Field of the Invention

[0001] The present invention generally relates to the field of methods and systems for detecting intrusion from outside into a closed entity like the load compartment of a truck or trailer, a container, a building and the like.

Background of the Invention

[0002] Intruder systems are known in the art and offered by several vendors. Various types of intruder systems exist, e.g. based on infrared detection, possibly with photo-electric beams, subsonic or ultrasonic systems, radio systems, etc ...

[0003] Reliable detection is obviously a key feature for such systems. Both false alarms and detection failures should be avoided as much as possible.

[0004] Prior art systems typically apply a low frequency analysis in order to determine an intrusion. One such solution is described in patent application EP1883910, where a system and method for intrusion detection is presented. It tackles the problem of false alarms. It is noted that alarm systems and detectors do not discern intruder activities from other activities, thereby causing frequent false alarms. In other words, the probable cause of the alarm signal is not discerned. As a solution a detection system is proposed comprising a microphone to detect infrasound signals. Infrasound is subaudible sound with frequencies less than 20 Hz. The detected infrasound signals are (assumed to be) caused by the intrusion, e.g. by the movement of a door or a window. Control parameters (threshold levels of probability of occurrence of predefined signal states) of the system are adaptable so that the alarm responses may be maintained to a predetermined temporal rate of false alarms. An element of this detection system employs the determination of the probability that particular information will occur within an ongoing temporal period. Such information correlates to desired detected activities and may include various signal characteristics, the source detecting the infrasound signals, and temporal relationships within and between detected signals.

[0005] WO2008/121041 is also concerned with vehicle intrusion detection. It proposes an improved surveillance system for trucks, trailers and the like. The solution comprises a multi-function day-running side-light unit. A number of such side-light units are mounted on opposite sides of the vehicle. [0006] Solutions based on low frequency analysis however suffer among other things from the important drawback that weather conditions (e.g. wind) may considerably affect the applied method. Consequently, there is a need for an intrusion detection solution wherein these drawbacks are avoided or overcome.

Aims of the invention

[0007] The present invention aims to provide a method and system for intrusion detection wherein the drawbacks of the prior art, specifically the influence of weather conditions, are avoided or overcome. A further aim is to provide a method and system with improved detection capability.

Summary

[0008] In a first aspect the present invention relates to a method for determining intrusion in a closed entity. The method comprises the steps of

- receiving measurements of mechanical vibrations measured inside or outside on the structure of on the closed entity,

- detecting impacts in the received measurements, and

- determining whether there is intrusion or not based on the detected impacts.

With impacts is meant a force or a shock applied on the closed entity over a short time period due to some collision, caused e.g. by an intrusion into the closed entity, the clicking of a lock, the cutting of a cable under tension, the impact of a step on the floor, a knock on the side, ... or by the meteorological conditions acting on the closed entity (mostly indirectly by loose parts moving by the wind, like the end of a cable, the wind playing with a soft sided truck, ...). By measuring the mechanical vibrations and detecting impacts in said measurements the signal analysis can be performed on signals with considerably higher frequency content. In other words, there is no need anymore to measure a signal of, say, less than 20 Hz. It is in this low frequency range that weather conditions like wind have most effect on the measurements. In the proposed method it is irrelevant whether the measurements are performed inside or outside on the closed entity's structure.

[0009] In one embodiment the intrusion is manifested as a single peak, i.e. one (very) large impact. Alternatively, the intrusion can appear as a series of impulses, most typically with smaller amplitude than in the case a single impulse is observed. The actual appearance of the intrusion in the observed signals largely depends on the modus operandi of the intruder.

[0010] In case a train of impulses is detected, the density of peaks in a given time frame is preferably taken into account in the step of determining intrusion. This may provide valuable information in determining whether the observed impulses are caused by an intrusion or not.

[0011] In an advantageous embodiment the method further comprises a step of categorizing the detected impulses as being caused by a known source or as intrusions. This categorizing is preferably performed based on statistical analysis of impulses detected before, i.e. in an earlier time frame. Alternatively, the categorizing is performed based on predetermined knowledge of a physical system producing impulses different from intrusions, a measurement of a parameter of the system and a comparison of that parameter with the single peak or the plurality of peaks.

