A VLSI-ORIENTED AND POWER-EFFICIENT APPROACH FOR

DYNAMIC TEXTURE RECOGNITION APPLIED TO SMOKE

DETECTION

Jorge Fern´andez-Berni, Ricardo Carmona-Gal´an and Luis Carranza-Gonz´alez

Institute of Microelectronics of Seville (IMSE-CNM)

Consejo Superior de Investigaciones Cient´ıﬁcas y Universidad de Sevilla

Avda. Reina Mercedes s/n 41012, Seville, Spain

Keywords:

Dynamic texture recognition, Power-efﬁcient VLSI implementation, Smoke detection, Forest ﬁre detection.

Abstract:

The recognition of dynamic textures is fundamental in processing image sequences as they are very common

in natural scenes. The computation of the optic ﬂow is the most popular method to detect, segment and analyse

dynamic textures. For weak dynamic textures, this method is specially adequate. However, for strong dynamic

textures, it implies heavy computational load and therefore an important energy consumption. In this paper,

we propose a novel approach intented to be implemented by very low-power integrated vision devices. It

is based on a simple and ﬂexible computation at the focal plane implemented by power-efﬁcient hardware.

The ﬁrst stages of the processing are dedicated to remove redundant spatial information in order to obtain

a simpliﬁed representation of the original scene. This simpliﬁed representation can be used by subsequent

digital processing stages to ﬁnally decide about the presence and evolution of a certain dynamic texture in the

scene. As an application of the proposed approach, we present the preliminary results of smoke detection for

the development of a forest ﬁre detection system based on a wireless vision sensor network.

1 INTRODUCTION

A temporal texture or dynamic texture (DT) is a

spatially-repetitive time-varying visual pattern whose

temporal variation presents certain stationarity (Nel-

son and Polana, 1992). An additional feature of a

DT is its indeterminate spatial and temporal extent.

Smoke, waves, a ﬂock of birds or tree leaves swaying

in the wind are some examples of DTs.

The recognition of DTs plays an essential role in

image processing as they are very common in natu-

ral scenes. Different methods have been proposed to

realize this process of recognition, as described in a

recent review (Chetverikov and P´eteri, 2005). The

methods based on optic ﬂow are currently the most

popular. Optic ﬂow is a computationally efﬁcient and

natural way to characterise the local dynamics of a

temporal texture. Specially, this is the case for weak

dynamic textures, that is, textures deﬁned by a lo-

cal moving coordinate system in which they become

static. However, the recognition of strong dynamic

textures implies a much greater computational effort.

For these textures, possessing intrinsic dynamics, the

brightness constancy assumption associated to stan-

dard optical ﬂow algorithms cannot be applied. More

complex approaches must be considered in order to

overcome this problem. Recently, interesting results

have been achieved by applying the so-called bright-

ness conservation assumption (Amiaz et al., 2007).

However, this method means heavy computational

load and the subsequent high energy consumption.

There are certain systems where a power-efﬁcient

implementation of DT recognition is totally manda-

tory. Wireless multimedia sensor networks (Akyildiz

et al., 2007) is an obvious example. These networks

are composed of a large number of low-power sen-

sors that are densely deployed throughout a region

of interest in order to capture and analyse video, au-

dio and environmental data from their surroundings.

The massive and scattered deployment of these sen-

sors makes them quite difﬁcult to service and main-

tain. Therefore, energy efﬁciency must be a major de-

sign goal of this kind of systems in order to prolong

the lifetime of the batteries as much as possible.

In this paper, we propose a novel approach to de-

tect DTs in scenes surveilled by very low-power inte-

grated vision devices. It is based on a signal process-

ing architecture where redundant spatial information

307

Fernandez-Berni J., Carmona-Galán R. and Carranza-González L. (2009).

A VLSI-ORIENTED AND POWER-EFFICIENT APPROACH FOR DYNAMIC TEXTURE RECOGNITION APPLIED TO SMOKE DETECTION.

In Proceedings of the Fourth International Conference on Computer Vision Theory and Applications, pages 307-314

DOI: 10.5220/0001766903070314

 SciTePress

is removedat the very ﬁrst stages of the processing by

means of a simple, ﬂexible and power-efﬁcient com-

putation at the focal plane. The ﬁnal decision about

the presence of a certain DT in the scene is realized

by analysing the reduced representation of the origi-

nal data obtained from the focal plane. Thus, the com-

putational load is greatly alleviated.

