Method of objects detection employing passive IR detectors for security systems
Tomasz Sosnowski*a, Henryk Maduraa, Mariusz Kasteka, Tadeusz Piątkowskia, Edward Powiadaa a Institute of Optoelectronics, Military University of Technology, 2 Kaliskiego Str., 00-908 Warsaw, Poland ABSTRACT PIR detectors used in security systems for people detection operate in far IR range (8÷14) µm. These detectors most frequently employ pyroelectric sensors. Application of a single pyroelectric sensor does not ensure distinguishing the phenomena of alarm character from, so-called, false alarms caused by, e.g., air turbulences or changes in a background temperature resulting from sun radiation. Thus, in PIR detectors, the sensors with two active elements are used (two sensors) and an alarm signal is determined on the basis of analysis of a difference (or a sum) and their output signals. Essential drawback of currently available PIR detectors is low efficiency of detection of slowly moving or crawling people. Efficiency of detection of slowly moving objects is low because radiation from such objects is close to background thermal noises. The presented signal analysis is based on determination of average moving value in three “time windows” of a defined wavelength. Moreover, a principle of “time windows” creation is given and an algorithm for determination of detection thresholds is described. In PIR detector, an adaptation detection threshold was taken following thermal changes of a background. Influence of sun radiation is taken into account in the algorithm of determination of adaptation detection threshold. Keywords: passive infrared detector, signal processing, probability of detection.
1. INTRODUCTION Main elements of security systems are PIR detectors. In general, detectors operating inside buildings have small detection range, small ranges of working temperature, and relatively simple algorithms of intruders detection. The detectors used for protection of objects or large areas (buildings, airports) have larger detection ranges and complex algorithms of signal processing on which significantly depend efficiency of their operation. Essential drawback of currently available PIR detectors is low efficiency of detection of slowly moving or crawling people. It is because radiation from such objects is similar to background thermal noises. Moreover, to detect slowly moving or crawling people, the lower limit frequency of a transfer band of PIR detector should be near zero. By fulfilling this condition, increase in low-frequency noises occurs causing next detector’s sensitivity decrease. Algorithms of intruder detection have to be different than these for typical PIR detectors. To detect crawling people, larger number of sensors should be used (more detection zones) what will cause increase in signal-to-noise ratio because each of sensors will “see” the smaller field of view. PIR detectors used in security systems for people detection operate in far infrared spectral range (8÷14 µm). In these detectors, the most frequently used are pyroelectric sensors allowing for detection of temperature changes as small as 10-6 K. Application of a single pyroelectric sensor does not ensure distinguishing the events of alarm nature from, socalled, false alarms caused by, e.g., air turbulences or background temperature changes resulting from sun radiation. Therefore in PIR detectors, the most frequently used are pyroelectric sensors with two active elements (two sensors) and an alarm signal is determined on the basis of analysis of a difference (or a sum) of their output signals1, 2. Usually, pyroelectric sensors are mounted, together with a transistor and a resistor polarizing its gate, in standard hermetic housings. A value of this resistor can be even up to 1011 Ω, in dependence on preamplifier configuration. The most frequently, as the amplifying elements, JFET or MOSFET transistors are used that are mounted near a detector. *
[email protected]; phone +4822 6839383; fax +4822 6668950; www.wat.edu.pl
2. CONSTRUCTION OF PIR DETECTOR Main elements of PIR detector are: objective (mirror or refraction one), set of pyroelectric sensors, and electronic systems (Fig. 1). The sensors convert an optical signal emitted from the “being observed” surface into an electrical signal. This signal is processed in the electronic systems (it is amplified, filtered, sampled) and next it is analyzed in a microprocessor system. Electronic Units
Unit of sensors with optical collectors Lens (germanium, gasir)
Optical Diaphragms
Fig. 1 Simplified diagram of PIR detector.
The presented PIR detector detects the crawling people at the distance of 140 m. High signal-to-noise ratio was obtained due to application of the larger number of pyroelectric sensors, i.e., larger number of detection zones (channels). Application of larger number of sensors forces the necessity to develop a complex optical system (Fig. 2). The optical system of PIR detector has to ensure such a position of detection zones to avoid presence of the areas which are “not seen” by a detector3-5. Channel 6
Channel 2
Channel 4
Channel 7
Channel 5 Channel 3
4,5m
Channel 1
1,3m 17m 25m
43m
94m
100m
140m
Dual pyroelectric sensor
Fig. 2 Detection zones of PIR detector in horizontal and vertical planes (top) and pyroelectric sensors with optical concentrators and general view of a detection set with seven two-segment sensors (bottom).
