Detection method for detecting alarm object in luggage

Abstract

The invention relates to a detection method for detecting an alarm object (AO) in a luggage item (B), comprising the following steps:-performing a transmission analysis (TA) on the luggage item (B), generating a transmission result (TR),-segmenting the transmission result (TR) in order to identify the object (O), generating a simulated detector signal (SDS) for at least one identified object (O), the alarm material (AM) is assigned by means of a diffraction simulation (DS) specific to the diffraction analysis (DA),-the diffraction analysis (DA) of the same luggage item (B) is carried out to generate a measurement detector signal (MDS),-a measurement diffraction result (MDR) is reconstructed from the measurement detector signal (MDS) generated for the at least one object (O), and-the at least one object (O) is detected by means of the reconstructed measurement diffraction result (MDR). -reconstructing a simulated diffraction result (SDR) having the same dimension as the measured diffraction result (MDR) on the basis of a simulated detector signal (SDS) generated for the at least one object (O),-comparing the measured diffraction result (MDR) with the simulated diffraction result (SDR),-outputting a comparison result (CR).

Description

[0001] The invention relates to a detection method for detecting an alarm object in a luggage item and to a computer program product comprising commands for executing such a detection method.

[0002] It is known that luggage monitoring needs to be performed in security-sensitive areas, for example at airports. At airports, this applies in particular to so-called checked luggage, i.e. luggage items that are to be transported in the cargo hold of an aircraft. In known solutions, a transmission analysis is usually carried out on the luggage items. This transmission analysis is, for example, an X-ray computed tomography method. The result of this transmission method can be described as a three-dimensional transmission result and serves to distinguish different objects within the luggage item and to detect at least partially the degree of danger of their presence. In the case of known security systems, on the basis of the transmission analysis and the evaluation of the three-dimensional transmission result, a decision is made as to whether it is possible that an alarm object is contained in the luggage item. This can also be understood as the output of an alarm signal.

[0003] In order to ensure that not every luggage item in which a possible alarm object has been found actually has to be opened and checked actually manually, many airports use a second, downstream detection method. This usually involves an optical examination of the transmission result by a person or possibly an algorithm. In some cases, a diffraction analysis using an X-ray diffraction device is also carried out. Every luggage item that triggers an alarm is sent to such a diffraction analysis. Here, an analysis of the luggage item is carried out by means of X-rays, wherein the X-rays diffracted within the luggage item are captured by a corresponding detector and output as a detector signal. The detector signal itself only has a limited significance and serves to reconstruct a diffraction result on the basis of the detector signal. This reconstruction is usually achieved by converting the detector signal from the detector coordinate system into the coordinate system of the luggage item. In a final step, in known methods, the diffraction result reconstructed in this way is compared with information from an alarm database in order to determine whether the material of the alarm object is an alarm material. In other words, in this way it is possible to confirm or cancel an alarm with regard to the material composition of the alarm object.

[0004] The disadvantage of the known solution is that the alarm database displays the diffraction results in an ideal manner, in particular in the form of impulse transmission functions. When performing a diffraction analysis in real life, the detector signals are influenced by a large number of environmental parameters, luggage parameters and device parameters of the diffraction device, so that there can sometimes be a considerable deviation between the ideal diffraction results from the alarm database on the one hand and the real measurement results as diffraction results on the other hand. For example, such a deviation can be caused by an absorption of parts of the radiation, so that the data contain photon noise. Another example is a mixing of signals from different materials. As a result, these differences lead to an increased number of false alarms, i.e. to an actually unnecessary manual inspection of the luggage items.

[0005] It is an object of the present application to remedy the above-mentioned disadvantages at least partially. In particular, it is an object of the present application to improve the measurement results in a cost-effective and simple manner and to avoid or at least reduce false alarms with a high degree of certainty.

[0006] The above-mentioned objects are achieved by a detection method having the features of claim 1 and by a computer program product having the features of claim 12. Further features and details of the present application are disclosed in the dependent claims, the description and the drawings. Of course, features and details described in connection with the detection method according to the present application also apply in connection with the computer program product according to the present application and vice versa, so that with regard to the disclosure content, mutual references can be made or can always be made with reference to the respective aspects of the present application.

