Infrared 7

Data Integration by considering spatial diversity in an IR barrier Fernando Álvarez

Juan Jesús García, Jesús Ureña, Manuel Mazo, Cristina Losada, Álvaro Hernández, Carlos De Marziani Department of Electronics. University of Alcalá Alcalá de Henares. Madrid. Spain {jesus, urena, mazo, losada, alvaro, marziani}@depeca.uah.es

Department of Electronics and Electromechanical Engineering. University of Extremadura. Cáceres. Spain [email protected]

does not always imply the existence of a dangerous object for the railway. The adverse climatology also gives the degradation of the optical channel, so the radiation lack in these circumstances can be confused with the existence of objects. To avoid the generation of alarms in this situation, the detection threshold is adapted to the state of degradation of the channel. Therefore two kind of information are available: the result of the correlation, and a threshold output indicating if the link is active. The emission encoding and the dynamic adaptation of the threshold have been previously described in [3] and [4] respectively. The information about the state of the links (on or off) is combined by means of the Dempster-Shafer evidential theory taking into account the channel degradation and the spatial diversity of the sensorial system structure, so a certainty value can be obtained about the existence objects larger than 50x50x50 cm. Regarding this problem, the following aspects are analyzed: a brief description of the designed structure of the sensorial system; sensors data fusion to obtain the existence certainty of dangerous obstacles; and finally, the most relevant conclusions.

Abstract This work proposes a data fusion technique to validate the detection of objects in railway environments. The system for obstacle detection is based on an infrared emitting barrier and another of receivers placed on opposing side in the railway, establishing different optical links between them. The interruption of one or several links in the sensorial system should produce an alarm. Since the detection is based on the radiation lack in the detectors, channel degradation can produce a similar situation as if there was one obstacle. Since the interruption of the links is the necessary condition, but not sufficient, for the existence of an object, this work proposes the fusion of the sensors information by means of the Dempster-Shafer evidential theory to obtain a certainty value of the existence of objects supposing a risk for the railway.

1. Introduction The great development of the railway as a mean of transport makes necessary more and more reliable the required safety systems. Among these systems, those called detectors of the fall of objects can be remarked, mainly demanded in the high-speed lines, to detect the existence of objects on the track supposing a risk for the railway circulation. The areas that should be scanned are those near tunnels and high steps. The minimum size of the object supposing a risk is 50x50x50 cm [1]. This type of systems is based on different sensorial elements; in [1] it is proposed one based on infrared barriers, where links are established between emitters and receivers. The detection of objects is carried out with the interruption of these links. To be able to discriminate the different emissions in a receiver, it is necessary to encode them. The detection of the emissions is carried out by correlation [2], where a detection threshold is defined to evaluate if the link is active or not. Because the detection of objects is based on the radiation lack at the receivers, this circumstance

1-4244-0681-1/06/$20.00 '2006 IEEE

2. Sensorial System 2.1. Sensorial structure The designed sensorial system is composed of two infrared barriers, one emitting and the other receiving, placed at both sides of the railway, as is shown in Figure 1. The most remarkable sensorial and functional features of the system are detailed in [3]. Summing them up, the minimum dimensions of the object to be detected are 50x50x50 cm, whereas the distance between contiguous transducers is 25 cm. In this way, if an object with minimum dimensions is in the scanned area, at least two links are interrupted. The distance between emitters and receivers is 14 m, given by the width of the railroads, although it could be higher.

