Intelligent security check system and method based on directional code detection
By introducing directional code detection and luggage graphic recognition technology into the security inspection system, the correlation matching of mobile phone feature code, face feature information and luggage feature information is achieved, and the problems of low efficiency and poor matching of existing security inspection systems are solved, and security inspection is improved The accuracy and efficiency of the baggage traceability function is provided.
Patent Information
- Application Number
- CN202410989578.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-29
AI Technical Summary
The existing security inspection system is inefficient and the probability that the inspected object does not match the luggage it carries is high, resulting in insufficient security inspection accuracy.
Using an intelligent security inspection system based on directional code detection, the directional code detection unit, face recognition unit and luggage graphic recognition unit are set at the security inspection entrance and exit, and the mobile phone feature code, face feature information and luggage feature information are combined with the central processing unit. The associated matching of the object to be inspected ensures the matching of the object to be inspected and the luggage location is notified through the luggage collection device.
It improves security inspection efficiency, reduces the probability that the inspected object does not match the luggage they carry, increases the accuracy of security inspection, and realizes the luggage traceability ability.
Smart Images

Figure CN119049164A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent security inspection, and in particular to an intelligent security inspection system and method based on directional code detection. Background Art
[0002] The existing security inspection systems and methods all involve manually checking the person and ID of the person being inspected at the security inspection entrance. After that, the person being inspected puts the luggage on the luggage conveyor belt and enters the luggage security inspection device. After the person being inspected passes the human body detection device, he is manually re-inspected. After the re-inspection, the luggage is picked up at the corresponding luggage claim belt. Some security inspection systems automate the person and ID inspection at the entrance, and complete identity verification by matching and verifying the extracted ID information through face recognition.
[0003] The existing security inspection system is restricted by the one-to-one correspondence between the security inspection entrance, baggage conveyor belt and security inspection exit, resulting in low security inspection efficiency. Summary of the invention
[0004] In order to solve the problem of low security inspection efficiency in the prior art, the present invention provides an intelligent security inspection system and method based on directional code detection. The present invention can improve the security inspection efficiency, reduce the probability of mismatch between the inspected object and the carried luggage, and increase the accuracy of security inspection. The technical solution adopted by the present invention is as follows:
[0005] An intelligent security inspection system based on directional detection code includes multiple security inspection entrances, multiple security inspection exits and a central processing unit. A baggage conveying detection system is arranged in the area between the security inspection entrances and the security inspection exits. The baggage conveying detection system includes a baggage conveyor belt and a baggage reclaiming device. The baggage reclaiming device is arranged at the rear end of the baggage conveyor belt. The baggage conveyor belt is provided with a baggage detection device.
[0006] The security check entrance and the security check exit are both provided with a directional code detection unit, a face recognition unit, and a luggage pattern recognition unit;
[0007] The directional code detection unit is used to simulate the induced signal of the mobile operator to realize the active code detection of the mobile communication terminals in the area and obtain the mobile phone feature code information;
[0008] The face recognition unit is used to recognize facial feature information and determine the identity of the face;
[0009] The luggage graphic recognition unit is used to identify luggage feature information and determine unique features of the luggage; the central processing unit associates the mobile phone feature code information, facial feature information, and luggage feature information obtained at the security check entrance to obtain associated information, and associates the mobile phone feature code information, facial feature information, and luggage feature information obtained at the security check exit to obtain associated information, and then matches the associated information at the security check exit with the associated information at the security check entrance. If there is matching associated information, the security check exit will release the luggage; otherwise, the security check exit will not release the luggage and output a warning prompt signal.
[0010] Optionally, the baggage retrieval device includes multiple baggage retrieval belts. The central processing unit identifies the retrieval belt position of the luggage based on the luggage feature information, and sends corresponding luggage location information to the mobile phone communication terminal corresponding to the mobile phone feature code information associated with the luggage feature information based on the associated information obtained at the security check entrance, to notify the person being inspected of the retrieval belt position of his or her luggage.