[0012] In a preferred embodiment the method further comprises a step of filtering the detected impulses in order to reduce or even remove the effect of noise. This can also be achieved by applying a denoising algorithm.

[0013] In another preferred embodiment the step of detecting impacts or impulses comprises performing a transformation on a set of digital samples obtained from the measurements to distinguish between samples indicative of a possible intrusion and other samples. In one embodiment such transformation comprises a non-linear summation of digital samples of said set or a statistical processing of digital samples. Such transformation indeed has as effect on the sample set that differences between sample values are stretched and portions corresponding to intrusions become more easily discernible.

[0014] The transformation may be preceded by a preprocessing step, wherein a frequency analysis or a wavelet analysis is applied to said digital samples. This preprocessing brings the sample set in an adequate form for subsequently applying the transformation. The preprocessing step may yield a quantity derived from said samples, said quantity further being used in the step of performing said transformation. Doing so is advantageous in that the sensitivity is enhanced.

[0015] The method comprises in an embodiment an initial step of measuring the vibrations on the closed entity and digitizing the measurements resulting in the set of digital samples. [0016] In one aspect the invention relates to a program, executable on a programmable device containing instructions, which when executed, perform the method as previously described.

[0017] In another aspect the invention relates to a processing device for use in the method for determining intrusion in a closed entity as described above, said processing device being operable for receiving measurements of mechanical vibrations measured inside or outside said closed entity, for detecting one or more impacts in said received measurements and for determining whether there is intrusion or not based on the detected impacts.

[0018] In a preferred embodiment the processing device is further arranged for outputting an alarm signal dependent on the outcome of the determining step.

[0019] In another embodiment the processing device comprises communication means for transmitting an output alarm signal.

[0020] The invention relates in another aspect to an intrusion detection system comprising a processing device as described and further comprising at least one sensing means for sensing mechanical vibrations.

Brief Description of the Drawings

[0021] Fig. 1 illustrates an embodiment of an intrusion detection system according to the invention.

[0022] Fig. 2 illustrates the application of the invention for a truck.

[0023] Fig. 3 represents a flow chart wherein various possible data analysis algorithms are shown. They can be applied separately or in combination.

[0024] Fig. 4 illustrates a normal distribution where the standard deviation is changed, so the exceptional values become more or less visible.

[0025] Fig. 5 illustrates the calculation of a decision parameter.

[0026] Fig. 6 illustrates a denoising operation.

[0027] Fig. 7 illustrates the relationship between (on the x-axis) the low frequency component defined by the wind on a truck, and (on the y-axis) the levels of the impulses.

[0028] Fig. 8 illustrates an algorithm for a tractor/trailer combination.

[0029] Fig. 9 illustrates an algorithm for analyzing the cause of a measured impulse.

[0030] Fig. 10 illustrates a full algorithm. [0031] Fig. 1 1 illustrates the detection of many relatively small impulses.

Detailed Description of the Invention

[0032] The present invention presents a method for detecting intrusion in a vehicle cabin, container, building, ... as well as a processing device for use in said method and an intrusion detection system comprising such a processing device.

[0033] The invention exploits the observation that intrusion detection can be performed based on measuring mechanical vibrations on the structure of the closed entity and detection of impacts or impulses in the measured signals. Such impulses are forces acting on the closed entity producing a finite change of momentum. They are characterised in the time domain as short discontinuities in the signal and in the frequency domain as broadband signals whereby they differentiate from other signals like resonances (that are only visible around one or a few frequencies, normally in the low frequency range) in the higher frequencies, e.g. above 100 Hz. This is opposed to the prior art solutions wherein one relies on low frequency measurements as explained in the background section. The invention further distinguishes from the prior art in that an advantageous data analysis approach is adopted that considerably improves the visibility of signal portions corresponding to an intrusion as compared to other portions of the measured signal, for example by applying a non-linear summation.

