Video Flame Detection Algorithm Based on Improved GMM and Multi-Feature Fusion

Chi Zhang; Qinghao Meng; Tao Jing

doi:10.3788/LOP202158.0410006

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 4 >Page 0410006 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 4, 0410006 (2021)

Video Flame Detection Algorithm Based on Improved GMM and Multi-Feature Fusion

Chi Zhang, Qinghao Meng, and Tao Jing^*

Author Affiliations

Institute of Robotics and Autonomous Systems, Tianjin Key Laboratory of Process Detection and Control, School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China

show less

DOI: 10.3788/LOP202158.0410006 Cite this Article Set citation alerts

Chi Zhang, Qinghao Meng, Tao Jing. Video Flame Detection Algorithm Based on Improved GMM and Multi-Feature Fusion[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0410006 Copy Citation Text

show less

Fig. 1. Detection flowchart based on improved GMM and multi-feature fusion

Download full size

Fig. 2. Frame capture of flame video sequence. (a) Indoor flame; (b) outdoor flame; (c) forest flame

Download full size

Fig. 3. Original and foreground renderings.(a)(b) Original image; (c)(d) traditional GMM foreground extraction map; (e)(f) improved GMM foreground extraction map

Download full size

Fig. 4. Feature extraction flowchart

Download full size

Fig. 5. Original images and LBP histograms. (a)(b) Flame images; (c)(d) fire-like images; (e) histogram extracted from Fig. (a); (f) histogram extracted from Fig. (b); (g) histogram of extracted from Fig. (c); (h) histogram of extracted from Fig. (d)

Download full size

Fig. 6. Experimental videos. (a) Outdoor flame; (b) indoor flame; (c) forest flame; (d) walking pedestrians; (e) flashing car lights; (f) flashing neon lights

Download full size

Algorithm	Fig. 2(a)	Fig. 2(b)	Fig. 2(c)	Average value
GMM	44.5	47.8	49.3	47.2
Improved GMM	33.7	30.4	31.6	31.9

Table 1. Comparison of algorithm processing speed unit: ms/frame

Algorithm	Fig. 2(a)	Fig. 2(b)	Fig. 2(c)	Average
GMM	1233	674	521	809
TBGMM	426	587	403	472
IGMM	269	391	137	266

Table 2. Comparison of false alarms

Algorithm	Fig. 2(a)	Fig. 2(b)	Fig. 2(c)	Average
GMM	563	601	731	632
TBGMM	411	398	759	523
IGMM	518	489	501	503

Table 3. Comparison of missed inspections

Target	1	2	3	4	5	6	7	Average value
Flame	0.7020	0.7764	0.6986	0.7356	0.7204	0.8016	0.7932	0.7468
Sunrise	0.5075	0.5108	0.5596	0.5845	0.5655	0.6019	0.4912	0.5459
Headlight	0.2301	0.2422	0.2499	0.3047	0.3881	0.1963	0.2943	0.2722
Candle flame	0.9457	0.8536	0.8983	0.8203	0.8915	0.8294	0.8231	0.8660

Table 4. Target similarity comparison

Video	Videoframe	Fireframe	Ref. [6]		Ref. [9]		Ref. [5]		Proposed method
Video	Videoframe	Fireframe	MF	TP /%	MF	TP /%	MF	TP /%	MF	TP /%
Fig. 6(a)	208	198	6	92.31	9	90.87	10	90.38	4	93.27
Fig. 6(b)	368	350	5	93.75	4	94.02	8	92.93	2	94.57
Fig. 6(c)	217	199	14	85.25	1	91.24	15	84.79	6	88.94
Average	--	--	--	90.44	--	92.04	--	89.37	--	92.26

Table 5. Comparison of flame video detection results

Video	Videoframe	Fireframe	Ref. [6]		Ref. [9]		Ref. [5]		Proposed method
Video	Videoframe	Fireframe	FF	FN /%	FF	FN /%	FF	FN /%	FF	FN /%
Fig. 6(d)	148	0	5	3.38	3	2.03	4	2.70	3	2.03
Fig. 6(e)	187	0	9	4.81	12	6.42	10	5.35	3	1.60
Fig. 6(f)	273	0	15	5.49	5	1.83	18	6.59	10	3.66
Average	--	--	--	4.56	--	3.43	--	4.88	--	2.43

Table 6. Comparison of non-fire video detection results

Chi Zhang, Qinghao Meng, Tao Jing. Video Flame Detection Algorithm Based on Improved GMM and Multi-Feature Fusion[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0410006

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享