Radiation signals caused by slowly moving, especially crawling, people are characterized by similar luminance amplitudes and velocities of its change in time such as fluctuations of background radiation. The signal amplitude from pyroelectric sensors is directly proportional to the velocity of a change of radiation signal in time (i.e., to the velocity of a moving object). Disadvantageous property of pyroelectric sensors is a voltage drift (pyroelectric detector is equipped with a field transistor operating as a voltage follower) which at low velocities of intruder movement can have temporal characteristics of signals, comparable with the characteristics originating from a moving person. Thus, in order that the sensor could efficiently detect slowly moving person, it is necessary to develop an algorithm distinguishing both
characteristic features of signal changes from a sensor caused by a photon noise and signal changes caused by a temperature drift of a pyroelectric sensor.
3. METHOD OF SIGNAL ANALYSIS An object in the sensor’s „observation zone” (inspection zone) is detected when a conventional detection threshold is exceeded by a signal level at the detection system output, caused by IR emitted from an object. In order to minimize the probability of false alarms, an adaptation detection threshold should be determined “following up” the changing atmospheric conditions causing changes of “thermal scene” parameters3, 6. N
a
1 M >> K is assumed.
Signals processing is carried out in the following way. The signal from a detection set is sampled and the voltages of successive signal samples of instantaneous values ai are added. The time window is formed, containing N samples, which in each consecutive cycle “shifts” by one sample. On the basis of an arithmetic average of voltage values of the samples in the window (Fig. 4), a value of the reference voltage level ϕi is determined
ϕi =
1 (ai− N +1 + ai− N + ... + ai ) , for i>N N
(1)
where i is the sample number which takes the values i = 1, 2, 3, ... N N
In following cycles of analysis N window is moving for one sample
N Signal
ϕi - reference level
5
15
20 30 Number of sample
35
45
i
Fig. 4 Sampled signal from a detection channel and the reference level ϕi calculated according to Eq. (1).
In the next window M, that is in the previously formed window N and containing M samples (for i≥N+M), the instantaneous values of the signal βi (Fig. 5) are determined as a difference of instantaneous values of voltages of consecutive samples of signal and the reference level voltage ϕi
β i = a i − ϕ i , for i≥N+M.
(2)
For further calculations, the absolute value βi is taken as
βi = ai − ϕi , for i≥N+M.
(3)
Deviation of signal value from reference level
βi
5
15
20 30 Number of sample
35
45
i
Fig. 5 Instantaneous deviation of the signal βi from reference level calculated according to Eq. (2).
On the basis of such determined data, the instantaneous value of the detection threshold voltage Di which for i≥N+M is calculated as:
Di = A
1 ( βi−M +1 + βi−M + ... + βi ) , M
(4)
where A is the coefficient considering detector design parameters. For typical solutions, the A coefficient takes the values 1÷5. For PIR detectors of several detection zones and extended operation range, the A coefficient of the higher values for the zone being near detector is taken and A of the lower values is assumed for distant zones. In order to diminish the influence of sun radiation, precipitation, and ambient temperature, the instantaneous corrected detection threshold Dki is determined:
Dki = Di ⋅ k s ⋅ k w ⋅ kt
i≥N+M
(5)
Signal
where ks is the correction coefficient considering sun radiation, kw is the correction coefficient of precipitation influence, and kt is the correction coefficient of ambient temperature. The correction coefficient ks increases the level of a detection threshold in case of sun illumination8-10. The values of changes in a detection threshold were determined experimentally from investigations on influence of sun illumination level on the increase in both object temperature and background thermal noises. Because the object is also illuminated with sun radiation, which increases its equivalent temperature, increase in a detection threshold value does not cause decrease in a sensor range and undesirable disturbances are eliminated. The value of ks coefficient was taken as 1÷1.5.
5
15
20 30 Number of sample
35
45
i
Fig. 6 Instantaneous value of the detection threshold voltage Di and the absolute values of βi signal.
The kw coefficient causes decrease in a detection threshold in case of precipitation in the detection zone of PIR detector. When intensity of precipitation (rain, snow) is higher, IR is much more attenuated what results in lower radiation reaching the detector, thus the signal at its output is weaker. When keeping the same value of a detection threshold, the sensor range is smaller. The kw coefficient is just to compensate the influence of precipitation in such a way to keep sensor range unchanged. The value of kw coefficient was taken in the limits 0.7÷1. In the next step, the samples in the time window K are analysed in which the parameter γi is determined as an arithmetic average of the values of the signal samples
γi =
βi :
1 (βi − K +1 + βi − K + ... + βi ) , for i≥N+M+K. K
(6)
An object is detected (Fig. 7) in the analysed signal when the parameter γi is higher than the instantaneous value of the corrected detection threshold Dki, i.e., when
γ i − Dki > 0 .