[0034] Fig.1 illustrates schematically an embodiment of an intrusion detection system comprising a processor device according to the invention. The detection system contains at least one sensing means (sensor) adapted to detect mechanical vibrations. This can be a vibration detector, an accelerometer, ... The sensing means can be axis sensitive, i.e. arranged for sensing in one or more specific directions according to its position. Each sensor defines a measurement channel. For every measurement channel digitizing means are provided, e.g. a conventional A/D converter to convert the measured analogue signal into a corresponding digital representation. This digitisation can be performed either inside the processing unit, in a separate A/D converter or in A/D conversion means integrated in the sensor. A separate A/D converter may be in connection with the processor either via a wireless or wired network link. The use of communication based on a protocol of the IEEE 802 standards family (e.g. ZigBee, Wi-Fi, etc) can be envisaged. The processor according to this invention is advantageously also provided with memory for storing data related to the measurement or the processing thereof.

[0035] In case a system is used with more than one sensor or when sensor signals have to be compared to avoid a false alarm because the mechanical vibration was caused on another unit (for instance the driver in the cabin), the intrusion detection system advantageously comprises a communication device, preferably arranged for short distance communication. The communication can be performed according to an IEEE 802 standard. For a system with more than one sensor, a module comprising a detector, an A/D convertor and a radio device is provided, completed with a small processor to be able to organize the communication. The analysis is carried out in another module provided with a more performing processing means. For a system where different units are to be protected (like a stack of containers), every module also contains a radio unit (e.g. ZigBee or another communication standard of the IEEE 802 family) to be able to communicate with the neighbouring units to analyze if a measured intrusion is direct or indirect from these neighbouring units.

[0036] Fig.2 shows an application for a truck. For trucks, the detection system is meant to protect the load, when the truck is parked, with or without somebody (driver, ...) in the cabin. One or more modules, optionally with simplified processor means, are placed on strategic places on the frame of the trailer/load compartment. These units have one, two or three-dimensional accelerometers as sensors. They communicate with the main module, i.e. the module with the processor that performs the bulk of the data analysis, which is placed in the driver's compartment or cabin.

[0037] Another application of the invention is in buildings, for instance integrated in walls or floors. Only modules with reduced processing power are preferably used throughout the building. They communicate wirelessly or through "IP over 220V" with a main module comprising processor means suitable for carrying out the data analysis. This module contains a more performing processor or may even be a multiprocessing system.

[0038] The detection can further be applied for protecting containers. Each container is provided with a sensing module and the modules are provided with communication means for communication with the other modules. At least one module is further arranged for communication with the outer world, e.g. for transmitting a silent alarm. [0039] Now the signal processing is discussed more in detail. The signal analysis can be performed in a statistical way, whereby an intrusion is defined as a non-expected signal. Alternatively, an analytical approach can be followed.

[0040] Whereas in the prior art solutions the classical RMS value of the samples is used in the analysis, in the present invention a different solution is presented. In the approach according to the invention the set of samples representing the measured mechanical vibration signal is so treated that the resulting transformed set of samples allows distinguishing much easier signal portions corresponding to an intrusion from other parts of the signal. Short signals due to an intrusion are thus made more visible in a noisy environment.

[0041] The transformation performed on the sample set can take various forms that each achieve the same goal. One advantageous solution is based on a nonlinear summation (e.g. based on the power of 10) of samples of the sample set :

A = Iog ₁₀ (∑l0 ^ai/a0 )

i

whereby a, is a digital representation of a measured amplitude and a ₀ a reference value that defines the amplification. In this way, the pulses within the total sum are amplified and make the signals due to an intrusion, however short, more visible.

[0042] Alternatively, the transformation of the data sample set is performed with a continuous wavelet transformation. Further, an approach can be adopted wherein a transformation is performed on the probability of having a measurement falling within a predetermined range, as explained below in detail.

[0043] Fig.3 represents a flow chart illustrating in each branch a different possible way of statistically processing the data samples representative of the measured mechanical vibration. In more specific embodiments of the invention several of the proposed options can be combined.