[0007] According to the present application, a detection method is used to detect an alarm object in a luggage item. To this end, the detection method comprises the following steps: - performing a transmission analysis on the luggage item, generating a transmission result, - segmenting the transmission result in order to identify objects, - generating a simulated detector signal for at least one identified object, assigning an alarm material by means of a diffraction simulation specific to the diffraction analysis, - performing a diffraction analysis on the same luggage item, generating a measured detector signal, - reconstructing a measured diffraction result for the at least one object from the generated measured detector signal, - reconstructing a simulated diffraction result having the same dimension as the measured diffraction result on the basis of the simulated detector signal generated for the at least one object, - comparing the measured diffraction result with the simulated diffraction result, - outputting a comparison result.

[0008] The detection method according to the application builds on known solutions, in which a combination of two different detection devices is used. As in the prior art, this involves a combination of a transmission device for performing a transmission analysis and a diffraction device for performing a diffraction analysis. This combination is also referred to as a system-of-systems (SoS) device.

[0009] As in the known solutions, the detection method begins with performing a transmission analysis, for example by means of a computed tomography device, in order to examine all items of luggage in the checked luggage items of an airport. During this performance, a transmission result is generated for each item of luggage, which in many cases is an image of the density distribution. As is also known, this, for example three-dimensional, transmission result is now segmented, so that individual objects within the transmission result are identified and distinguished from other objects. During this segmentation for object detection, for example individual volume elements, so-called voxels, which together form a homogeneous volume, can be combined and thus assigned with high probability to a single specific object. As a result, one or more objects within the transmission result can be identified on the basis of this transmission result.

[0010] The idea according to the application is based on the fact that at least one identified object is subjected to a simulation. This simulation is a diffraction simulation, i.e. a simulation is made to generate a simulated detector signal, for example the signal that would be expected if the object were subjected to a real diffraction analysis in a second step. In other words, in this step, the subsequent diffraction analysis is simulated and a simulated detector signal is generated by this simulation. This means that for this object within the transmission result, at least one alarm material is assumed for the subsequent diffraction simulation. Thus, a projection into the future is made in the case that the object consists of or at least contains the assumed alarm material. This projection results in a simulated detector signal, for example a signal that would be generated if the diffraction analysis were performed in real life. The diffraction simulation thus simulates the actual analysis sequence of the diffraction analysis to be made subsequently and thus makes a prediction based on at least one assumed alarm material.

[0011] After this simulation step according to the application, or at the same time, the diffraction analysis can now actually be performed on the diffraction device, so that a measured measurement detector signal is generated. In the next two steps, as has already been explained in the introduction, a simulated diffraction result and a measured diffraction result are generated from the respective detector signals, i.e. from the simulated detector signal and from the measurement detector signal, by a reconstruction step. This reconstruction may, for example, include the above-mentioned back-projection from the detector coordinate system into the coordinate system of the item of luggage. The simulated diffraction result and the measured diffraction result preferably have the same dimension, i.e. they are, for example, both three-dimensional. In principle, other dimensions can also be envisaged, for example to perform the method in 2 dimensions or even in more than 3 dimensions.

[0012] The core idea of the present application is that in the last comparison step, the measured diffraction result no longer has to be compared to a standardized, in particular idealized, alarm database. Rather, a comparison is made to a separate simulated diffraction result specific to the exactly identified object with the assumed alarm material. That is, in the comparison step, not only is the diffraction result determined separately for the current detection situation, but also the comparison value in the form of the simulated diffraction result is determined separately. Finally, the comparison result is output, in particular in a qualitative manner, i.e. with a statement as to whether the object actually comprises the assumed and assigned alarm material.

[0013] As an advantage over known methods, the idealized alarm database is thus no longer used for the final execution of the comparison to determine whether an alarm object is present. Rather, the comparison value is also specifically and separately generated for the detected detection situation itself. This generation is ensured by the simulation step explained above, which ensures that the diffraction simulation takes into account specific properties and individual distortions in the diffraction analysis being carried out, even with the assumed alarm material. In other words, by means of the diffraction simulation, the comparison value is adapted to the individual situation of the object, so that it can predict in an improved manner what the assumed alarm material can actually look like in the form of a simulated detector signal for this object.