1149

Receivers

14 m Rx-7

Rx-6

Rx-5

Rx-4

Rx-3

Rx-2

Rx-1

25 cm Rx

Rx+1 Rx+2

Rx+3

Rx+4

Rx+5

Rx+6

Emitters 25 cm Receivers

14 m

25 cm

Figure 1. Scheme of the design sensory system. Because the infrared emission is not punctual, every emitter excites a group of receivers. This allows multiple links to be established in the sensorial system, apart from those on the axial axis shown in Figure 1. For an infrared emitter with an aperture angle of ±2°, with a distance between emitters and receivers of 14 m, this emitter excites up to 5 receivers, establishing therefore five optical links for every emitter, as depicted in Figure 2.

a

E x-5

±2°

Emitterx

Figure 2. reception.

aperture

 yk(1,1  ( 2 ,1  yk Yk =  yk( 3,1  ( 4 ,1  yk  y ( 5,1  k

the

[

yk( 2 ,x

yk( 3,x

yk( 4 ,x

y

]

e

125 cm

(2)

" yk(1, x " y k(1, X   " y k( 2, x " y k( 2, X  " yk( 3, x " y k( 3, X   " y k( 4, x " y k( 4, X  " yk( 5, x " yk( 5, X  5 x X

(3)

To evaluate if a link is interrupted, every component of the matrix Yk is compared to a detection threshold. Since the meteorology or the solar radiation [5] can affect to the obtained correlation values, this threshold is dynamically adapted to these circumstances by means of the use of the H∞ estimator [6][4], obtaining a vector of five measurements for every receiver corresponding to the state of each link: on or off, as shown in (4) and (5).

2.2. Emission encoding Since a multi-mode operation is carried out in the barrier (simultaneous multiemission and multireception), it is necessary to encode every emission to avoid interferences among the different emissions, and to discriminate them at the receiving block. For that, mutually orthogonal complementary sets of sequences have been used [2][3]. In Figure 3, the different colours of the established links represent different codes in order to avoid interferences between emissions. The detection of the different emissions is carried out by means of a correlation process, where the output from every receiver provides a vector with five measurements, corresponding to the correlation values obtained for each link, as shown in (1).

y (kx = y k(1,x

d

E x+3 E x+4

where G is the process gain, according to the encoding scheme; T is the emission period; θk is the atmospheric attenuation; φn,k is the noise component after the correlation; and j=1, 2, 3, 4 and 5. In (3) the matrix Yk contains the correlation output of all the links in the barrier.

Figure 3 shows the links for a segment with 10 emitters (2.25 m). In this sensorial structure, whenever an object with dimensions larger than the minimum ones appears, at least 10 links are interrupted.

( 5, x T k

c

i= 0

Receiverx+1

in

b

E x+1 E x+2

i= ∞

Receiverx+2

Emission

a

Ex

y k( j , x = G ⋅ θ k ⋅ ∑ δ [k − i ⋅ T ] + φη ,k

25 cm

14 m

e

E x-1

The emitters transmit periodically, being the correlation output obtained for every link as follows:

Receiverx

25 cm

d

E x-2

Figure 3. Links at a segment of 2.25m.

Receiverx-1

25 cm

c

E x-3

Emitters

Receiverx-2 25 cm

b

E x-4

[

z (kx = z k(1, x ( j,x

z k( 2, x

z k( 3, x

z k( 4 , x

]

T

(4)

( j ,x

zk = 1 si yˆ k ≥ Threshold ( j,x ( j ,x zk = 0 si yˆ k < Threshold ( j ,x

z k( 5, x

(5)

Where yˆ k is the estimation of the correlation ( j ,x output yk carried out by the H∞ filter at k instant. Similarly as (3), a Zk matrix is obtained representing the state of all the links. Next section deals how this matrix is processed by Dempster-Shafer’s rule, to obtain the

(1)

where x is the position of the receiver in the barrier; and k is the instant when data are captured.

1150

certainty of the existence of objects larger than 50x50x50 cm.

by an object, it is necessary to take into account the channel degradation. Furthermore, the probability of the object interrupting a link to be in the area A(x) depends on the percentage of the range of the link placed in such area. For the link between emitter x and receiver x the probability is one, but for the rest is 0.5 or 0.25. According to the above considerations, the certainty of interruption of a link between the emitter e and the receiver r by an object is the following:

3. Data fusion to obtain the existence certainty of dangerous obstacles In order to obtain the existence certainty of obstacles larger than 50x50x50 cm (dangerous obstacles for the trains) the detection area has been divided into 25 cmwide influence areas according to the receivers. The influence area of the receiver Rx is represented by A(x), as shown in Figure 4. As was said in previous section, the link colours represent different orthogonal codes in order to discriminate the emissions.