[0044] In a first branch of the flow chart of Fig.3, a preprocessing frequency analysis is carried out. An intrusion is mostly characterized as an impulse or impact and is in general independent of the characteristics of the truck, trailer, .... This means that in an FFT analysis an intrusion is detectable in the higher frequencies. Therefore a high pass filter is used to detect the impacts. Before this FFT analysis, if the sensors create a large background noise, it can be useful to perform a noise cancelation, for instance based on a wavelet denoising algorithm. For the high frequency band, for every sample, instead of the classical RMS value, a non-linear summation is calculated. In this way, one can amplify the pulses within the total sum and make the signals due to an intrusion, however short, more visible :

A = log ₁₀ (∑lO ^fli/fl0 )

i

whereby a, is a measured amplitude of the sample and a ₀ a reference value that defines the amplification.

The further analysis can be done either directly on the measured signal in digital form or on a quantity derived from that signal, e.g. the proportion of the measured signal on the trailer divided by the signal on the cabin :

truck

^RMS

~ _y cabin

^RMS

This improves the sensitivity on the trailer. When there is a normal external impact from the wind or from other external sources as the vibrations due to the traffic, this has an impact on the trailer and an impact on the cabin. The ratio of both is relatively continuous in a certain situation. When there is an intrusion in the load compartment or trailer, this ratio works as a sort of amplification factor and makes the intrusion more visible.

[0045] A second branch in Fig.3 illustrates a signal processing based on wavelet analysis rather than frequency analysis. A continuous wavelet transform is applied. Different wavelet types are used. The Daubechies D2 to D6 wavelet transforms give results with a good discrimination possibility. Daubechie wavelets are well known in the art, for example from the handbook "Ten Lectures on Wavelets" (I. Daubechie, CBMS-NSF Regional Conference Series in Applied Mathematics, no.61 , 1 992). As the invention concerns a security system, only relative short signals (e.g. 0.5, 1 or 2 seconds) can be used in order to be able to detect an intrusion in a time period as short as possible. Therefore relatively large time steps can be employed to speed up the calculations. The intrusions are easily recognized in the higher wavelet scales. Therefore, as basic criterion for the detection the sum of the coefficients of higher scale or the RMS value of these coefficients is used. An alternative is to work with a denoising algorithm based on wavelets.

[0046] The third branch of the flow chart shows a "more channel" frequency analysis, where a correlation between more signals is performed. [0047] The basic problem with the detection is that the external disturbances in a more disturbed environment can be higher than the real intrusions in a less disturbed environment. This implies that some normalization needs to be performed. For this a chi-squared cumulative distribution function can be used to calculate the probability that a certain measurement is equal to a predetermined average. The smaller the probability, the higher the chance that this measurement is abnormal and is an intrusion. To amplify this effect and to get a workable parameter, the probability p of a measurement is transformed to a useable detection parameter D.

This chi-squared cumulative distribution analysis is based on a First In - First Out (FIFO) database (i.e. a database wherein stored measurement data are taken in the order they have entered the database) of the measurements of the last t minutes. This analysis offers two problems when all measurements are used :

- The sensitivity of the system, especially in more disturbed situations, reduces dramatically as the influence of an intrusion is not too exceptional and therefore detection parameter D is not very large.

- The measurements of a real intrusion are in the database and become part of "the normal situation". This means that after a while the intrusions are not detected again.

Therefore a pre-analysis is performed on every measurement to determine if it can be stored in the database or not. Two possible criteria that can be used, are :

If the average value μ increases too much due to a single measurement, this measurement is not allowed.

If the standard deviation σ increases too much due to a single measurement, this measurement is not allowed either.

The result of this is shown in Fig.4. Obviously, the sensitivity for abnormal situations increases by reducing the database size by not allowing outlying measurements.

[0048] Note that other algorithms can be applied as well. For example, the transformation can in a possible embodiment be a weighing of the digital samples of the measured mechanical vibrations.

[0049] The basic assumption for the detection of an intrusion is that this intrusion comprises one (very) large impact or a lot of small impacts. Combinations of these characteristics are possible too. Which characteristic actually occurs basically depends on the modus operandi of the intruder.

[0050] To take this into account, the detection measurements are added together to a final decision parameter FD. When the detection is large enough, it is added to the value of FD. When there is no detection, or when the detected signal is too small, FD's value is reduced with a predetermined amount. This is illustrated in the flow chart of Fig.5. In this way one gets a good weighing of both possible types of intrusion.

