Fast Smoke Detection from Videos using CUDA
Smoke detection is key component of preventing disaster and fire accident. Over a last decade smoke has been detected by using various sensors like infrared sensor, ion sensor and optical sensors. Such detection approach sometimes give false alarm and have low accuracy and low response rate. This paper presents a smoke detection technique from video by using color, shape and motion characteristics of the smoke. For this Background subtraction is performed by using Gaussian Mixture Model, calculate the color probability of moving object, histogram of optical flow. Some part of the smoke detection algorithm is performed by using CUDA GPUs for both low resolution and high resolution videos.
Keywords - Color Probability, Gaussian Mixture Model (GMM), Smoke Detection, CUDA, GPGPU