As an application of this approach, we study the

smoke detection within a wireless vision sensor net-

work devotedto forestﬁre detection. The results point

to a very high reliability and robustness in the process

of detection.

2 BINNING PROCESS

In general, existing research on DT recognition is

based on global features computed over the whole

scene. A clear sign of this fact is that practically all

of the sequences composing the reference database

DynTex (P´eteri et al., 2006) contain only close-ups of

DTs. For these sequences, it does make sense to apply

strategies of global feature recognition over the whole

scene. However, there are interesting applications of

DT recognition, e.g. video-surveillance, where tex-

tures can appear at any location of the scene. In this

case, a previous detection and subsequent analysis of

candidate regions to contain a certain DT could re-

duce the computational load by progressively reduc-

ing the amount of data to process. To detect such can-

didate regions, we can take advantage of the spatial

repeatability of patterns in DTs. In this way, we are

going to divide the scene into blocks, or bins, whose

size S

is deﬁned, in pixels, as:

= W × H (1)

The fundamental concept of the proposed ap-

proach is that each bin of the scene, labelled as (i, j),

can be considered as an independent entity capable

of detecting the presence of a determined spatial pat-

tern within it. When a bin detects the spatial pattern,

it is marked as a candidate bin to contain a part of a

DT whose spatially-repetitive pattern coincides with

the detected pattern. A subsequent phase of spatio-

temporal analysis of the candidate bins will eventu-

ally conﬁrm or dismiss the presence of the DT. Note

that, during this phase, the amount of data to be pro-

cessed is reduced by a factor W × H with respect to

the previous phase of candidate bin detection.

Every bin will be represented by only one value.

This value must be deﬁned in such a way that the pres-

ence of a certain spatial pattern within it can be eas-

ily detected. We propose to deﬁne the representative

value of the bins in terms of spatial frequency infor-

mation. For example, consider the detection of a ﬂock

of birds passing along the sky. A ﬂock of birds in con-

trast with sky entails signiﬁcant information at spatial

frequencies greater than 0, which expresses the origin

of coordinates in the Fourier space. Let us suppose

that each bin (i, j) of the scene is represented by the

following value:

(k > 0)

(k)

(2)

where P

(k > 0) is the power at the bin (i, j) for all

of the spatial frequencies other than k = 0 and P

(k)

is the power at the bin (i, j) for all the spatial fre-

quencies including k = 0. The bidimensional vector

k represents the 2π-normalized wavenumber vector.

Applying this expression, the recognition of a spatial

pattern similar to that of a ﬂock of birds is straightfor-

ward. In Fig. 1 a grayscale binned image of a ﬂock of

birds is depicted along with its representation using

Eq. 2. It can be seen that even a recognition of zones

with different density of birds can be carried out from

the value of B

. A global pixel-based analysis is not

necessary.

Figure 1: Example of processing to detect candidate bins.

This example allows to understand how the detec-

tion of DTs can be simpliﬁed by means of a binning

process. However, a thorough analysis to establish an

optimum value of S

has not been realized. It is an es-

sential aspect of the proposed approach. The value of

must be adequate to detect a certain spatial pattern

as well as to track its temporal dynamics across the

scene. Therefore, it will depend not only on the spa-

tial pattern but also on the spatio-temporal dynamics

VISAPP 2009 - International Conference on Computer Vision Theory and Applications

308

of the texture. Later in this paper we describe the tun-

ning of S

, along with other parameters, for smoke

detection within the framework of forest ﬁre detec-

tion.

3 SIGNAL PROCESSING

ARCHITECTURE

The only element missing now is an efﬁcient compu-

tation of the spatial frequency components present in

each and every bin. Thus, moving part of the heaviest

computational effort to the focal plane would result in

a ﬂexible and power-efﬁcient architecture. This ap-

proach is inherent to biological vision systems, like

the human retina, where visual information is cap-

tured and preprocessed, alleviating the data ﬂow sent

to the visual cortex (Roska and Werblin, 2001). In

this way, low-level tasks, according to the classiﬁca-

tion in (Pirsch and Stolberg, 1998), whose require-

ments of accuracy are not too demanding, are realized

in parallel by highly efﬁcient, although moderately

coarse, analog hardware at the focal plane. Physi-

cal implementations based on this processing archi-

tecture achieve higher performance with less cost and

power (Carmona et al., 2003). In our case, we are go-

ing to convey the detection of candidate bins to the

focal plane. A microprocessor would make the ﬁnal

decision about the presence or absence of the DT by

analysing the preprocessed and therefore reduced in-

formation represented by these candidate bins. Note

that this processing architecture simply removes re-

dundant data at the focal plane in order to deliver to

the microprocessor just the necessary information to

track and detect the texture.