[0014] By preferably the same reconstruction of the simulated diffraction result generated on this basis, the detection accuracy is significantly improved, so that the number of false alarms can be reduced and the detection reliability of the alarm material is significantly improved. The diffraction simulation can predict the noise that the diffraction measurement would exhibit, which is also beneficial. This allows a distinction between statistical and systematic deviations, which will improve the detection performance.

[0015] In the detection method according to the application, it is advantageous if, in addition to the at least one identified object with the assigned material, a simulated detector signal is additionally generated for the adjacent region. Since the interaction with the adjacent region around the object also causes impairment of the expected measurement detector signal, this impairment has already been taken into account in the diffraction simulation. Thus, not only the object but also the adjacent volume portion, also referred to as adjacent voxels, are related to the measurement detector signal and thus also have to be taken into account for generating the simulated detector signal in this embodiment. In this way, the accuracy of the simulation can thus be improved, so that the accuracy of the subsequent comparison and the associated reliability of the alarm material detection can also be correspondingly improved.

[0016] In the detection method according to the application, it is also advantageous if a preliminary analysis is carried out on the at least one identified object, in particular with regard to at least one of the following parameters: - the morphology of the at least one identified object, - the density of the at least one identified object.

[0017] The above list is a non-exhaustive list. Of course, in such a preliminary analysis, two or more parameters can also be taken into account in combination. These parameters serve to decide in the preliminary analysis whether the object in principle can be an alarm object. Such a preliminary analysis can therefore also be referred to as a pre-alarm, which leads to the luggage item being directed to a second checking stage in the form of a diffraction analysis. If an alarm object is excluded with a high probability in the preliminary analysis, the luggage item can be classified as "safe" and can bypass the diffraction analysis. Furthermore, the preliminary analysis can provide an indication as to which potential alarm material can be involved. For example, if the morphology, density or other parameters indicate that an explosive material is an alarm material, the alarm material for the generation step in the diffraction simulation can be selected accordingly depending on the result of the preliminary analysis.

[0018] If, in the detection method according to the application, a further analysis is carried out depending on the result of the preliminary analysis, advantages can also arise. Thus, in the simplest case, such a pre-alarm can decide in principle whether the luggage item is to be passed through for a diffraction analysis. However, a corresponding appropriate selection of alarm material or a defined amount of alarm material can also be made. The analysis settings of the diffraction device and / or the reconstruction mechanism can also be adjusted.

[0019] Furthermore, it can be advantageous if, in the detection method according to the application, the diffraction simulation takes into account at least one of the following parameters: - the angular resolution of the diffraction analysis to be carried out, - the spatial resolution of the diffraction analysis to be carried out, which leads to signal mixing, - the attenuation of the scattered signal to be carried out by the diffraction analysis of the other objects in the luggage item and thus of the expected distribution of the (Poisson) noise variance on the momentum transfer axis (energy axis), - the attenuation of the detector response during the diffraction analysis to be carried out.

[0020] The above list is a non-exhaustive list. The parameters to be taken into account depend in particular on the diffraction device actually used and are selected for it. These parameters can also be specifically adapted to the respective luggage item and / or the respective object in order to further improve the accuracy of the diffraction simulation result and thus the quality of the detection method according to the application.

[0021] In the detection method according to the application, it can be further advantageous if a back-projection or reconstruction is performed in order to generate the measured diffraction result and / or the simulated diffraction result. As already explained, this back-projection for the reconstruction step is, for example, a back-projection from the detector coordinate system to the luggage item coordinate system. In particular, the same back-projection algorithm can be used for both the simulated diffraction result and the measured diffraction result. In particular, this is the case if a separate segmentation can be performed on the measured diffraction result to provide a further identification of the object.

[0022] In the detection method according to the application, it is advantageous if the step of segmenting the transmission result is performed on the basis of a homogeneous volume. Of course, other segmentation options can also be envisaged. However, the application of a homogeneous volume provides a simple and quick starting point for performing an effective segmentation in order to identify the object. As already explained with reference to the preliminary analysis, an indication about the respective material can also be provided for selecting the alarm material in the simulation step.

[0023] If, in the detection method according to the application, the detection is performed on at least two different identified objects in the luggage item, advantages also arise. In particular, if not only a single object in the luggage item is detected by segmentation, but also several objects, and these are also identified as potentially dangerous, it can be advantageous to perform the further simulation, reconstruction and comparison steps on all identified alarm objects. When performed in a computer program, this is in particular performed by the respective computer program in a sequential manner, such that all objects are checked step by step for the respective assigned alarm material.