[0011] Optionally, the central processing unit further assigns feature codes to the baggage feature information, and associates the feature codes with detection patterns of the baggage detection device.
[0012] Optionally, the number of the security check entrances and security check exits is N, where N is an integer greater than 1, and each security check entrance is provided with a baggage conveyor belt, and the N baggage conveyor belts converge together and pass through the baggage detection device together, and the number of the baggage retrieval belts is greater than N.
[0013] Optionally, the central processing unit executes the following program:
[0014] Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s;
[0015] No. i The description features of all sampling points corresponding to the sample unit are:
[0016] Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points;
[0017] Step 2: Model the training set for the example unit and define the training set
[0018] Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients
[0019]
[0020] Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type;
[0021] according to use Function updates parsing dictionary
[0022]
[0023] in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition;
[0024] according to Update dictionary
[0025]
[0026] Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class;
[0027] If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ];
[0028] Calculate the comprehensive dictionary D * =w1D+w2P;
[0029] Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P;
[0030] Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s};
[0031] ψ(uni i )=[Z(uni1), Z(uni2),...,Z(uni s )];
[0032] Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame;
[0033] Z(uni s )={z p :p};
[0034]
[0035] z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ;
[0036] c j =argmin||X j -D * c j ||+λ||c j ||;
[0037] c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ;
[0038] Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
[0039] An intelligent security inspection method based on directional code detection is implemented based on any of the above systems and includes the following steps:
[0040] Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s;
[0041] No. i The description features of all sampling points corresponding to the sample unit are:
[0042] Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points;
[0043] Step 2: Model the training set for the example unit and define the training set
[0044] Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients
[0045]
[0046] Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type;
[0047] according to use Function updates parsing dictionary
[0048]
[0049] in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition;
[0050] according to Update dictionary
[0051]
[0052] Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class;
[0053] If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ];
[0054] Calculate the comprehensive dictionary D * =w1D+w2P;
[0055] Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P;
[0056] Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s};
[0057] ψ(uni i )=[Z(uni1), Z(uni2),...,Z(uni s )];
[0058] Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame;
[0059] Z(uni s )={z p :p};
[0060]
[0061] z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ;
[0062] c j =argmin||X j -D * c j ||+λ||c j ||;
[0063] c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ;
[0064] Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
[0065] The beneficial effects of the present invention are:
[0066] 1. The present invention can improve the efficiency of security inspection, reduce the probability that the inspected object does not match the luggage carried, and increase the accuracy of security inspection.
[0067] 2. The present invention can trace the source of the safety problem of the luggage of the inspected object. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] The present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0069] Figure 1 Schematic diagram of the intelligent security inspection system based on directional detection code. DETAILED DESCRIPTION
[0070] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0071] An intelligent security inspection system based on directional code detection includes:
[0072] N security check entrances, N-in-N baggage security check conveyor belts connected to the security check entrances, a baggage safety detection device connected to the baggage security check conveyor belts, an N-in-N baggage conveyor belt connected to the back of the baggage security check detection device, and a baggage reclaiming device arranged at the rear end of the N-in-N baggage conveyor belts, where N is a positive integer greater than 1;
[0073] The security entrance and the security exit are both provided with a directional code detection unit, a face recognition unit, a baggage pattern recognition unit and a central processing unit;
[0074] The device is used to simulate the induced signal of the mobile operator, realize the active code detection of the mobile communication terminals in a certain area, and obtain the mobile phone feature code information;
[0075] The face recognition unit is used to recognize facial feature information and determine the identity of the face;
[0076] The luggage graphic recognition unit is used to identify the luggage feature information and determine the unique features of the luggage; the central processing unit matches and associates the mobile phone feature code information, the face feature information, and the luggage feature information;
[0077] The central processing unit at the security check exit matches and correlates the mobile phone feature code information, the face feature information, and the luggage feature information. If there is a match, the luggage is released; if there is no match, the luggage is not released and a warning prompt signal is output.