[0051] For every characteristic criterion a decision parameter FD is calculated. This FD has to be compared with a reference. Due to the manipulation of the distribution functions as illustrated in Fig.4, the detection parameter D - and thus the final decision parameter FD - are not fully independent of the external disturbing situation. Analyzed in another way, the method of calculating this decision parameter is not normalizing the measurements for 100%. To take this into account the criterion to compare FD is adapted by using a regression based on a non-filtered, non- manipulated database of the measurement levels. This gives a regression coefficient of more than 90%, so it is a good indication for the level of disturbances.

[0052] Practical measurements show that no matter which basic criterion that is used and analyzed, intrusions may occur which give rise to signal levels considerable higher than any value measured when there is no intrusion. Hence, in such situation there is no need for a statistical analysis and the intrusion is detectable from the measurement itself. Therefore, next to the statistical analysis and decision making, one can sometimes employ a direct decision approach based on a simple comparison of measured parameters and a safe fixed threshold.

[0053] Instead of a statistical analysis, also an analytical approach can be followed. Based on the dynamic characteristics of a truck/trailer and based on the modus operandi of possible intrusions, there are two relevant signal types to detect

- High impacts due to fast and brutal intrusions

- Many (very) small impacts during a longer time period due to very careful intrusions

[0054] On a trailer there are two relevant axes to measure. On the vertical axis impacts and low frequency response are to be determined, whereas on a horizontal axis perpendicular to the length of the trailer impacts can be measured. On the cabin only the vertical axis and the impacts are relevant. [0055] In order to recognize the impacts a sampling rate of at least 500 samples/s is necessary. When measured with an accelerometer, a wavelet denoising - for instance Daubechie wavelet transform D08 with a minimum level 5 - gives the relevant impacts as can be seen on Fig.6. The low frequency response is found by filtering the signal between 1 to 6 Hz and 8 to 30 Hz. This can be measured with an accelerometer or a geophone.

[0056] The measurement samples each have a duration of at least 100 msec. Again, as in the above described statistical approach, a transformation is performed on the sample set so that portions of the processed signal corresponding to an intrusion become more visible. Instead of calculating and directly applying a RMS value for the sample, a non-linear summation is used. Various possibilities exist. Two options are explained more in detail below.

[0057] A sample being processed is denoted s,. A constant s _t represents a predefined constant. If s, < s _t then s, is set to zero, otherwise it keeps its value. Next a transformation is performed on the non-zeroed s,.

S = 0 ^{si , sre} f whereby s _re f represents a predefined reference. Then :

1000

for each S,- ; if max(Si) = 0 then S = 0 else S = log(∑ S _t )

i=\

[0058] In a second option the same first step is performed as above, i.e. if s, < s _t then s, is set to zero, otherwise it keeps its value. Then, the following product is calculated :

Note that the product works much faster on most embedded processors than a third power. The third step then is

f looo λ ³

for each S,- : if max(Si) = 0 then S = 0 else S = log(∑ S _t )

V *=i J

The impact signals on the trailer are added together. The low frequency signals are not combined. They are only used as an indication of the wind speed, as they are defined by the whole body movement of the trailer on its suspension system. [0059] The intrusion detection is preferably determined by the parallel operation of different algorithms, based on the intrusion modus operandi. As the skilled person readily understands, only one algorithm can be applied as well.

[0060] The intrusion detection is done using the impacts and is based on two principles. Firstly, some of the signals on the trailer are not relevant, because they are clearly caused by other causes. They are removed from the signal. Two examples are given that demonstrate this first principle :

- When the impact signal in the cabin is clearly higher than the impact signal in trailer, the signal in the trailer is caused by a movement in the cabin, for instance due to a movement of the driver.

- When the low frequency movement of two or more sensors on extreme positions on the trailer are in the same order, the movement of the trailer is caused by a global external source (like wind).

Secondly, the residual signal level increases when the weather conditions are worse, i.e. when there is more wind. The impacts are not directly caused by the wind, but indirectly for instance by some loose binding cable that is blown up by the wind.

There is a direct relation between the level of the wind and the low frequency movement of the truck. This low frequency level is used as a prediction for the maximum level due to impacts. An example is given on Fig.7. When the measured impact level is higher than the predicted level, an intrusion alarm is given.