At thispoint, we need to deﬁne a ﬂexible hardware

structure at the focal plane in order to be able to detect

any DT. Such a structure must satisfy two conditions:

1. The size of the bins S

can take any value

2. Information about any particular band of spatial

frequencies can be extracted at every bin

A structure fulﬁlling both conditions is depicted

in Fig. 2. It consists of a M × N grid where the value

of each pixel is stored in a capacitor. These capaci-

tors are 4-connected to the neighboring capacitors by

means of analog switches. These switches are con-

trolled by the corresponding row or column selection

signal. When selected, i. e. the control signal is high,

the switch behaves as a resistor connecting the two

nodes, whose utility will be explained in short. If the

control signal is low the switch is in the boundary of

a bin. Thus the particular distribution of 0’s and 1’s

in the set of row and column selection signals estab-

lishes the size and amount of bins in which the image

plane is divided.

Figure 2: Processing structure at the focal plane.

Once S

is determined, a resistive grid is estab-

lished within every bin taking into account the resis-

tance of the switches when they are on. Such a re-

sistive grid can carry a linear diffusion of the pixel

values within the bin as long as the switches remain

ON, let us say a period of time ∆t. Consider y

(x, y)

as a function deﬁning the values of the pixels within

the bin (i, j). The linear diffusion can be expressed as

(Jahne et al., 1999):

(k, ∆t) = Y

(k, 0)e

−4π

D∆t|k|

(3)

where Y

(k, ∆t) is the spatial Fourier transform of the

subimage contained in the bin after ∆t seconds and

(k, 0) is the transform at time t = 0, that is, just

before starting the diffusion. The constant D is the

diffusion coefﬁcient.

From Eq. 3, the power of each frequency compo-

nent during the process of diffusion is:

(k, ∆t)|

= |Y

(k, 0)|

−8π

D∆t|k|

(4)

which can be expressed as:

(k, ∆t)|

= [1 − α(k, ∆t)]|Y

(k, 0)|

(5)

where α(k, ∆t) is the attenuation undergone by each

component at frequency k after ∆t seconds of linear

A VLSI-ORIENTED AND POWER-EFFICIENT APPROACH FOR DYNAMIC TEXTURE RECOGNITION APPLIED

TO SMOKE DETECTION

309

diffusion. This attenuation increases along time and

is more signiﬁcant for higher frequencies.

Therefore, by controlling ∆t we can obtain infor-

mation at different bands of spatial frequencies. Con-

sider the simplest case, a lowpass ﬁlter. We can spec-

ify a maximum attenuation for the passband α

, with

a cut-off spatial frequency of |k

|. At the same time,

for the stopband, a minimum attenuation α

is re-

quired starting at frequency |k

|. Then, the period of

time after which the linear diffusion must be stopped

in order to carry out the so deﬁned ﬁltering is between

these two bonds:

|ln(1− α

8π

D|k

≤ ∆t ≤

|ln(1− α

8π

D|k

(6)

Then, providing the means for storing and com-

bining two consecutive samples of the diffused im-

age, taken after ∆t

and ∆t

from the starting point of

the diffusion, we can compute a bandpass, a highpass

or a bandreject ﬁlter. Keep in mind, however, that the

highest frequency feature that can be consideredis de-

termined by the size of the pixel, and therefore a real

highpass ﬁlter is actually a bandpass ﬁlter.

A crucial property of the structure just deﬁned is

that the diffusion operation is really a simple charge

redistribution which does not consume energy, i. e. it

is realized by a passive network. That is to say, the

signal processing within the structure just described

is massively parallel and ultra power-efﬁcient.

