Review on Smoke Detection Algorithms for Video Images

Changyou Chen; Jiansheng Yang

doi:10.3788/LOP202158.0400003

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 4 >Page 0400003 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 4, 0400003 (2021)

Review on Smoke Detection Algorithms for Video Images

Changyou Chen and Jiansheng Yang^*

Author Affiliations

College of Electrical Engineering, Guizhou University, Guiyang, Guizhou 550025, China

show less

DOI: 10.3788/LOP202158.0400003 Cite this Article Set citation alerts

Changyou Chen, Jiansheng Yang. Review on Smoke Detection Algorithms for Video Images[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0400003 Copy Citation Text

show less

Fig. 1. Flow chart of smoke detection for video images

Download full size

Fig. 2. Discrete diagram for estimating smoke motion direction ^[52]

Download full size

Fig. 3. Examples for smoke detection under different scenes^[15]

Download full size

Fig. 4. Smoke detection based on multi-feature fusion

Download full size

Metric parameter	Formula
Mean	μ= $\frac{1}{N_{1}} \overset{N_{1} - 1}{\sum_{a = 0}}$ x₁(a)
Uniformity	U= $\overset{N_{1} - 1}{\sum_{a = 0}} x_{1}^{2}$ (a)
Entropy	E=- $\overset{N_{1} - 1}{\sum_{a = 0}}$ Q_a(ln Q_a)
Standard deviation	σ= $\sqrt[]{\frac{1}{N_{1}} \overset{N_{1} - 1}{\sum_{a = 0}} {[x_{1} (a) - μ]}^{2}}$

Table 1. Formulas for calculating statistical metric parameters of texture features

Parameter	Formula
Energy	A_ASM= $\sum_{e} \sum_{f}$ P(e,f)²
Contrast	C_CON= $\sum_{e} \sum_{f}$ (e-f)²P(e,f)
Correlation	C_COR= $\frac{\sum_{e} \sum_{f} (e -) (f -) P (e, f)}{σ_{x} σ_{y}}$
Homogeneity	I_IDM= $\sum_{e} \sum_{f} \frac{1}{1 + {(e - f)}^{2}}$ P(e,f)

Table 2. Statistical parameters for characterizing GLCM^[32]

Method	Pre-processing	Smoke feature	Classification	Accuracy/%	Falsepositive/%	Falsenegative/%
Method inRef. [57]	Color segmentation technique	Texture, energy, and motion	Neural network(NN)	80	-	-
Method inRef. [15]	Color segmentation technique	Energy and texture	SVM	96.29	3.71	1.74
Method inRef. [5]	Motion segmentation technique	Color , fuzzy feature, shape irregularity, and motion	PSO-SVM	97.16	2.84	-
Method inRef. [52]	Motion segmentation technique	Color, texture, energy, HOG feature, shape irregularity, and motion	AdaBoost-RBFSVM	91.25	0.31	-
Method inRef. [67]	Motion segmentation technique	Shape irregularity, diffusion property, and texture	ANFIS-Bayes-MR	99.09	1.94	-

Table 3. Smoke detection algorithm for video images

Reference	Network model	Accuracy/%	False positive/%
Method in Ref. [89]	Deep normalization and CNN(14 layers)	97.52	0.6
Method in Ref. [90]	RCNN+3D CNN	95.23	0.39
Method in Ref. [91]	CNN(AlexNet)	99.44	0.44
Method in Ref. [92]	CaffeNet CNN	98	1.7
Method in Ref. [93]	DN-CNN	96.37	0.60
Method in Ref. [87]	YOLOv2	94.7	5.2
Method in Ref. [87]	GMM+YOLOv2	98.1	0.8

Table 4. Smoke detection results for video images based on different deep learning network models

Changyou Chen, Jiansheng Yang. Review on Smoke Detection Algorithms for Video Images[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0400003

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information