σ e ,r = α e ,r · ρ e ,r

where αe,r is the channel degradation before the interruption of the link; and ρe,r is the probability of the object to be in the area A(x). The value of αe,r is empirically obtained according to Figure 5.

50 cm

A(x-2) Rx-4

Rx-3

A(x-1) A(x)

Rx-2

Rx-1

A(x+1) Rx

Rx+1

Rx+2

Rx+3

(6)

Rx+4

α e, r 1 αmax 14 m

0 TH-min

3·TH-min

( j ,x

yˆ k −1

G

Figure 5. Calculation of αe,r. b

c

c

a

a

b

b

c

In Figure 5 αmax is between 0 and 1 (empirically it has been chosen to be 0.5) ; TH-min is the minimum threshold;

c

E x-4,t1 E x-3,t2 E x-2,t1 E x-1,t2 E x,t1 E x+1,t2 E x+2,t1 E x+3,t2 E x+4,t1 25 cm

( j ,x

G is the process gain -see (2)-; and yˆ k −1

150 cm

is the

estimation of the correlation carried out by the H∞ filter at k-1 instant (before the interruption of the link) [4]. This estimation can be considered as a channel degradation measurement, and it corresponds to the link between emitter e and receiver r. Once σe,r is obtained, the value of the existence certainty of obstacles in the area A(x) is computed taking into account the eleven links that exist in every influence area. If it is assumed that P(li ) = σ e ,r , then cA(x) is obtained

Figure 4. Influence areas. Every influence area is analysed, and a value of the existence certainty of obstacles is obtained. This value is represented by cA(x) ∈ [0,1]. If there exists a minimum dimension obstacle, it occupies two consecutive areas. Then, consecutive influence areas are combined by the Dempster-Shafer’s rule [7] to obtain a value of the existence certainty of obstacles larger than 50 cm in any side.

as the union probability of the independent events:

P(l ) = 0; for i = -5;i ≤ 5;i + + P(li ∪ l ) = P(li ) + P(l ) − P(li ) ·P(l )

3.1. Analysis of the existence certainty of obstacles in A(x) As Figure 4 shows, there exist eleven links at every influence area, established between five emitters and five receivers. These links cross several areas, except from the link between the emitter x and the receiver x, that belongs only to the area A(x). To know if a link is interrupted, it is only necessary to evaluate the state (on, off) of the corresponding element in matrix Zk. Due to the fact that the channel degradation (solar radiation, weather conditions, etc) can generate a lack of signal in the detector, this situation can be mistaken by the existence of an object. For this reason, if at any k instant zk(j,x was null –existence of an obstacle-, but the channel was very degraded at the k-1 instant, it is very unlikely that the lack of signal was produced by an object. Therefore, to obtain the certainty of the link interrupted

P(l ) = P(li ∪ l )

(7)

endfor c A( x) = P(l ) 3.2. Consecutive areas data fusion Once the value of the existence certainties of obstacles is available, they can be combined between consecutive areas, in order to obtain the existence certainty of objects larger than 50 cm on any side. As a result, it is obtained the following vector:

1151

[

CO = co ,1 " co , x " co ,( N Z −1)

]

the channel degradation affects them. Though the theoretical results show the feasibility of the proposed solutions, real tests have to be carried out with an IR barrier that validate the described algorithms. To improve the safety level required in this application, it is necessary to incorporate new sensory systems, as can be cameras or ultrasounds to make up for the IR deficiencies.