4 AN APPLICATION TO SMOKE

DETECTION

As an application of the approach previously ex-

plained, we present a vision algorithm suitable for a

forest ﬁre detection system based on a wireless vision

sensor network. The vision sensors will be placed on

top of poles in order to focus the canopy of small veg-

etation areas. Each sensor of the network will run

on-site a vision algorithm in order to detect smoke

arising among the vegetation. When a sensor detects

smoke, a warning message is sent to a control center

by multihopping. This structure of the system, based

on a careful placing of the sensors, reduces signiﬁ-

cantly the sources of false alarms with respect to its

counterparts based on lookout towers with automatic

surveillance. On the contrary, because of the neces-

sary dense deployment of vision sensors, energy efﬁ-

ciency is a key point impacting the system cost.

The smoke detection algorithm is based on the

analysis of the sequence of images captured by each

sensor. We deﬁne the time interval between two con-

secutively captured images as T

. These images are

comparedto a reference image, the background, in or-

der to detect changes generated by smoke dynamics.

This method of motion detection, called background

subtraction (Hu et al., 2004), is suitable for scenes

with a relatively static background. In the case of the

proposed system, the visual ﬁeld is basically com-

posed of vegetation, and therefore the background

will hardly suffer signiﬁcant sudden changes. It will

experience, however, gradual illumination changes

throughout the day. Taking this fact into account, the

reference image will be updated every time interval

Regarding the binning process, note that the ap-

pearance of smoke in any scene means the equaliza-

tion of the R, G and B components in the pixels af-

fected by smoke (Chen et al., 2006). Therefore, it

could be a criterion to detect candidate bins. How-

ever, we propose a more data-efﬁcient option based

only on the detection of sudden increases in the B

component with respect to the background. The use

of the B component is owing to its greater sensitiv-

ity in natural scenarios with vegetation to the changes

generated by smoke when compared to the R and G

components and combined luminance. This will also

discard changes introduced by sources different than

smoke, like the motion of tree leaves.

Consider Fig. 3. We have highlighted different

zones in three scenes where the background is mainly

constituted by vegetation. These frames correspond

to different parts of the associated sequences, before

and after a trace of smoke has appeared in the ﬁeld of

view. Table 1 shows the normalized average increase,

referred to the image without smoke, undergone by

each RGB component and the combined luminance

of the pixels within the marked zones in presence of

smoke. It can be seen that, in all cases, the appearance

of smoke amongvegetation conveysa greater increase

in the B component than that observed in the R and G

components and the combined luminance.

Therefore, each bin will be represented by the av-

erage value of the B component of its pixels. In terms

of spatial frequencies, it is equivalent to say that we

are interested in the information contained at k = 0.

The value of S

will depend on the average variation

in the number of smoke pixels during their appear-

ance in the scene: the less the average variation, the

smaller the necessary size of the bins to track the dy-

namics. Consider a sequence of k consecutive images

containing smoke captured by a sensor. The average

variation in the number of smoke pixels will be:

V =

∑

i=1

| U

[i] −U

[i− 1] |

(7)

where U

[i] is the total number of smoke pixels of the

VISAPP 2009 - International Conference on Computer Vision Theory and Applications

310

(a) (b) (c)

Figure 3: Three typical scenes with vegetation background without smoke and in presence of smoke.

Table 1: Normalized average increase, referred to the case without smoke, suffered by each component in the zones marked

in Fig. 3.

R component G component B component Luminance

Scene (a) 13.8% 14.5% 20% 14.9%

Scene (b) 12.5% 13.4% 19.5% 13.8%

Scene (c) 11% 12.8% 16.4% 12.7%

i−th image of the sequence (supposeU

[0] = 0). Ac-

cording to this expression, the smoke dynamics repre-

sented by these images can be correctly tracked when-

ever S

fulﬁlls the following condition:

≤

V (8)

Finally, the condition which must be satisﬁed in

order to consider a foreground bin B

as a candidate

bin to contain smoke is:

− B

100

MAX

− B

MIN

) (9)

where p

represents the threshold percentage of the

B component signal range, set by (B

MAX

− B

MIN

which the B component of a foregroundbin B

must

rise with respect to the corresponding background bin

Once the candidate bins are detected, the algo-

rithm analyses them looking for the spatio-temporal

dynamics which is characteristic of smoke. This stage

of the algorithm is divided into two phases: the detec-

tion and the conﬁrmation phase. The detection phase

starts when the ﬁrst candidate bins are discovered, an

instant denoted as t

, and ﬁnishes at t = t

. Then the

conﬁrmation phase is started, and will ﬁnish, if the

result is positive, at time t = t

, by sending an alarm

message. The internal processing at both phases is

described next.