(7)
In Fig. 7, also short lasting disturbances of high amplitudes are illustrated (samples 18 and 45) which are not interpreted as object detection.
Value of parameters γ i , D ki
8 6
Detection level Dki
Detection of object γi - average value of |βi | in window K
4 2
5
15
20 30 Number of sample
35
45
i
Fig. 7 Object is detected when a value of γi parameter exceeds a value of corrected detection threshold (in figure for samples No 26÷No 31).
4. RESULTS OF INVESTIGATIONS OF METHOD USED IN PIR DETECTOR During the measurements, carried out under various metrological conditions, the data were registered allowing for checking the correctness of algorithm operation, i.e., of a detection threshold determination made in a microprocessor system. The results of signals registration and the calculated values of detection thresholds were registered using the software developed for communication between PIR detector and computer. This software provides registration of signals from particular detection zones (channels) and registration of signals from particular stages of signal analysis. Figures 8 and 9 present selected records of signals from a PIR detector which were produced due to a crawling person in an inspection zone8, 10. Fig. 8 shows the test field and PIR detector during the tests. The presented results illustrate a change of a signal in the detector channels caused by a crawling person and adaptive change of a detection threshold resulting from the change of the analysed signal.
Fig. 8. The test field and PIR detector during the tests.
400 Channel 1 350 300
Signal
250 200 150 100 50 0 100
150
200
250
300 t [s]
350
400
450
500
Fig. 9. Signal in detection channel 1 of a PIR detector (ambient temperature 31ºC, velocity of a crawling person 0.06 m/s, distance 136 m). 400 Channel 2 350 Signal 300
Signal
250 200 Treshold detection
150 100 50 0 100
150
200
250
300 t [s]
350
400
450
500
Fig. 10. Signal in detection channel 2 of PIR detector (ambient temperature 26ºC, velocity of a crawling person 0.1 m/s, distance 140 m). 500 450
Signal
Channel 1
400 350 Signal
300 250 200 Treshold detection
150 100 50 0 50
60
70
80
90 t [s]
100
110
120
130
Fig. 11. Signal in detection channel 2 of PIR detector registered at night (ambient temperature 25ºC, velocity of a moving person 1 m/s, distance 140 m).
The investigation results presented in the above figures confirmed proper operation of a detector, especially correctness and high efficiency of signal analysis method. Figure 10 shows a detection threshold following thermal changes of a background which, for this measuring case, were caused by sun radiation. Figure 11 presents the recorded changes of a signal in detection channel 1 of PIR detector caused by a person walking with a velocity of 1 m/s, what proves universality of the used algorithm of a signal analysis allowing for detection of very slowly or very quickly moving people. Efficiency of a signal analysis method, the aim of which is to determine a threshold value of a signal allowing for object detection, can be confirmed by adequately high probability of object detection and low probability of false alarm.
5. DETERMINATION OF A PERSON DETECTION PROBABILITY A signal coming from an object is disturbed with the noise generated by a background and electronic systems of PIR detector. It can be assumed that the noise at the detector output is a white noise, while instantaneous values of voltages can be described with a normal distribution with the expected value xN equal to zero and the standard deviation σ N . For a signal with Gaussian distribution of probability density of instantaneous values of voltages, the detection probability is described from a relation
PD = 1 −
1 σN
⎛−x 2 ⎞ ⎟ dx , exp ⎜ ⎜ 2σ 2N ⎟ 2π xD∫− xS ⎝ ⎠ ∞
where xD is the value of a detection threshold and x
S
(8)
is the average value of the signal.
Under the same conditions, the probability of a false alarm is expressed as
PFA =
1 σN
∞
⎛ − x2 ⎜⎜ exp 2 2π x∫D ⎝ 2σ N
⎞ ⎟⎟ dx . ⎠
(9)
For low probability of false alarms and high probability of object detection when it is fulfilled, Eqs. (8) and (9) can be transformed into the following form
⎛ − ( xS − xD )2 ⎞ ⎟, PD = 1 − exp⎜⎜ 2 ⎟ σ 2 2 π ( xS − x D ) N ⎝ ⎠ σN
PFA =
(10)
σN
⎛−x 2 ⎞ exp ⎜⎜ D2 ⎟⎟ . 2 π xD ⎝ 2σ N ⎠
(11)
These probabilities depend on two parameters, the ratio of a detection threshold value to the standard noise deviation xD/σN as well as the ratio of an average value of a signal to the standard noise deviation xS / σ N . The detection probability can be presented as a function of signal-to-noise ratio by introducing the notation SNR =
xS σN
and transforming Eq. (11) we have 2
⎛ x ⎞ − ⎜⎜ SNR − D ⎟⎟ σN⎠ 1 PD = 1 − exp ⎝ . 2 ⎛ xD ⎞ ⎟⎟ 2 π ⎜⎜ SNR − σ N ⎠ ⎝
(12)
For calculations of detection and false alarm probability, a standard noise deviation should be determined basing on the changes of the Sk(m) parameter, and a threshold value should be substituted instead of the xD value.
