tchsieh / earlyfiredetection Goto Github PK

                            VISION-BASED FIRE DETECTION USING VIDEO SEQUENCES

Abstract

Fire, if improperly used, could pose great threats to peoples’ security, life, and property. Motivated by the requirement to detect fire at its early stage, we aimed to develop an automatic system for vision-based fire detection using video sequences. Our system included four major steps, namely image preprocessing, foreground region analysis, fire dynamic behavior analysis, and fire flow energy analysis. Overall, our system could achieve the detection rates of over 91% in either indoor or outdoor environments. In addition, our system could achieve the system response time within 1 second (average delay of ~25 frames) once the fire occurred. In summary, our system could be used in surveillance systems, leading to prevent damage to peoples’ security, life, and property.

System software environment

Microsoft Visual Studio C++ 2010 With OpenCV2.4.6

Conclusion

We proposed a vision-based fire detection system using video sequences which could potentially be used for early detection of fire. Based on the assumptions of fire properties and dynamics (e.g., color, region, Eddies effect, etc.), our system was designed using image/video processing techniques and included four major steps: image preprocessing, foreground region analysis, fire dynamic behavior analysis, and fire flow energy analysis. Our experiment results showed that our system had been evaluated using a video database containing fire and/or non-fire scenarios. Overall, our system could achieve the detection rates of over 91% in either indoor or outdoor environments. In addition, our system could achieve the system response time within 1 second (average delay of ~25 frames) once the fire occurred. Therefore, the fire detection techniques being developed could be quite effective than traditional fire detectors (e.g., temperature, smoke detectors, etc.)
Despite the high accuracy yielded by our system, our system may still generate false alarms when a moving object does mimic the image properties of fire. However, we anticipate that these situations are rather rare in real scenarios. In this paper, we have developed fire-detection methods (algorithms) and have presented a fire detection system that could be used to automatically detect fire at its early stage. In addition, we have carefully evaluated the fire dynamic behavior with respect to its image characteristics in real scenarios. In summary, our system could be used in conjunction with other fire detectors (e.g., smoke detectors and the like), leading to prevent damage to peoples’ security, life, and property.

For more information please visit "VISION-BASED FIRE DETECTION USING VIDEO SEQUENCES.pdf".