First of all, in order to consider that smoke is

present at the scene, a minimum number of candi-

date bins must exist. Let us deﬁne N(t) as the number

of candidate bins at time instant t. This parameter

changes every T

, that is, with every captured image.

During the conﬁrmation phase, the following expres-

sion must be fulﬁlled:

N(t) ≥ N

MIN

{t ∈ [t

]}

(10)

where N

MIN

represents the minimum necessary num-

ber of candidate bins to consider smoke.

Another important element of the smoke dynam-

ics is its gradual appearance into the scene. Once the

ﬁrst candidate bins are detected, new candidate bins

must gradually appear until reaching at least N

MIN

t = t

. This fact can be described by means of two

conditions. The ﬁrst one is:

− t

≤ T

MAX

(11)

where T

MAX

represents the maximum time interval

within which smoke must appear into the scene once

the ﬁrst candidate bins are detected. The second con-

dition is:

N(t) − N(t − T

) ≤ G

MAX

{t ∈ [t

]}

(12)

where G

MAX

expresses the maximum permitted

growth of candidate bins between two consecutive

captured images during the smoke dynamics.

Finally, smoke does not appear as candidate bins

scattered throughout the scene. On the contrary, it

A VLSI-ORIENTED AND POWER-EFFICIENT APPROACH FOR DYNAMIC TEXTURE RECOGNITION APPLIED

TO SMOKE DETECTION

311

is formed by compact regions of candidate bins. Let

us deﬁne Z(t) as the number of 8-connected regions

of candidate bins. Just like N(t), Z(t) changes with

every captured image. The compactnessof smoke can

be described as:

Z(t) ≤ Z

MAX

{t ∈ [t

]}

(13)

being Z

MAX

the maximum permitted number of 8-

connected regions during the smoke dynamics.

The duration of the conﬁrmation phase will be

ﬁxed, that is:

− t

= T

(14)

Therefore, the spatio-temporal dynamics of

smoke can be summarised as follows: a minimum

number of candidate bins N

MIN

must appear once the

ﬁrst candidate bins are discovered at t = t

. The time

instant at which N

MIN

is reached, t = t

, establishes

the end of the detection phase and the beginning of the

conﬁrmation phase. Besides, t

must fulﬁll Eq. (11),

accounting in this way for the time scale of the grad-

ual appearance of smoke in the scene. If the limit es-

tablished by T

MAX

is reached without having accom-

plished the minimum candidate bins ﬁxed by N

MIN

then the process starts again as the detection was trig-

gered by something that was not smoke. If the de-

tection is correct conﬁrmation phase starts. During

this phase, whose duration is determined by Eq. (14),

the number of candidate bins must be always above

MIN

, i. e. they must satisfy Eq. (10). In this way,

we take into account that smoke does not disappear

suddenly from the scene. Finally, the growth rate of

smoke, deﬁned in Eq. (12), and the compactness of

the smoke traces in the scene, representedby Eq. (13),

are checked during both the detection phase and the

conﬁrmation phase. Failure in holding these condi-

tions means tracking some dynamics belonging to a

source different from smoke, what stops the process

returning to the beginning of the detection cycle.

5 PRELIMINARY TESTS

In order to test the vision algorithm, we realized some

recordings in natural scenarios. They were carried

out in 9 different locations under different illumina-

tion conditions using three different cameras. Ap-

proximately 80 minutes were recorded containing 16

sequences of gradual appearance of smoke following

its natural evolution in scenes whose background is

basically composed of vegetation and numerous se-

quences without smoke in order to check the false

alarm rate. The resolution of the frames is 720× 576

px and the frame rate is 25 frames per second in all

the recordings. As an example, the images in Fig. 3

correspond to three different tests.

To set the values of the parameters of the algo-

rithm, we analysed 9 of the 16 different sequences

where smoke appears gradually (Fern´andez-Berni,

2008). The other 7 sequences will be employed to test

the algorithm later. Speciﬁcally, 25 seconds of every

sequence were analysed. The starting point is just be-

fore smoke begins to appear. Only the B component

of the frames was used for the adjustment.

