基于时空局部二值模式的火灾视频识别

Fire video recognition based on volume local binary pattern

为了获得较高识别率,引入时空局部二值模式,用以识别火灾视频。先以时空局部二值模式提取视频中的表观特征和运动特征后,再利用支持向量机对多种场景的火灾视频进行分类识别。针对白天火灾、夜晚火灾、室内火灾、室外火灾、森林火灾五种场景进行测试实验,结果显示,对夜晚火灾视频和室外火灾视频识别率可达到100%,对白天火灾、室内火灾、森林火灾视频的识别率分别为94.117 6%、95.238 1%、94.444 4%,这表明所提方法有效,其识别率不易受场景光照条件或复杂背景影响,具有鲁棒性。

In order to get higher recognition rate for fire video, an approach based on Volume Local Binary Pattern （VLBP） is proposed. Firstly, the features of fire video are modeled with VLBP by combining appearance and motion. Secondly, Support Vector Machine （SVM） is used to classify the fire video under various scenes. Test experiments results for five scenes of daytime fire, nighttime fire, indoor fire, outdoor fire and forest fire show that the recognition rate for nighttime fire and outdoor fire can reach 100~, and the recognition rate of daytime fire, indoor fire, forest fire are 94. 117 6~, 95. 238 1%, 94. 444 4%, respectively. It shows that the proposed algorithm can be used to classify fire video and also the recognition rate will not be influenced by the changes of the light condition and the complex background. This shows that the algorithm is robust.