Fire warning efficiency is based on machine vision technique

Recently, the potential research group of the Department of Information Technology and Military Technical Institute has researched and created a fire detection and warning system based on machine vision technology. This system is effectively applied in harsh observation conditions such as arsenal, arsenal, combustible hazardous materials storage, confidential documents, places where early fire warnings are needed.

Fire - a big disaster

Currently in the world as well as in Vietnam, there are thousands of fires every year, causing great loss of life and economy. Meanwhile, most of the fires that were discovered were fires that appeared, developed long ago (tens of minutes to hours). Therefore, the search and development of methods for early detection of soon-to-burn and small burning areas is really urgent.

Currently, there have been many studies proposing fire detection and warning measures such as using fire detectors, smoke detectors and fire detectors. Fire detection and warning system based on traditional fire detectors has been effective in many situations, warning of accurate fire when temperature, smoke spreads to the fire detector reaching the operating threshold of the device.

However, this method is still limited by the fire detectors only work when the temperature, the smoke has spread to the sensor head and reached the operating threshold, then often the fire has grown large. Therefore, the system is only effective in small and enclosed spaces (as in the room), detecting fire when the fire breaks out not quickly, while the monitoring area has open space such as corridors, closed rooms, photos. If the wind is used, the system will be inefficient. The detection and warning of fire using detectors is also limited because it depends entirely on the time of heat and smoke propagation since the onset of fire until the smoke or temperature spreads to the head. detector.

In recent years, an open direction in fire warning research is to use techniques to recognize photos through camera observation system. Studies on applying image and video processing techniques to real-world camera systems for the purpose of detecting and warning fire have developed strongly and have achieved certain results. Most of the proposed solutions for the fire detection problem use image processing techniques, digital video is now based on the observable properties of the fire such as color, position change. The pixel of the fire over time. However, most studies only stop at analyzing the characteristics of the flame, the results are still at the experimental level, the accuracy is not high; study fireproof materials; Fire fighting method .

Stemming from the aforementioned practice, together with the support from the Ministry of Science and Technology, the research team of the Faculty of Information Technology and the Military Technical Institute conducted a number of studies and achieved the possible results It is important to detect early fire warnings through the analysis of color characteristics, characteristic of the 'shaking' of the fire. The group has excellently completed the project: 'Researching and proposing models, detection and warning solutions based on visual techniques for fire and explosion prevention'.

Artwork: VnExpress

Wide applicability

Dr. Tong Minh Duc, Faculty of Information Technology, Military Technical Academy said that the group aims to research, test models, detection solutions, early fire warning based on images obtained from CCTV , when the fire began to form smoke, small flames, temperatures and smoke did not spread to the camera. At the same time, to build an early fire detection and warning system, the group built a fire video data set; study and build fire detection model, smoke detection model, fire detection and warning model; building fire detection and warning system; experiment; finishing products.

After a period of implementation, the topic has built a mathematical model of the pixel of the flame; color mathematical model of pixels belonging to the smoke cloud; algorithm to identify moving objects with input images obtained from CCTV; flame identification model is based on a combination of color characteristics and flutter of fire; smoke cluster recognition model based on the combination of color characteristics and the movement of smoke smoke; fire detection model in low light environment; Fire detection model with infrared camera; mathematical model recognizes images of flames based on spatial structure, 'rim' structure and 'tops' structure for a still image; model design, help system to detect early fire within a narrow space (in closed rooms) through images from CCTV. The system allows setting up warning parameters suitable for monitoring environment.

The research team has successfully built a fire detection and warning system with CCTV conditions in the room, the protection area is a cone shaped with a = 450 angle and a height of 5m, the working environment temperature from 10-40 degrees Celsius, ordinary fire material; dataset of classification of video clips of fires. The system was tested at the Fire Alarm and Firefighting Laboratory, Department of Automation and Technical Vehicles, University of Fire Protection.

The research topic has proposed models, detection solutions and fire warning using machine vision techniques. The research results of the thesis can be applied in early warning at narrow observation space with conventional fire materials. The system will be effectively applied in the harsh observation conditions such as arsenal of arsenal, military warehouses, combustible hazardous materials storage, confidential documents, where early warning is needed.
The system, if combined with a fire alarm system using traditional fire detectors, will operate more efficiently. In many situations, the traditional fire alarm system is dependent on space, weather, wind direction . the new system can overcome some disadvantages and detect fire early. On the contrary, in situations where the fire is concealed, the density of the smoke cloud is sparse, the traditional system is effective, determining the exact fire.

It can be said that the research model has demonstrated the ability to access machine vision techniques to identify and warn fire to warn early fire in some conditions of narrow space, observation distance, materials Conventional fire is right. According to Dr. Tong Minh Duc, in the coming time, the group will continue to study and coordinate with the University of Fire Protection to study specific conditions and situations, combined with the system of using traditional fire detectors and construction. system, conduct trial production, complete database of fire video images to have a standard data source for research and learning at training institutions to study and research fire protection.