Flame detection based on temporal-spatial block covariance matrix fusion features

A flame recognition method is proposed based on video image´s color and other spatial-temporal blending features in CMYK´s space. It demarcates the video frame into spatial-temporal cubic block and moves characteristics into integrated covariance matrix descriptor features. It firstly presents a feature selecting method by evaluating corrective value distribution corresponding each two correlation coefficient of a covariance matrix. It puts forward a feature selection method by analyzing change trend of Riemannian Manifolds distance of covariance matrix. The flame feature and classification method are analyzed and compared by covariance metric using Riemannian Manifolds distance and Support Vector Machine(SVM) classifier. It is demonstrated by experiments that the flame detection system based on covariance matrix hierarchical feature has higher accuracy and recognition efficacy.