[0078] The working principle of this system is as follows:
[0079] This system changes the current scheme of detecting people and luggage together, and subverts the current one-to-one mode of security checkpoints, luggage conveyor belts, and luggage detection devices. The association of people, ID cards, and mobile phone feature signals begins at the security checkpoint. When placing luggage, the association of luggage features is added. When detecting luggage, the association of luggage detection graphic features (generally photo and perspective images) is added. This improves the security inspection efficiency of the inspected object and prevents the inspected object from taking or missing luggage by mistake when leaving the security checkpoint.
[0080] Furthermore, the matching and correlation information of the mobile phone feature code information, the face feature information and the luggage feature information is made consistent at the security check entrance and the security check exit, and the preferred solution prevents the problem of taking the wrong luggage.
[0081] Furthermore, the baggage claim device includes KB baggage claim belts, and the central processing unit sends the matched baggage location information to the mobile communication terminal corresponding to the mobile phone feature code information matching the baggage feature information according to the matching association information, so as to notify the inspected object of the location information of the baggage claim belt, KB ≥ N. The preferred solution can further improve the inspected object's ability to know the baggage location and the security inspection efficiency.
[0082] Furthermore, the central processing unit also assigns feature codes to the baggage feature information, and associates the feature codes with the detection patterns of the baggage safety detection device. Adding the detection patterns to the association family can improve the accuracy of baggage matching.
[0083] Furthermore, the central processing unit executes the following program:
[0084] Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMGi Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s;
[0085] No. i The description features of all sampling points corresponding to the sample unit are:
[0086] Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points;
[0087] Step 2: Model the training set for the example unit and define the training set
[0088] Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients
[0089]
[0090] Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type;
[0091] according to use Function updates parsing dictionary
[0092]
[0093] in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition;
[0094] according to Update dictionary
[0095]
[0096] Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class;
[0097] If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ];
[0098] Calculate the comprehensive dictionary D * =w1D+w2P;
[0099] Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P;
[0100] Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s};
[0101] ψ(uni i )=[Z(uni1), Z(uni2),...,Z(uni s )];
[0102] Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame;
[0103] Z(uni s )={z p :p};
[0104]
[0105] z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ;
[0106] c j =argmin||X j -D * c j ||+λ||c j ||;
[0107] c j ∈RM×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ;
[0108] Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
[0109] In the preferred embodiment, the image frames or image sub-segments required by the present invention can be extracted through a simple and efficient image cutting and classification method, thereby improving the efficiency and accuracy of the security inspection algorithm. An intelligent security inspection method based on directional detection code includes:
[0110] Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s;
[0111] No. i The description features of all sampling points corresponding to the sample unit are:
[0112] Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points;
[0113] Step 2: Model the training set for the example unit and define the training set
[0114] Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients
[0115]
[0116] Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type;
[0117] according to use Function updates parsing dictionary
[0118]
[0119] in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition;
[0120] according to Update dictionary
[0121]
[0122] Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class;
[0123] If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ];
[0124] Calculate the comprehensive dictionary D * =w1D+w2P;
[0125] Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P;
[0126] Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s};
[0127] ψ(uni i )=[Z(uni1), Z(uni2),...,Z(uni s )];
[0128] Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame;
[0129] Z(uni s )={z p :p};
[0130]
[0131] z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ;
[0132] c j =argmin||X j -D * c j ||+λ||c j ||;
[0133] c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ;
[0134] Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
[0135] Furthermore, step 2 also includes tracking the baggage graphic frame, and matching and adding the baggage security detection device detection graphic information after the baggage enters the baggage security detection device;
[0136] Step 3 also includes adding the baggage security detection device detection graphic information into the matching process.
[0137] Furthermore, the mobile phone communication terminal of the inspected object is not placed on the luggage conveyor belt after inspection; the central processing unit sends its matched luggage location information to the mobile phone communication terminal corresponding to the mobile phone feature code information matching the luggage feature information based on the matching association information, thereby notifying the inspected object of its luggage reclaim belt location information.
