A method for detecting a long-distance hidden camera device based on sparse representation

Abstract

The present application belongs to the technical field of photoelectric countermeasure, and relates to a detection technology for a hidden camera device, and particularly provides a long-distance detection method for a hidden camera device based on sparse representation, which is used to overcome the problem of low recognition accuracy in the prior art; in the present application, after sparse representation of an image through wavelet transform, a measurement matrix group is obtained by compressing and projecting the sparse image using a random measurement matrix; a measurement matrix without static background and dynamic background information is obtained by performing a series of difference addition operations on the matrix group; then, the orthogonal matching pursuit algorithm is used to reconstruct and recover the measurement matrix to obtain a recovered image; finally, the cat-eye target area is screened out according to the image pixel value and the connected area area judgment method, so as to realize the cat-eye effect target recognition method based on sparse representation; in this way, the image data redundancy is greatly reduced, the recognition accuracy of the long-distance cat-eye target is effectively improved, and the anti-detection of the hidden camera device is realized, so as to eliminate the potential information leakage danger.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of photoelectric countermeasure, and relates to a detection technology for a hidden camera device, and particularly provides a long-distance detection method for a hidden camera device based on sparse representation. BACKGROUND

[0002] With the wide application of pinhole cameras, miniature cameras and other photoelectric detection equipment, various activities are facing more and more serious information leakage risks, and the leakage of confidential information will cause great losses in military, social and commercial fields. At present, the most common method for detecting a hidden camera device is a radio wave detection method, which mainly judges whether a hidden camera device exists by detecting radio wave field strength. However, the main problem of this method is that the detection wavelength range is too wide, and it cannot be targeted at the most needed 2.4GHz WIFI signal and SIM card signal, which is easy to cause false positives and results in low detection accuracy. SUMMARY

[0003] The present application aims to solve the problem of low recognition accuracy caused by false positives of the existing radio wave detection method, and provides a long-distance detection method for a hidden camera device based on sparse representation. The present application uses a wavelet transform matrix to perform sparse representation on an image, and then uses a random measurement matrix to compress and project the sparse image to obtain a measurement matrix group. A series of difference addition operations are performed on the matrix group to obtain a measurement matrix without static background and dynamic background information. Then, an orthogonal matching pursuit algorithm is used to reconstruct and recover the measurement matrix to obtain a recovered image. Finally, a cat-eye target area is screened out according to the image pixel value and connected area area judgment method, so as to realize a cat-eye effect target recognition method based on sparse representation. The present application greatly reduces the image data redundancy and effectively improves the recognition accuracy of the long-distance cat-eye target, thereby realizing the counter-detection of the hidden camera device and eliminating the potential information leakage risk.

[0004] To achieve the above object, the technical scheme adopted by the present application is as follows:

[0005] A long-distance detection method for a hidden camera device based on sparse representation, characterized in that it comprises the following steps:

[0006] Step 1: two pairs of active images and passive images are jointly collected by a camera and a laser, which are active image P1 and passive image N1, active image P2 and passive image N2, respectively.

[0007] Step 2: the same wavelet transform is performed on the active image P1 and the passive image N1, the active image P2 and the passive image N2 by using a wavelet transform matrix, to obtain sparse matrices PP1 and NN1, PP2 and NN2.

[0008] Step 3, according to the sparse representation theory, the same compression projection is carried out on the sparse matrices PP1 and NN1, PP2 and NN2 by using a Gaussian random matrix, and measurement matrices PPP1 and NNN1, PPP2 and NNN2 are obtained;

[0009] Step 4, difference operation is carried out on the measurement matrices PPP1, NNN1, PPP2 and NNN2, and difference matrices Y1, Y2 and Y3 are obtained, specifically Y1=PPP1-NNN1, Y2=PPP2-NNN2 and Y3=PPP2-NNN1;

[0010] Step 5, the dynamic background matrix is calculated according to the difference matrices Y1, Y2 and Y3:

[0011] Y background = Y1-Y2+Y2-Y3+|Y1-Y3|

[0012] Further, the measurement matrix Y only-cateye = Y1+Y2+Y3-Y background is obtained without dynamic background.

[0013] Step 6, the orthogonal matching pursuit recovery algorithm is used to reconstruct and recover the measurement matrix Y only-cateye , and a recovered image Isub is obtained.

[0014] Step 7, for the recovered image Isub, the target information is screened according to the highest pixel value in the image and the 3 / 4 determination method, and a recovered image Iend is obtained.

[0015] Step 8, for the recovered image Iend, the area of the connected region is determined, and then the cat eye target position information is screened out.

[0016] Step 9, according to the cat eye target position information, the cat eye target is marked in the original active image.

[0017] Further, in step 7, the 3 / 4 determination method is specifically: all pixel values less than 3 / 4 times the highest pixel value of the image are set to 0, and pixel points higher than or equal to 3 / 4 times the highest pixel value are kept unchanged.

