FIREWATCHER is a smart wildfire smoke detection solution to analyze
multi-channel CCTV videos with the use of deep-learning based video analysis technology
and to take initial measures for extinguishing wildfire whose smoke is detected in CCTV videos.
Expectancy effect
-
Applied technology
- Simultaneous analysis on multi-channel CCTV videos
- Deep-learning model with the dataset of Korean mountain conditions
- Division into a total of 10 detection classes including smoke, fog, and cloud
-
Work / Function
- Short messaging service (SMS) when wildfire smoke is detected
- 24-hour real-time monitoring through CCTV video analysis
- Estimation of the real position of the smoke detection area in video
-
Management efficiency
- System state monitoring
- Event history and statistics
- CCTV and admin (for SMS) information registration
Main functions
-
Multi-channel
CCTV video analysis -
AI based object
identification and recognition
(smoke, cloud, and
steam classification) -
Real-time alarms
for events -
Immediately initial
measures for
fire/wildfire
Solution features
-
01
Early detection of fire through the detection of ‘smoke’, rather than ‘fire’
-
02
Deep-learning model of minimizing errors, such as the detection of cloud and fog looking like smoke
-
03
Real-time simultaneous analysis and monitoring of multiple channels
Solution environment
S/W spec.
Server
OS : Ubuntu 20.04 LTSDB : MongoDB v3.6.3
Client
OS : Windows 10 Pro 64 bitBrowser : Chrome 102.0 or higher
H/W spec.
Server
CPU : Intel(R) Xeon(R) Gold 6334 CPU@ 3.60GHz or moreGPU : Nvidia A10 (GPU Memory 24GB) or more
RAM : 64GB DDR4 2933Mhz or more
Client
CPU : Intel(R) Core(TM) i7-4702MQ CPU2.20 GHz or moreRAM : 8GB or more
HDD : 500GB or more
Solution components
Main display
-
Main display
- CCTV channel and event information
- Event statistics data
- Display of public weather environment data
- System sever operation information (CPU temperature)
- Event & history management
- Display of GIS based camera position information
-
Real-time monitoring for wildfire smoke detection
- Minimization of errors through the division into a total of 10 classes including smoke, fog, and cloud
- Deep-learning model with the dataset of Korean mountain conditions
- 24-hour real-time monitoring through CCTV video analysis for smoke detection
