A Low Cost Machine Vision System for Real-Time Fire Detection
|
 |
|
Post a Comment
|
|
|
|
|
CONTRIBUTORS:
|
|
|
JOURNAL:
|
|
|
YEAR:
|
2007
|
|
PUB TYPE:
|
Journal Article
|
|
SUBJECT(S):
|
machine vision; real-time fire detection; image processing; fire physics; fire spectral characteristics; fire temporal characteristics; fire spatial characteristics
|
|
DISCIPLINE:
|
Engineering and Applied Sciences
|
|
HTTP:
|
http://www.ics.uplb.edu.ph/node/226
|
|
LANGUAGE:
|
English
|
|
PUB ID:
|
103-444-126
(Last edited on
2008/07/19 02:12:47 GMT-6)
|
|
SPONSOR(S):
|
|
|
ABSTRACT:
Progress in fire detection technologies has been substantial over the last decade due to advances in sensor, microelectronics and information technologies, as well as a greater understanding of fire physics. However, acquiring systems that integrate these technologies is still costly, while most of these systems cater only to fire events that are not common locally. We designed, implemented and tested a system for real-time detection of fire for a wide range of local fire and false alarm events using robust algorithms that integrate state-of-the-art software technologies and running under cheap off-the-shelf hardware. The spectral, spatial and temporal properties of fire events are automatically extracted by the system using color video streams captured from a cheap USB-mounted web camera. The use of color video streams has significant advantages over the traditional ultraviolet and infrared fire detectors due to the former’s improved detection and fewer false alarms, while additional descriptive information about fire location, size, and growth rate can be obtained. We used the color probability density of fire pixels to represent the spectral model of fire events. We “trained” a machine vision algorithm by creating, normalizing, and thresholding the color histogram of collected video sequences of fire to produce a color look-up table that will determine the fire-colored pixels. Our spatial and temporal models respectively capture the spatial structure and the temporal signature of a fire region. The ratio between the intersection and difference of fire-colored pixels in consecutive video frames served as a criterion for deciding if the group of fire-colored pixels possess the fire’s spatial and temporal behavior. This criterion can be adjusted to improve detection under a specific environment. The system uses an audio stream to output alarm signals of varying loudness appropriate for the detected rate of fire growth. Additionally, the system can record the detected fire events to help decision makers on how to avoid future fire damages and to aid arson and forensic investigators. We tested our system under different local indoor and outdoor fire events consisting of thousands of image frames. Our system detected real fire events and ignored non-fire events 84% of the time. The system detected no-fire events as fire events (i.e., false alarms) 12% of the time and ignored real fire events 4% of the time. The ignored fire events, however, are from controlled fire such as the blue flame from a torch welder and a motion-less flame from a gas stove. Based on our tests, our vision-based fire detection system from off-the-shelf hardware can be a cheap yet flexible alternative to traditional ones.
|
|
|
|
STATISTICS
|
|
Click on # to view
|
|
Citations
|
|
0
|
|
References
|
|
0
|
|
Comments
|
|
0
|
|
Quality
|
|
0/0.00
|
|
Interest
|
|
0/0.00
|
|
View(er)s
|
|
1/75
|
|
|
|
|
|
|
| Prev |
Next |
|