[0024] In the detection method according to the application for at least one identified object, it is also advantageous if at least two different alarm materials can be assigned for the respective generation of at least two different simulated simulated detector signals. In general, not only a single alarm material is possible for an identified object. For example, if there is an indication of explosives, it can be useful to check different explosive materials as alarm materials. With reference to the preceding paragraph, it should be noted here that for different potentially alarm objects within the luggage item, different alarm materials or combinations of alarm materials can of course also be assigned to these steps of the detection method according to the application. In addition to alarm materials, a negative test can in principle also be envisaged, in which a non-alarm material is assigned to this simulation step and the final comparison. In other words, by actively assigning an object to a non-alarm material, the presence of an alarm material can be ruled out depending on the comparison result.

[0025] A further advantage is achieved in that, in the detection method according to the application, when performing the comparison, a classification method is used, in particular one of the following: - correlation-based classification (in particular, one-dimensional feature vectors), - support vector machines (in particular, multi-dimensional feature vectors), - neural networks (in particular, abstract non-linear (and high-dimensional) feature extraction by convolutional layers and their interpretation by fully connected layers).

[0026] The above list is a non-exhaustive list. Of course, combinations of the individual classifications can also be considered.

[0027] Correlation-based binary classification describes the classification of a substance based on a similarity value (similarity measure). Exceeding a defined threshold leads to the generation of an alarm for the material being sought. Falling below the threshold leads to no alarm being generated. One example of a similarity value can be the normalized scalar product of the signal with the corresponding reference.

[0028] Support vector machines (SVM) classify spectra based on several features. These features (e.g. several different similarity measures) are pre-extracted from the signals to be classified. The number of features specifies the dimension of the "feature space" (vector space). The SVM now tries to find the optimal hyperplane (maximum distance of support vectors) in this feature space to separate the objects that generate an alarm from the objects that do not generate an alarm. The separation by the generated hyperplane can be linear (linear kernel) as well as non-linear (polynomial, radial).

[0029] In CNNs (convolutional neural networks), the manual feature extraction is dispensed with. Instead, the weights of a large number of convolution kernels (one-dimensional vectors or matrices) are trained. These convolutions summarize the local environment by filters (convolutions). Pooling layers ensure that the filtered data is reduced, since irrelevant information is discarded. The result is a high-dimensional, non-linear (non-linear using non-linear activation functions: sigmoid, ReLu,...) feature space that has been trained on the filters so that the subsequent classification is as optimal as possible. These features can in theory be used for classification by SVMs. However, usually a "fully connected neural network" is used to interpret the high-dimensional and non-linear feature space and to generate the classification result. Thus, the neural network consists of two parts, the first part for feature extraction (convolutional neural network) and the second part (fully connected neural network) separates the high-dimensional feature space. Both parts are trained simultaneously in the learning process.

[0030] It can be further advantageous if in the detection method according to the application the comparison generates a similarity value which reflects the similarity between the measured diffraction result and the simulated diffraction result. This similarity value is in particular an abstract value which as a single value can simply be compared to a limit value to generate a comparison result. For example, if this similarity value is high, it can be assumed that the similarity is low, and a low value of the similarity value can indicate that the similarity is high. It is thus possible to specify a single defined limit value for the evaluation and output of the comparison result, whereby the similarity is sufficient to confirm the correspondence with the assigned alarm material.

[0031] The subject matter of the application further comprises a computer program product comprising commands which, when the program is run on a computer, cause it to carry out the steps of the detection method according to the application. The computer program product according to the application thus has the same advantages as explained in detail with reference to the detection method according to the application.

[0032] Further advantages, features and details of the application are explained in the following description, in which exemplary embodiments of the application are described in detail with reference to the drawings. The features mentioned in the claims and the specification can in each case, both individually and in any combination, be of importance for the application. In each case schematically: Figure 1 a schematic diagram of a method according to the prior art is shown, Figure 2 an embodiment of the detection method according to the application is shown, Figure 3 a further embodiment of the detection method according to the application is shown, Figure 4 a further embodiment of the detection method according to the application is shown.

