摘要：Video smoke detection plays a significant role in real-time fire alarms. In this paper, aiming at the problems of high false positives and high false negatives in complex environments, a video smoke detection algorithm based on multi-feature fusion and modified random forest (RF) is proposed. Firstly, the motion region is detected by ViBe. The noise and holes are processed by morphology. Then, the candidate smoke region is filtered in HSV color space. LBMP, wavelet energy, and smoke growth rate features are extracted from the candidate area. Finally, the multi-features are fused and input into RF classifier for smoke detection. The experimental results show that the modified RF selects the decision trees with high AUC and small correlation through ranking and clustering, which has higher detection accuracy. Furthermore, the proposed method combines the static and dynamic characteristics of smoke, which improves the detection speed effectively and has good robustness. It can avoid high false positives and high false negatives in complex environments.
关键词：modified random forest; ViBe; HSV color space; LBMP
