Article Preview
TopIntroduction
The current trends of research have drawn attention to the research area that is combined for both fire detection and human detection and trying to develop new technology for saving lives caused by the fire. The evolution in technologies in the field of computer vision (Szeliski, 2010) led to much new advancement in digital image processing. Schalkoff (2020) presented an idea to supplant the traditional and typical fire detection alarms with computer vision-based systems using image processing. It comprises three basic steps: the classification of the fire pixel, segmentation of moving objects, and then analyzing candidates region as the algorithm is based on image processing then there is less chance of raising false alarms. To distinguish moving pixels from non-moving pixels, a backdrop subtraction breakthrough with a frame differencing algorithm is applied to the frame buffer filled with successive frames of the input video. To raise a fire alarm, the moving pixels that are also detected as fire pixels are further studied in subsequent frames.
It is difficult to detect humans due to the differences in outfits, motion background, illumination, clutter, noise, and other factors that add to the complexity of human detection. There are so many algorithms exist like (Viola & Jones, 2001) for human identification, which are accurate enough to detect humans but had a few limitations that cause failure in real- time operation as stuck into high false detection rate (Dalal and Triggs, 2005). Later, a new approach called Histogram of Oriented Gradient was developed to extract features from local cells. It is unaffected by geometric and photometric changes (HOG) except for object orientation (Dalal and Triggs, 2005). On the other hand, Viola-Jones is reliant on geometric modification yet produces better results and is commonly used for facial recognition and detection. (Barnouti et al., 2016) applied this technology in Autobots and surveillance to save lives and avoid fatalities, the technology may be advanced and used much more efficiently.
Due to unexpected failures in the intended deliverables of the present fire security management system, our idea of detecting humans in fire using modern machine learning algorithms forms its shape. The objective of this work is to serve a great purpose of saving lives, although currently, this work has some limited boundaries it has a great future scope ahead if it gets incorporated with modern tools like robots and drones. The proposed model in this work, is a result of a complete cycle of collecting data, training a model and testing over multiple examples to achieve better efficiency at each step.
Fires are posing an increasingly severe threat to people's lives and property. For the detection of fire, a combination of methods of motion detection and color detection of the flame was used as preprocessing step. In screening the fire candidate pixels, this strategy saves a significant amount of computing time. Second, despite its irregularity, the flame has a significant resemblance to the image's sequence. Many fire detection sensors are available nowadays to quickly detect fire and trigger fire alarms. Still, they can also generate false alarms as in case of cigarette smoke which generate chemical particles in the air that the sensors pick up. In order to overcome from this problem, computer vision and image processing based methodology was used to identify fire using the YCbCr color standard and to recognize humans in real-time, the Histogram of Oriented Gradient (HOG) and Support Vector Machine (SVM) based classifier was utilized. This strategy, however, can be improved by incorporating this technology into a robot that can rescue persons trapped in a fire and save them from destruction. The proposed methodology in this work has two phases, one is detection of fire more precisely than the conventional systems available and other is majorly focusing on rescuing human beings stuck in a fire, which is very common to observe. Experimental results show that it could improve the accuracy and reduce the false alarm rate when compared to conventional method.