The proposed algorithm presents a basic parame-

ter: T

. It establishes the temporal scale of the smoke

dynamics. In order to determine its value, we ﬁrst de-

tected the pixels affected by smoke. Indeed, smoke

is the most signiﬁcant foreground motion in the 9 se-

quences analysed. Therefore, we can consider that a

pixel represents smoke if:

(x, y) − p

(x, y) > ε (15)

where p

(x, y) is the foreground value of the pixel

(x, y), p

(x, y) is its background value and ε is the

pixel noise of the camera.

Once the smoke pixels are detected, the parame-

ter T

must reﬂect the temporal scale of its dynam-

ics. At the same time, this parameter must be as large

as possible to reduce the processing load. Consider

Fig. 4. It shows a magnitude that is very sensitive

to the smoke dynamics: the number of smoke pix-

els per frame of every sequence. If we examine the

evolution of these curves in the frequency domain, it

can be seen that most of the power is concentrated at

very low frequencies. In fact, at f = 0.5Hz the power

of the DFT is approximately three orders of magni-

tude smaller than the power at f = 0Hz for all the

sequences. This means that the essential characteris-

tics of the dynamics can be tracked by only analysing

the frequency interval [0, 0.5]Hz. In terms of T

, it is

translated into T

= 1s, that is, a sample frequency of

1Hz. Therefore, the adjusment of the rest of parame-

ters will be realized taking into account that T

= 1s.

To determine S

, we represent

V for each se-

quence in Fig. 5. It can be seen that the minimum

value is approximately 200 smoke pixels per second.

According to Eq. (8), this is the maximum number of

pixels which must contain every bin in order to track

the smoke dynamics. Taking into account the resolu-

tion of the frames, the size of the bins must be 15× 12

pixels (180 pixels). It is important to emphasize that,

thanks to the binning process, the initial information

composed of 0.4Mpx is reduced to only 2.3Kbins .

Besides, in this case, every bin can be represented by

only one bit indicating if it is marked as a candidate

bin or not.

VISAPP 2009 - International Conference on Computer Vision Theory and Applications

312

0 5 10 15 20 25

x 10

Time (seconds)

Number of smoke pixels

Figure 4: Number of smoke pixels per frame.

1 2 3 4 5 6 7 8 9

200

400

600

800

1000

1200

1400

1600

1800

2000

2200

2400

2600

2800

3000

3200

3400

Sequence

Figure 5: Average variation in the number of smoke pixels

for each sequence.

Regarding the percentage p

, consider Fig. 6. It

represents the average increase, normalized to the sig-

nal range, of the pixels affected by smoke with regard

to the background. It can be seen that the average

increase exceeds 10% for most of the time. We are

going to extrapolate this result to the candidate bins

by setting p

= 10%.

0 5 10 15 20 25

Time (seconds)

Average increase (%)

Figure 6: Normalized average increase of the pixels affected

by smoke.

Applying Eq. (9) with the values of S

and p

just

set, we obtain the number of candidate bins of the se-

quences, as depicted in Fig. 7. According to this rep-

resentation, the minimum number of candidate bins

reached in the sequences is 17. We set N

MIN

= 14

in order to concede a margin of three candidate bins.

This choice implicitly sets T

MAX

= 20s and T

= 4s.

Therefore, the detection of smoke will be carried out

within a maximum interval of 24s after the discovery

of the ﬁrst candidate bins.

0 5 10 15 20 25

100

110

120

130

140

150

160

170

180

190

200

210

220

230

240

250

260

270

280

290

300

Time (seconds)

N(t)

Figure 7: Number of candidate bins at every second of the

sequences.

At this point, it is easy to adjust the value of

MAX

. Consider Fig. 8, where the maximum growth

rate of candidate bins is represented for each se-

quence. Among these maximum rates, the greatest

is 47 bins per second. We set G

MAX

= 50, adding a

small margin of three bins.

1 2 3 4 5 6 7 8 9

Sequence

Maximum growth rate (candidate bins per second)

Figure 8: Maximum growth rate of candidate bins in each

sequence.

There are two parameters left: Z

MAX

and T

. To

determine Z

MAX

, consider the Fig. 9 where the num-

ber of 8-connected regions of candidate bins is de-

picted throughout the sequences. We can see that the

maximum value is 6. Besides, it is only reached once

in one of the sequences. Therefore, we set Z

MAX

= 6,

without additional margin.

Finally, with respect to T

, longer recordings are

mandatory in order to estimate an approximate value

of this parameter. Even for the longest sequence anal-

ysed, whose duration is 458s (around 8 minutes),

smoke detection was possible without updating the

initial reference image.