[0033] Figure 1It is schematically shown how a detection takes place in the past in a detection system 10, which is here in the form of a System of Systems arrangement. In this case, the detection system 10 is equipped with a transmission arrangement 20, for example a computed tomography scanner. This transmission arrangement 20 performs a transmission analysis TA on a baggage item B, as a result of which a three-dimensional transmission result TR is obtained. By means of segmentation, an object O can be identified, for example in this case a single object O can be identified. In the case of an alarm, the baggage item B can be passed on for further analysis, in this case by means of a diffraction arrangement 30. This diffraction arrangement 30 is in the form of an X-ray diffraction system, whereby the X-rays that are diffracted when the baggage item B is scanned are detected by a corresponding detector and output as a result of a diffraction analysis DA, whereby a measured diffraction result MDS is output. By means of reconstruction, the measured detector signal MDS is back-projected into a measured detector result MDR, whereby ultimately the potential alarm object AO can be compared with an alarm database ADB and the ideal comparison values stored therein. As has already been explained, this has the disadvantage that the actual measurement errors or impairments of the measured detector signal MDS and / or the measured diffraction result MDR are not taken into account in the idealized alarm database ADB.

[0034] Figure 2 One embodiment of the application is shown, which is intended to solve the above-mentioned problems. Here too, a transmission analysis TA is performed by means of the transmission arrangement 20, whereby a three-dimensional transmission result TR can be provided. This enables the object O to be identified again by means of segmentation, and in particular the potential alarm object AO to be identified by means of a preliminary analysis. For example, in the case of such a preliminary analysis, an indication can also be provided about the material group for which the identified potential alarm object AO is checked in a subsequent step.

[0035] In the detection method according to the application, a simulated detector signal SDS is now generated for the identified potential alarm object AO by means of a diffraction simulation DS. In order to perform this diffraction simulation DS, not only is the subsequent diffraction analysis DA simulated, but also an alarm material AM is assumed and assigned to the alarm object AO. This results in a simulated detector signal SDS, which can again be reconstructed by means of back-projection to generate a simulated diffraction result SDR. In other words, this is a simulation part, which outputs a prediction and thus the expected simulated diffraction result SDR of the potential alarm object AO with the assigned alarm material AM.

[0036] The diffraction analysis DA is also carried out in a known manner in order to be able to detect the measured detector signal MDS. This measured detector signal MDS is also reconstructed again by means of back-projection, preferably in the same or essentially the same manner as the simulated detector signal SDS, into a measured diffraction result MDR. In contrast to known solutions, for example as described in the above-mentioned patent application, the measured diffraction result MDR is now compared with the simulated diffraction result SDR.Figure 1 As shown, instead of a comparison with an idealized alert database ADB, a comparison is made between a specific measured diffraction result MDR for the alert object AO and a specific, and thus individual, simulated diffraction result SDR for the baggage item B and the alert object AO. Thus, for the same alert material AM, for different alert objects AO and / or different baggage items B, the simulated diffraction result SDR will be different from the current case. Thus, the simulated diffraction result SDR changes for each baggage item B and each object O, so that for each diffraction analysis DA and final comparison, there is always a simulated diffraction result SDR that is individualized for this case. In a final step, a comparison result CR is output, which indicates in particular whether the alert object AO has a sufficiently high similarity to the assumed and assigned alert material AM.

[0037] Figure 3 A possible further development of the embodiment of Figure 2 is shown. Here, when considering and applying the diffraction simulation DS, not only the alert object AO itself is considered, but also the adjacent region adjacent to the alert object AO. Here, the three-dimensional transmission result TR is schematically represented by individual cubic-shaped voxels, so that in the simulation and generation of the simulated detector signal SDS, the adjacent voxels adjacent to the alert object AO can now also be considered in the diffraction simulation DS. In this case, the adjacent region adjacent to the alert object AO is preferably introduced into the diffraction simulation DS together with the material M as non-alert material.

[0038] In the embodiment of Figure 3 , the comparison result CR is additionally defined more in detail. In this case, this is provided with a similarity value SV in particular having a scalar form. This similarity value can be a simple parameter, which can be compared with a limit value to indicate the similarity and thus easily distinguish whether the alert object AO has a similarity to the assigned alert material AM.

