An accurate 3-D fire location method based on sub-pixel edge detection and non-parametric stereo matching

Fire disasters will cause severe damage to human properties and bring terrible mental and physical injury if they cannot be detected and extinguished in time. As traditional fire detectors, usually acting as alarms, fail to automatically locate fires and extinguish fires, fire detection and location based on binocular stereo vision has attracted much attention recently. But current three-dimensional (3-D) location methods based on binocular stereo vision have less accuracy with regards fire location due to imprecise camera calibration and unstable stereo matching, a novel 3-D location method based on sub-pixel edge detection and non-parametric stereo matching technology was explored. Firstly, to improve the camera calibration accuracy, Zernike moments operator was applied to relocate the sub-pixel edges from the feature points detected by Canny operator. Secondly, regional matching combined with epipolar constraint was proposed for fire stereo matching. Epipolar constraint was used to reduce the search area from two dimensions to one. And Census transform based on non-parametric transform was employed for exact regional matching. The experimental results indicated that the proposed method had high performance in 3-D fire location with high accuracy and strong robustness. An automatic fire-fighting system based on the proposed fire location method was developed and successfully applied to a coal to chemical plant in Yunnan province, China. (C) 2014 Elsevier Ltd. All rights reserved.