A VLSI-ORIENTED AND POWER-EFFICIENT APPROACH FOR DYNAMIC TEXTURE RECOGNITION APPLIED

TO SMOKE DETECTION

313

0 5 10 15 20 25

Time (seconds)

Z(t)

Figure 9: Number of 8-connected regions of candidate bins

during the sequences.

Once the parameters were set, we applied the al-

gorithm, to all the recordings. Smoke was detected

in the 16 smoke sequences, the 9 employed to set the

parameters, obviously, and the other 7. Besides, in

the rest of sequences where smoke was not present,

no false alarm was detected. Therefore, the algorithm

achieves the highest reliability when applied to all our

recordings.

6 CONCLUSIONS

A new generic approach to detect dynamic textures

has been presented. It is specially suitable for real-

time applications where the power requirements de-

mand very low energy consumption. As an applica-

tion, we propose the smoke detection within a forest

ﬁre detection system based on a wireless vision sen-

sor network. A set of recordings was carried out in

natural scenarios in order to validate the feasibility of

the proposed approach for this application. The re-

sults point to a very high reliability and robustness

in the process of detection. Our primary next objec-

tive is the physical implementation of a vision chip

whose focal plane includes the structure proposed in

this paper. We also intend to extend the applicabil-

ity of the proposed approach by deﬁning new vision

algorithmswhich can take signiﬁcant advantage of in-

cluding both binning and spatial ﬁltering within each

bin.

ACKNOWLEDGEMENTS

This work is funded by Junta de Andaluc´ıa (CICE)

through project 2006-TIC-2352.

REFERENCES

Akyildiz, I., Melodia, T., and Chowdhury, K. (2007). A

survey on wireless multimedia sensor networks. Com-

puter Networks, 51(4):921–960.

Amiaz, T., Fazekas, S., Chetverikov, D., and Kiryati, N.

(2007). Detecting regions of dynamic texture. In

International Conference on Scale Space and Varia-

tional Methods in Computer Vision (SSVM’07), pages

848–859.

Carmona, R., J´ımenez-Garrido, F., Dom´ınguez-Castro, R.,

Espejo, S., Roska, T., Rekeczky, C., and Rodr´ıguez-

V´azquez, A. (2003). A bio-inspired 2-layer mixed-

signal ﬂexible programmable chip for early vision.

IEEE Transactions on Neural Networks, 14(5):1313–

1336.

Chen, T., Yin, Y., Huang, S., and Ye, Y. (2006). The

smoke detection for early ﬁre-alarming system based

on video processing. In IEEE International Confer-

ence on Intelligent Information Hiding and Multime-

dia Signal Processing (IIH-MSP’06), pages 427–430,

California, USA.

Chetverikov, D. and P´eteri, R. (2005). A brief survey of dy-

namic texture description and recognition. In Interna-

tional Conference on Computer Recognition Systems

(CORES’05), pages 17–26, Rydzyna Castle, Poland.

Fern´andez-Berni, J. (2008). Video database:

Smoke propagation in natural scenarios.

http://www.imse.cnm.es/∼berni/.

Hu, W., Tan, T., Wang, L., and Maybank, S. (2004). A

survey on visual surveillance of object motion and be-

haviors. IEEE Transactions on Systems, Man. and Cy-

bernetics, 34(3):334–352.

Jahne, B., Hauβecker, H., and Geiβler, P. (1999). Hand-

book of Computer Vision and Applications, volume 2,

chapter 4. Academic Press.

Nelson, R. and Polana, R. (1992). Qualitative recognition

of motion using temporal texture. CVGIP: Image Un-

derstanding, 56(1):78–89.

P´eteri, R., Huskies, M., and Fazekas, S. (2006). Dyn-

tex: A comprehensive database of dynamic textures.

http://www.cwi.nl/projects/dyntex/.

Pirsch, P. and Stolberg, H. (1998). VLSI implementations

of image and video multimedia processing systems.

IEEE Transactions on Circuits and Systems for Video

Technology, 8(7):878–891.

Roska, B. and Werblin, F. (2001). Vertical interactions

across ten parallel, stacked representations in the

mammalian retina. Nature, 410:583–587.

VISAPP 2009 - International Conference on Computer Vision Theory and Applications

314