Probability of detection PD [%]
100 90 89
PFA = 0.2
80
PFA = 0.1
70 60
PFA = 0.01
50 39
40 PFA = 0.05
30 20 10 0 1
3
2
5
4
SNR
6
SNR = 4.95
SNR = 2.83
Signal
3.00 2.75 2.50 2.25 9.8 10.0 10.2 10.4 10.6 10.8 11.0 11.2 3 Number of sample 10
Fig. 12. Probability of object detection for a primary signal (at the amplifier output) and the signal processed with the proposed analysis method.
Figure 12 presents the diagrams of detection probability as a function of the signal-to-noise ratio for a few assumed probabilities of a false alarm. According to Eq. (12), the probability of false alarm depends on the xD/σN ratio. In a method of signal analysis used for PIR detector, one can take the higher value of xD (higher SNR). After filtration, also the lower value of σN is obtained what ensures lower values of false alarm probability. This probability has been determined for a real signal and the noise shown in Fig. 13. For a signal at the amplifier output (Fig. 3), a signal-to-noise ratio is significantly lower than for a course obtained after the signal processing by analysis algorithm and therefore the detection probability (for the same probability of a false alarm) should be higher11, 12. 0
10 PFA
-1
10
-2
10
-3
10
-4
10
-5
10
-6
10
-7
10
-8
10
Fig. 13. False alarm probability.
1
2
3
4
5
6
7 XD /σN
6. CONCLUSIONS To recapitulate, it should be stated that the developed PIR detector detecting slowly moving and crawling people has the parameters allowing for its application in security systems. Satisfactory results of laboratory and field investigations of the developed and performed model of a PIR detector have confirmed that the developed detector detects the objects moving with the velocities from 5 cm/s to 5 m/s. A detection range of crawling people (for the performed model) is of the order of 140–150 m. In a method of signal analysis used for PIR detector, gives lower values of false alarm probability. The new method of signal processing gives a higher the detection probability than PIR with normal signal processing.
REFERENCES [1] [2] [3] [4] [5]
[6] [7] [8]
[9] [10] [11] [12]
ELTEC Instruments Inc., Catalogue of firm products, P.O. Box 9610, Central Business Park, Daytona Beach, Florida 32120-9610, USA, 1996. Madura H., Sikorski Z., Polakowski H., Kastek M., „Long-range passive IR sensor”, Quantitative Infrared Thermography 5 QIRT’2000, Reims, France, pp. 92-96, (2000). Kastek M., „Method of objects detection in IR system of elongated detection zone”, Doctors Thesis, Warsaw, The Military University of Technology Library, 2002 (in Polish). Madura H., Sikorski Z., Polakowski H., Kastek M., „Automated stand for measurement of parameters of long-range passive IR sensor”, Quantitative Infrared Thermography 5 QIRT’2000, Reims France, pp. 118-121, 2000. Kastek M., Madura H., Morawski M., Piatkowski T., Powiada E., Polakowski H., „Test bed for measurement of angular parameters of passive infrared sensors” Infrared Physics & Technology, Volume 49, No 3, pp. 198-201, 2007. Leach G., „A single performance measure for perimeter intruder detection systems”, European Conference on Security and Detection, Conference Publication No 437, pp. 114-119. 28-30.04. 1997. Madura H., Kastek M., Powiada E., „Method of objects detection by means of IR sensors”, Polish Patent Office, Patent Application P-360064, Warsaw 2003 (in Polish). Polakowski H., Piątkowski T., Madura H., Kastek M., Stojak Z., Powiada E., Chmielewski K., Kulewski R., „Method of disturbances correction in IR receivers”, Polish Patent Office, Patent Application P-352889, Warsaw 2002, (in Polish). Madura H., Kołodziejczyk M., „Influence of sun radiation on results of non-contact temperature measurements in far infrared range”, Opto-Electronics Rewiev, No 13 (3), pp. 253-257, 2005. Madura H., Piątkowski T., Powiada E., „Multispectral precise pyrometer for measurement of seawater surface temperature”, Infrared Physics & Technology Vol. 46, pp. 69-73, 2004. Madura H., „Method of signal processing in passive infrared detectors for security systems”, Computational Methods and Experimental Measurements, WITPress, Southampton, Boston, UK, pp.757-768, 2007. Kastek M., Sosnowski T, Piątkowski T., „Passive infrared detector used for detection of very slowly moving or crawling people” Optoelectronics Review Vol. 16 (3), pp. 77-84, 2008.