[0138] Here is a more specific example:
[0139] An intelligent security inspection system based on directional detection code, such as Figure 1 The intelligent security inspection system comprises: N security inspection entrances, N-in-N luggage security inspection conveyor belts connected to the security inspection entrances, a luggage safety detection device connected to the luggage security inspection conveyor belts, an N-in-N luggage conveyor belt connected to the back of the luggage security inspection detection device, and a luggage retrieving device arranged at the rear end of the N-in-N luggage conveyor belts, where N is a positive integer greater than 1;
[0140] The security entrance and the security exit are both provided with a directional code detection unit, a face recognition unit, a baggage pattern recognition unit and a central processing unit;
[0141] The device is used to simulate the induced signal of the mobile operator, realize the active code detection of the mobile communication terminals in a certain area, and obtain the mobile phone feature code information;
[0142] The face recognition unit is used to recognize facial feature information and determine the identity of the face;
[0143] The luggage graphic recognition unit is used to identify the luggage feature information and determine the unique features of the luggage; the central processing unit matches and associates the mobile phone feature code information, the face feature information, and the luggage feature information;
[0144] The central processing unit at the security check exit matches and correlates the mobile phone feature code information, the face feature information, and the luggage feature information. If there is a match, the luggage is released; if there is no match, the luggage is not released and a warning prompt signal is output.
[0145] This embodiment has subverted the current one-to-one mode of security checkpoints, luggage conveyors, and luggage detection devices. The association of the characteristic signals of people, ID cards, and mobile phones begins at the security checkpoint. When placing luggage, the association of luggage characteristics is added. When detecting luggage, the association of luggage detection graphics (generally perspective images) is added. This improves the security inspection efficiency of the inspected object and prevents the inspected object from taking the wrong luggage or missing luggage when leaving the security checkpoint.
[0146] The directional code detection unit in the embodiment includes a directional code detection antenna and a frame code detector, which mainly realizes the function of obtaining mobile phone signals from a specific area and parsing the mobile phone number. It simulates the mobile communication network of China Mobile, China Unicom, and China Telecom operators, sends out an induced signal, and realizes active code detection of mobile phone terminals in its control area.
[0147] Preferably, the matching association information of the mobile phone feature code information, the facial feature information, and the luggage feature information is consistent at the security check entrance and the security check exit. The preferred solution prevents the problem of missing luggage. Preferably, the luggage retrieval device includes KB luggage retrieval belts, and the central processing unit sends the matching luggage location information to the mobile phone communication terminal corresponding to the mobile phone feature code information matching the luggage feature information according to the matching association information, so as to notify the inspected person of the location information of the luggage retrieval belt, KB≥N. The preferred solution can further improve the inspected person's ability to know the luggage location and the security inspection efficiency.
[0148] Preferably, the central processing unit also assigns feature codes to the baggage feature information, and associates the feature codes with the detection patterns of the baggage security detection device. Adding the detection patterns to the association family can improve the accuracy of baggage matching.
[0149] Preferably, the central processing unit executes the following program:
[0150] Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s;
[0151] No. i The description features of all sampling points corresponding to the sample unit are:
[0152] Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points;
[0153] Step 2: Model the training set for the example unit and define the training set
[0154] Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients
[0155]
[0156] Among them, D gis the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type;
[0157] according to use Function updates parsing dictionary
[0158]
[0159] in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition;
[0160] according to Update dictionary
[0161]
[0162] Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class;
[0163] If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ];
[0164] Calculate the comprehensive dictionary D * =w1D+w2P;
[0165] Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P;
[0166] Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s};
[0167] ψ(uni i )=[Z(uni1), Z(uni2),...,Z(uni s )];
[0168] Z(unis ) is the sparse coding feature of the s-th sub-block in the graphics frame;
[0169] Z(uni s )={z p :p};
[0170]
[0171] z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ;
[0172] c j =argmin||X j -D * c j ||+λ||c j ||;
[0173] c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ;
[0174] Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
[0175] In the preferred embodiment, the image frames or image sub-segments required by the present invention can be extracted through a simple and efficient image cutting and classification method, thereby improving the efficiency of the security inspection algorithm.
