摘要
火灾视觉特征的提取是视觉火灾探测中的关键问题.我们主要研究色彩、纹理以及轮廓脉动等特征的提取,并提出一种度量轮廓脉动信息的距离模型,该模型在规格化的傅立叶描述子空间能够准确地度量这种时空闪烁特征.实验结果表明,该方法具有比较好的鲁棒性,有助于提高视觉火灾探测的准确率、降低误报漏报率.
Based on investigating color, texture and temporal features for vision based fire detection, a distance model of contour fluctuation between two successive frames in the normalized Fourier descriptors domain was presented to measure this time varying contour fluctuation feature of flame. The model of contour fluctuation is effective and robust for fire recognition. To further reduce false alarms, several features extracted according to color, texture and the distance model were together regarded as a joint feature vector for artificial neural network to detect fire. Experiments show that the algorithm is effective and robust, and that it is significant for improving accuracy and reducing false alarms.
基金
国家自然科学基金专项基金(50323005)
中国博士后科学基金(2004036155)资助
关键词
视觉火灾探测
背景模型
傅立叶描述子
计算机视觉
vision based fire detection
background model
Fourier descriptor
computer vision