[0018] Further, in step 8, the specific process is: all pixel values of the connected region with an area less than 1 / 2 times the maximum area and the connected region with an area less than or equal to 2 pixel points are set to 0, and the remaining non-zero connected region is taken as the cat eye target region to obtain the cat eye target position information.

[0019] Based on the above technical solutions, the present application has the following advantages:

[0020] The application provides a method for detecting a long-distance hidden camera based on sparse representation, which comprises the following steps: sparse representation of an image through wavelet transform, compression projection of the sparse image by using a random measurement matrix to obtain a measurement matrix group, a series of difference addition operations on the matrix group to obtain a measurement matrix without static background and dynamic background information, reconstruction of the measurement matrix by using an orthogonal matching pursuit algorithm to obtain a recovered image, and screening of a cat-eye target area according to the pixel value of the image and a connected area area judgment method, so as to realize a cat-eye effect target recognition method based on sparse representation, greatly reduce the image data redundancy, effectively improve the recognition accuracy of the long-distance cat-eye target, and further realize anti-detection of the hidden camera and elimination of potential information leakage danger. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 It is a flowchart of the method for detecting a long-distance hidden camera based on sparse representation.

[0022] Figure 2 It is two pairs of active images and passive images in step 1 in the embodiment.

[0023] Figure 3 It is two pairs of sparse matrix groups in step 2 in the embodiment.

[0024] Figure 4 It is two pairs of measurement matrix groups in step 3 in the embodiment.

[0025] Figure 5 It is a difference matrix Y1, Y2 and Y3 in step 4 in the embodiment.

[0026] Figure 6 It is a measurement matrix Y without background information in step 5 in the embodiment. only-cateye .

[0027] Figure 7 It is a recovered image Isub in step 6 in the embodiment.

[0028] Figure 8 It is a result of the 3 / 4 judgment method screening in step 7 in the embodiment.

[0029] Figure 9 It is a cat-eye target area position information map in step 8 in the embodiment.

[0030] Figure 10 It is a cat-eye target detection result map in step 9 in the embodiment. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical scheme and beneficial effects of the application more clear, the application is further described in detail below with reference to the drawings and embodiments.

[0032] When light irradiates a hidden camera device, a small part of the light will return to the original path in reverse with an intensity usually 2 to 4 orders of magnitude stronger than diffuse reflection, emitting a bright light, which is like a cat's pupil shrinking into a thin slit, this phenomenon is called "cat eye effect", and such hidden camera device is called cat eye target; based on this, the embodiment provides a method for remotely detecting a hidden camera device based on sparse representation, which utilizes the sparse features of the cat eye effect in a natural background, adopts a cat eye effect target recognition method based on sparse representation to process the active and passive images captured by an imaging system, and finally realizes positioning and labeling of the cat eye target; the flow of the embodiment is shown in Figure 1 , and specifically includes the following steps:

[0033] Step 1: Collecting two pairs of active and passive images by a camera and a laser, wherein the image collected when the laser is turned on is an active image, and the image collected when the laser is turned off is a passive image; as shown in Figure 2 , wherein P1 and N1 are the first pair of active and passive images, P2 and N2 are the second pair of active and passive images, P1 and P2 are active images, and N1 and N2 are passive images; simultaneously collecting two pairs of active and passive images for cat eye detection can to some extent ignore the influence of dynamic background on the detection result, effectively improving the detection accuracy;

[0034] Step 2: Using a wavelet transform matrix to perform the same wavelet transform on the two pairs of active and passive images in step 1 to obtain two pairs of sparse matrix groups; as shown in Figure 3 , PP1 is the sparse matrix after wavelet transform of P1, NN1 is the sparse matrix after wavelet transform of N1, and PP2 and NN2 are the same; wavelet transform can sparsely represent the image, remove a large amount of invalid information, and reduce the image data storage amount;

[0035] Step 3: According to the compressed sensing theory, using the generated Gaussian random matrix to perform the same compression projection on the two pairs of sparse matrix groups in step 2 to obtain two pairs of measurement matrix groups; as shown in Figure 4 , PPP1 is the measurement matrix obtained by compression projection of PP1, NNN1 is the measurement matrix obtained by compression projection of NN1, and PPP2 and NNN2 are the same; the purpose of this step is to compress and project the matrix group, and the original image in step 1 is truly sparsely represented; in the embodiment, when the image compression rate is set to 0.5, the width of the compressed image is only half of that of the original image;

[0036] Step 4: Performing three difference operations on the two pairs of measurement matrix groups in step 3 to obtain difference matrices Y1, Y2 and Y3; three difference operations can filter out static background information to highlight the cat eye target, as shown in Figure 5The difference operation is specifically: Y1=PPP1-NNN1, Y2=PPP2-NNN2, and Y3=PPP2-NNN1; for the dynamic background case, the positions of the cat-eye target information in the difference matrices Y1, Y2, and Y3 are the same, and thus the dynamic background can be filtered out through subsequent operations;

[0037] In step 5, the cat-eye target information and unknown dynamic background information in each difference matrix in step 4 have the same position and the same strong form (the pixel value is generally the highest pixel value of the whole image); therefore, the difference between any two difference matrices contains only dynamic background information; based on this, the sum of the absolute values of the difference between any two difference matrices in step 4 is taken as a dynamic background matrix, and the dynamic background matrix is specifically:

[0038] Y background =|Y1-Y2|+|Y2-Y3|+|Y1-Y3|

[0039] Further, a measurement matrix Y only-cateye =Y1+Y2+Y3-Y background is obtained, as shown in Figure 6 .