[0039] Figure 4It is also shown that several possibilities for further developments of the detection method according to the application are possible. On the one hand, it can be clearly seen here that the diffraction simulation DS is not performed only once for the alarm object AO. Rather, for example, different alarm materials AM are assigned here, so that the diffraction simulation DS is also performed several times. This results in two different simulated detector signals SDS and thus also two different simulated diffraction results SDR. In the final result, one measured diffraction result MDR is now compared several times with the different simulated diffraction results SDR, so that different comparison results CR are obtained for each alarm material AM. In this way, the potential alarm object AO can be checked for different alarm materials AM with the same effort, since the diffraction analysis DA only has to be performed once and the corresponding checking cases are integrated into several applications of the diffraction simulation DS.

[0040] The above description of embodiments only describes the application in the context of examples. Individual features of the individual embodiments can be freely combined with one another, if technically expedient, without departing from the scope of the application.

[0041] List of reference signs 10 detection system 20 transmission device 30 diffraction device TA transmission analysis TR transmission result DA diffraction analysis MDR measured diffraction result SDR simulated diffraction result CR comparison result SV similarity value DS diffraction simulation SDS simulated detector signal MDS measured detector signal M material AM alarm material O object AO alarm object B baggage item ADB alarm database

Claims

1. A detection method for detecting an alarm object (AO) in a baggage item (B), comprising the following steps: - performing a transmission analysis (TA) on the baggage item (B), generating a transmission result (TR), - segmenting the transmission result (TR) in order to identify objects (O), - generating a simulated simulated detector signal (SDS) for at least one identified object (O), assigning an alarm material (AM) by means of a diffraction simulation (DS) specific to a diffraction analysis (DA), - performing a diffraction analysis (DA) on the same baggage item (B), generating a measured detector signal (MDS), - reconstructing a measured diffraction result (MDR) from the measured detector signal (MDS) generated for the at least one object (O), - reconstructing a simulated diffraction result (SDR) having the same dimension as the measured diffraction result (MDR) on the basis of the simulated detector signal (SDS) generated for the at least one object (O), - comparing the measured diffraction result (MDR) with the simulated diffraction result (SDR), - outputting a comparison result (CR).

2. The detection method according to claim 1, wherein, The simulated simulated detector signal (SDS) is generated for adjacent regions in addition to the at least one identified object (O) with the assigned material (M).

3. The detection method according to one of the preceding claims, wherein, A preliminary analysis is performed on the at least one identified object (O), in particular with respect to at least one of the following parameters: - the morphology of the at least one identified object (O), - the density of the at least one identified object (O).

4. The detection method according to claim 3, wherein, The further analysis is performed depending on the result of the preliminary analysis.

5. The detection method according to one of the preceding claims, wherein, The diffraction simulation (DS) takes into account at least one of the following parameters: - the angular resolution with which the diffraction analysis (DA) is performed, - the spatial resolution with which the diffraction analysis is performed, which can result in signal mixing, - the attenuation of the scattered signal of the diffraction analysis (DA) to be performed on other objects (O) in the baggage item (B) and the resulting attenuation of the variance of the (Poisson) noise in the expected distribution over the momentum transfer axis (energy axis), - the attenuation of the detector response during the diffraction analysis (DA).

6. The detection method according to one of the preceding claims, wherein, A back-projection or reconstruction is performed in order to generate the measured diffraction result (MDR) and / or the simulated diffraction result (SDR).

7. The detection method according to one of the preceding claims, wherein, The step of segmenting the transmission result (TR) is performed on the basis of a uniform volume.

8. The detection method according to one of the preceding claims, wherein, The detection is performed for at least two different identified objects (O) in the baggage item (B).

9. The detection method according to one of the preceding claims, wherein, For the at least one identified object (O), at least two different alarm materials (AM) are assigned for the respective generation of at least two simulated simulated detector signals (SDS).

10. The detection method according to one of the preceding claims, wherein, When performing the comparison, a classification method is used, in particular one of the following classification methods: - classification on the basis of correlations, - support vector machines, - neural networks.

11. The detection method according to one of the preceding claims, wherein, The comparison generates a similarity value (SV) which reflects the similarity between the measured diffraction result (MDR) and the simulated diffraction result (SDR).

12. A computer program product comprising commands which, when said program is run on a computer, cause the computer to perform the steps of the detection method having the features of one of claims 1 to 11.

Images