[0061] Some flow charts are added that describe a program specific for a tractor/trailer combination.

In Fig 8 an algorithm is illustrated wherein based on the movement of the truck, measured through a low frequency component (G1 and G2 are the measurements of two sensors in this case), the criterion for a certain algorithm is defined by a regression based on the graph of Fig. 7, for every individual algorithm. The values A, B and C are defined on the relevant graph of the type of Fig 7.

Fig 9: illustrates an algorithm to analyze if a measured impulse is caused by something in the cabin or not. If the impulse is relevant, it will be used to analyse if it can be caused by an intrusion or part of an intrusion (i.e., a single pulse in a train of impulses in a certain time), if it is caused by something in the cabin, the cause is known and the impulse on the trailer is not used for further analysis. The algorithm gives as result the relevant impulses. As explained before, there are two possible ways for an intrusion: either one important impulse or a train of consecutive impulses in a certain time period. Fig. 10 describes the analysis for the first type and Fig. 1 1 for the second type.

For the analysis of a maximum, two criteria are used to see if the impulse can be caused by an impact on the trailer due to an intrusion. In the first place one analyses if it can be caused by a source in the cabin, in the second place, the geophone is giving a low frequency image of the truck movements. With two geophones, one can know if the movement of the truck on its natural frequency is caused by a global impact (like the wind) or a local impact like an intrusion or the tapping of a loose cable. When the impact measurement is relevant, it needs to be compared with the criterion, which is changing following the algorithm of Fig.8 in function of the disturbance level of the environment.

Fig 1 1 gives one detection method of an intrusion, i.e. the detection of many relatively small impulses in a certain time period through a sliding sum. In the first place every individual impulse is analyzed to check if it is caused by the cabin or not. If not, it is added to a "floating sum", a sum of the impulse levels over a predefined time. This value is than compared with a criterion based on the algorithm of Fig. 8.

[0062] Some applications of the proposed invention are the following. A first application concerns intrusion from outside in the load compartment of a truck or trailer for protection against theft of load, protection against the entrance of stowaways, protection against people putting smuggling material in the load. The way of entering the trailer is not relevant: through the door, the roof, the side walls if soft sided, ... The system works independently of the type of truck, the load, the weather conditions, the traffic conditions around the truck, ... The system only reacts to intrusions in the trailer and not in the cabin, so the driver can be in the cabin, enter or leave, sleep in it, ... without activating the alarm. A second application may be intrusion from outside in containers. Again this is a solution to track the entrance of thiefs, stowaways, smuggling material, ... The way the container is entered is irrelevant: through the door or by drilling a hole in the side, the bottom or the top, ... The system works independently from the type of container, the load, the traffic conditions around the container, the weather, ... The containers can be piled up, as is normally done. Through communication between the systems on different containers, it is detected where the intrusion occurs and only for that container the alarm isl set off, if relevant. A third application is in buildings where the system can be fitted into the wall or into a floor. The entrance of an intruder, thief or other, is detected. The system adapts itself to the environmental conditions that can vary for instance due to a different traffic situation outside the building. [0063] Although the present invention has been illustrated by reference to specific embodiments, it will be apparent to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied with various changes and modifications without departing from the scope thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. In other words, it is contemplated to cover any and all modifications, variations or equivalents that fall within the scope of the basic underlying principles and whose essential attributes are claimed in this patent application. It will furthermore be understood by the reader of this patent application that the words "comprising" or "comprise" do not exclude other elements or steps, that the words "a" or "an" do not exclude a plurality, and that a single element, such as a computer system, a processor, or another integrated unit may fulfil the functions of several means recited in the claims. Any reference signs in the claims shall not be construed as limiting the respective claims concerned. The terms "first", "second", third", "a", "b", "c", and the like, when used in the description or in the claims are introduced to distinguish between similar elements or steps and are not necessarily describing a sequential or chronological order. Similarly, the terms "top", "bottom", "over", "under", and the like are introduced for descriptive purposes and not necessarily to denote relative positions. It is to be understood that the terms so used are interchangeable under appropriate circumstances and embodiments of the invention are capable of operating according to the present invention in other sequences, or in orientations different from the one(s) described or illustrated above.