(8)

where every component co,x is the result of the fusion between the areas A(x) and A(x+1); and NZ the number of the influence areas. According to the Dempster-Shafer’s theory, if cA(x) is considered as the probability mass of the existence certainty of obstacles in the area A(x), the component co,x can be obtained as follows:

co, x =

c A( x ) ⋅ c A( x+1) 1 − (1 − c A( x ) ) ·c A( x +1) − (1 − c A( x +1) ) ·c A( x )

Acknowledgment (9) The work described in this paper has been possible by funding from the Ministry of Public Works -through the VIATOR project (ref. 70025/T05)- and the Community of Madrid -through the ANESUS project (ref. CAMUAH2005/016)-.

Figure 6 shows an example of the explained algorithm, where there are a small object and a dangerous one. As shown, firstly the existence certainties of obstacles in areas (cA(x)) are computed (first row in the table in Figure 6), and then consecutive areas have been combined using (9). The higher the value of co,x is, the larger the certainty of obstacles larger than 50x50x50 cm is. A(1)

R-1

R0

A(2)

R1

A(3)

R2

A(4)

R3

References [l]

A(5)

R4

R5

R6

R7

[2] Object larger than 50x50x50 cm

[3] Small object

b

c

c

E -1,t1 E 0,t2

E 1,t1

a

E 2,t2

a

b

E 3,t1 E 4,t2

b

E 5,t1

c

E 6,t2

c

E 7,t1

[4] A(1)

A(2)

A(3)

A(4)

A(5)

CA(1) =0.77

CA(2) =0.88

CA(3) =0.77

CA(4) =0,05

CA(5) =0,64

Co,1 =0.96 Co,2 o,1 =0.96 Coo,13 =0.15 Co,4 o,1 =0.08

[5]

Figure 6. Example of sensors data fusion.

4. Conclusions The sensorial system is an IR barrier, required to detect the existence of obstacle on the tracks. Due to the fact that detection is based on the lack of radiation in the receivers, the channel degradation can be mistaken with the existence of obstacles. For this reason, it is necessary validation algorithms. Sensor data fusion based on the evidential theory has been applied in order to obtain an existence certainty of obstacles dangerous for the railway traffic. The proposed fusion algorithm takes into account the spatial diversity of the links that are established in the barrier and how

[6]

[7]

1152

“System for falling obstacle detection on the railway. Technical and functional requirements”. GIF, 2001 (Published in Spanish). Shu-Ming Tseng; Bell, M.R.; “Asynchronous multicarrier DS-CDMA using mutually orthogonal complementary sets of sequences”, IEEE Transactions on Communications, Volume 48, Issue 1, Jan. 2000, pp 53 – 59 J. Jesús García, Jesús Ureña, Álvaro Hernández, Manuel Mazo, J.Carlos García, Fernando Álvarez, J. Antonio Jiménez, Patricio Donato, Carmen Pérez.“IR sensor array configuration and signal processing for detecting obstacles in railways”. Third IEEE Sensor Array and Multichannel Signal Processing Workshop SAM’04. Juan J. García, Cristina Losada, Felipe Espinosa, Jesús Ureña, Álvaro Hernández, Manuel Mazo, Carlos de Marziani, Ana Jiménez, Fernando Álvarez, José A. Jiménez. “Optimal estimation techniques to reduce false alarms in railway obstacle detection”. IEEE International Conference on Industrial Technology. ICIT 2005. pp 459-464 Scott Bloom, AirFiber; Eric Korevaar, MRV Communications; John Schuster, Terabeam; Heinz Willebrand, LightPointe Communications, “Understanding the performance of free-space optics [Invited]” Journal of Optical Networking, June 2003, Vol 2 nº 6, pp 178-200 Simon, D. “To game theory approach to constrained minimax state estimation”. Report of the Department of Electrical Engineering, Cleveland State University. January, 2005. Lawerence A. Klein, Data and sensor fusion: a tool for information assessment and decision making, Spie Press, 2004.