Experimental Results

The proposed method, Method 1, is implemented on a PC with an Intel Pentium 4, 2.40 GHz processor. It is tested for a large variety of conditions in comparison with the method utilizing only the color and temporal variation information, which we call Method 2, described in [19]. The scheme described in [5] is also implemented for comparison and it is called as Method 3 in the rest of the article. The results for some of the test sequences are presented in Table 1.

Method 2 is successful in determining fire and does not recognize stationary fire-colored objects such as the sun as fire. However, it gives false alarms when the fire-colored ordinary objects start to move, as in the case of a realistic scenario. An example of this is shown in Fig.  (a). The proposed method does not give any false alarms for this case (Fig.  (b)). The fire-colored strip on the cargo truck triggers an alarm in Method 2 when the truck starts to move. Similarly, false alarms are issued with Method 2 in Movies 3, 7 and 9, although there are no fires taking place in these videos. The moving arm of a man is detected as fire in Movie 7 (Fig.  (c)), and a red parking car is marked as fire in Movie 9 with Method 2 (Fig.  (d)).

Method 3 gives similar detection results for fire. However, it also suffers from inefficient analysis of the motion of fire colored objects. Fire-colored ordinary moving objects causes Method 3 to give false alarms in Movies 1, 3, 7 and 9. If Method 1 is used, moving fire-colored ordinary objects do not cause an alarm to be raised. This is because the cyclic movement of flames is taken into account in our method, as well as the spatial variation in the color/brightness values of the moving fire-colored regions. Method 1 successfully detects fire in videos covering various scenarios, including partial occlusion of the flame. Sample images showing the detected regions are presented in Fig. .

In Movie 11, a man wearing a fire-colored shirt intentionally waves his arms to mimic the quasi-periodic flicker behavior in flames. Although all of the methods produce false alarms in this Movie, Method 1 significantly decreases the number of false positives relative to Methods 2 and 3.

These methods are also compared to each other in terms of computational cost (as shown in Table 2). Movies in Tables 1 and 2 are all captured at 10 fps with a frame size of 320 by 240 pixels. The average processing times per frame are 16.5 msec, 12.5 msec and 14.5 msec, for our method, Method 2, and Method 3, respectively. Our method is computationally more demanding due to additional wavelet analysis based steps. Since only shift and add type operations take place when convolving signals with the wavelet filters, additional cost is not high. Our implementation works in real-time for videos with frame size 320 by 240 pixels, captured at 10 fps or higher in a PC.

The video clips that we tested our method contain a total of 83,745 frames in 61 sequences. In 19 of the sequences fire takes place. Our method is successful in detecting fire in all of these sequences. This corresponds to a fire detection rate of 1.0. A fire contour recognition rate of $0.999$ is reported in [15] which corresponds to a fire detection rate of $0.999$. Our overall false alarm (false positive) rate is 0.001. It is reported that non-fire contour recognition rate is 1.0 in [15] which corresponds to a false alarm rate of 0. The video sequences containing fire in [15] are not publicly available. Therefore we used our own data set. We also test our method with the data set of the EU funded CAVIAR project/IST 2001 37540, publicly available at URL: http://homepages.inf.ed.ac.uk/rbf/CAVIAR/. Although there are a lot of clips with moving fire-colored objects, none of the clips in this data set contains fire. Our method gives no false alarms in any of these sequences.