Reducing false alarm of video-based smoke detection by support vector machine

Techniques used in video smoke detection systems have been discussed noticeably in past few years. With the advantage of early fire alarm in large or specific spaces such as studio and tunnels, the video-based smoke detection systems would not have time delay as conventional detectors. In contrast, how to reduce false alarm and increase the generalization ability is the key issue for such state-of-the-art systems. In this paper, examples consisting of features extracted from a real time video are collected for the training of a discriminating model. A prototype of support vector machine (SVM) is therefore introduced for the discriminating model with the capability in small sample size training and the good generalization ability. In order to reduce the false alarm, the prototype is then extended to a class-imbalanced learning model to deal with rarity of the positive class. A number of assuming data are used for imbalanced test to cope with the real world of fire safety. The technique is optimistic to enhance accuracy and reduce false alarm in video-based smoke systems.