detection technology in firefighting field.In view of the video smoke detection,most of the researches at home and abroad in the light of the better environment,in low-light conditions,although the study is still in a budding stage,the prevention of fire has a very important guiding significance.The fire in low-light conditions not only causes more casualties,but also causes a serious harm to public property.The main work of this paper is to study the smoke images of low-light indoor videos,and to detect and identify the smoke targets in the images to realize the fire warning under the conditions.In low-light indoor conditions,images got by infrared cameras have much random noise,poor visual effects and other bad characteristic.In order to eliminate the noise and highlight the image target,a new modified denoising algorithm based on the adaptive median filtering algorithm for low-light video images is presented in this paper.In order to reduce the impact of low-light indoor light changes on target detection and to obtain target images that have less noise,the detection of the smoke target is achieved by improving the learning rate in the hybrid Gaussian model method.After getting the smoke target,opening operation is performed firstly and then expansion operation is processed to fill the inside hole of the target and remove the small pieces of meaningless area of the target images.After obtaining the accurate and complete target,in order to distinguish the smoke target,the area diffusion characteristic,the movement direction characteristic and the texture characteristic are studied respectively.For the area diffusion feature,the average value of the absolute area variation rate is characterized to denote the diffusion.For the movement direction characteristic,it is characterized by the ratio of the main moving direction of the center,the center is the smallest peripheral rectangle of the target contour.For the texture feature,it is characterized by the contrast and the angular second moment of the rectangle smoke area in the original images.In order to improve the recognition rate of smoke targets,the support vector machine based on radial basis function is used to fusion and judge the features of the smoke target.In order to test the performance of the proposed algorithm,the algorithm is carried out in different scene,different monitoring distance,different indoor illumination,different smoke concentration and different interference.The test results show that the smoke recognition algorithm can distinguish smoke and disturbances in most cases,the smoke detection is high,the anti-interference is stronger and the ability to adapt the scene is better.