[0040] In step 6, the measurement matrix Y only-cateye in step 5 is reconstructed and recovered by using an orthogonal matching pursuit (OMP) recovery algorithm, and the recovered image Isub is as shown in Figure 7 , which has the same size as the original image in step 1 but a greatly reduced data volume, and the static background information and most of the dynamic background information are filtered out, highlighting the main feature cat-eye target information, as shown by the white bright spots in Figure 7 .

[0041] In step 7, for the image Isub recovered in step 6, the target information is selected according to the highest pixel value in the image and the 3 / 4 determination method, and all pixel values lower than 3 / 4 times the highest pixel value are set to 0, and those higher than or equal to 3 / 4 times the highest pixel value are retained; as shown in Figure 8 , the left subgraph Iend is the result of the selection according to the 3 / 4 determination method, and after the selection, two different connected regions are retained, and most of the noise information is removed, and the vertical axis Z of the right subgraph represents the pixel value of the x,y pixel point, and it can be seen that the pixel values of the two connected regions are different;

[0042] In step 8, for the image Iend in step 7, the area of the connected region is determined, and the cat-eye target information is selected; all pixel values of the connected regions with an area less than 1 / 2 times the maximum area and an area less than or equal to 2 pixel points are set to 0 and are not regarded as target regions, and the remaining non-zero connected regions are regarded as cat-eye target regions, and the cat-eye target position information is obtained; as shown in Figure 9As shown, the white bright spot area in the left sub-image corresponds to the area with the highest Z value in the right sub-image, which is the cat's eye target area finally determined by the algorithm;

[0043] Step 9: Based on the cat's eye target location information from Step 8, mark the cat's eye target in the active image P1 from Step 1, such as... Figure 10 As shown.

[0044] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. A method for long-distance detection of covert filming devices based on sparse representation, characterized in that, Includes the following steps: Step 1: The camera and laser jointly acquire two pairs of active and passive images, namely active image P1 and passive image N1, and active image P2 and passive image N2. Step 2: Use wavelet transform matrices to perform the same wavelet transform on active image P1 and passive image N1, and active image P2 and passive image N2 respectively, to obtain sparse matrices PP1 and NN1, PP2 and NN2. Step 3: Based on the compressed sensing theory, use Gaussian random matrices to perform the same compressed projection on sparse matrices PP1 and NN1, PP2 and NN2 respectively to obtain measurement matrices PPP1 and NNN1, PPP2 and NNN2. Step 4: Perform difference operations on the measurement matrices PPP1, NNN1, PPP2 and NNN2 to obtain difference matrices Y1, Y2 and Y3, specifically: Y1 = PPP1 - NNN1, Y2 = PPP2 - NNN2, Y3 = PPP2 - NNN1; Step 5: Calculate the dynamic background matrix based on the difference matrices Y1, Y2, and Y3. Y background =|Y1-Y2|+|Y2-Y3|+|Y1-Y3| This yields a measurement matrix without dynamic background: Y only-cateye =Y1+Y2+Y3-Y background ; Step 6: Apply the orthogonal matching pursuit recovery algorithm to the measurement matrix Y. only-cateye Reconstruction and restoration are performed to obtain the restored image Isub; Step 7: For the recovered image Isub, based on the highest pixel value in the image, filter the target information according to the 3 / 4 judgment method to obtain the recovered image Iend; Step 8: For the restored image Iend, determine the area of ​​the connected regions, and then filter out the target location information of the cat's eye; Step 9: Mark the cat's eye target in the original active image based on the cat's eye target location information.

2. The method for long-distance detection of covert filming devices based on sparse representation as described in claim 1, characterized in that, In step 7, the 3 / 4 determination method specifically involves setting all pixel values ​​that are less than 3 / 4 times the highest pixel value of the image to 0, while keeping the pixel values ​​that are greater than or equal to 3 / 4 times the highest pixel value unchanged.

3. The method for long-distance detection of covert filming devices based on sparse representation as described in claim 1, characterized in that, In step 8, the specific process is as follows: set all pixel values ​​of connected regions with an area less than 1 / 2 times the area of ​​the largest region and connected regions with an area less than or equal to 2 pixels to 0, and use the remaining non-zero connected regions as the cat eye target region to obtain the cat eye target location information.

Images