[0176] This embodiment also provides an intelligent security inspection method based on directional code detection, the method comprising:
[0177] Step 1: Complete the collection and matching of the identity information, face information, and mobile phone feature code information of the inspected person at the security check entrance;
[0178] The collection of mobile phone feature information codes is completed by the directional code detection unit, the identity information is completed by collecting the ID card information submitted by the inspected object, and the face information is completed by the face recognition unit;
[0179] Step 2, tracking the inspected object. When the inspected object places the luggage on the luggage conveyor belt, the luggage visual information is matched and added based on the matching association relationship in step 1; the luggage visual information is collected by the luggage graphic recognition unit;
[0180] Step 3, track the inspected object and the luggage graphics respectively. When the inspected object arrives at the security check exit, detect whether the identity information, facial information, and mobile phone feature code information of the inspected object match. If they match, the object will be released. If they do not match, the object will not be released and a warning prompt signal will be output.
[0181] Specifically, the central processing unit performs the following steps:
[0182] Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s;
[0183] No. i The description features of all sampling points corresponding to the sample unit are:
[0184] Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points;
[0185] Step 2: Model the training set for the example unit and define the training set
[0186] Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients
[0187]
[0188] Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type;
[0189] according to use Function updates parsing dictionary
[0190]
[0191] in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition;
[0192] according to Update dictionary
[0193]
[0194] Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class;
[0195] If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ];
[0196] Calculate the comprehensive dictionary D * =w1D+w2P;
[0197] Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P;
[0198] Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s};
[0199] ψ(uni i )=[Z(uni1), Z(uni2),...,Z(uni s )];
[0200] Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame;
[0201] Z(uni s )={z p :p};
[0202]
[0203] z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ;
[0204] c j =argmin||X j -D * c j ||+λ||c j ||;
[0205] c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ;
[0206] Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
[0207] Specifically, step 2 also includes tracking the baggage graphic frame, and after the baggage enters the baggage security detection device, matching and adding the baggage security detection device detection graphic information;
[0208] Step 3 also includes adding the detection graphic information of the luggage security detection device into the matching process. Specifically, the mobile communication terminal of the inspected object is not placed on the luggage conveyor belt after being inspected; the central processing unit sends its matched luggage location information to the mobile communication terminal corresponding to the mobile phone feature code information matching the luggage feature information according to the matching association information, so as to notify the inspected object of its luggage claim belt location information.
[0209] Although the above describes the illustrative specific embodiments of the present invention so that those skilled in the art can understand the present invention, the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, all inventions and creations using the concepts of the present invention are protected.
Claims
1. An intelligent security inspection system based on directional code detection, characterized in that: It includes multiple security check entrances, multiple security check exits and a central processing unit. A baggage conveying detection system is arranged in the area between the security check entrances and the security check exits. The baggage conveying detection system includes a baggage conveyor belt and a baggage reclaiming device. The baggage reclaiming device is arranged at the rear end of the baggage conveyor belt. The baggage conveyor belt is provided with a baggage detection device. The security check entrance and the security check exit are both provided with a directional code detection unit, a face recognition unit, and a luggage pattern recognition unit; The directional code detection unit is used to simulate the induced signal of the mobile operator to realize the active code detection of the mobile communication terminals in the area and obtain the mobile phone feature code information; The face recognition unit is used to recognize facial feature information and determine the identity of the face; The luggage graphic recognition unit is used to identify luggage feature information and determine unique features of the luggage; the central processing unit associates the mobile phone feature code information, facial feature information, and luggage feature information obtained at the security check entrance to obtain associated information, and associates the mobile phone feature code information, facial feature information, and luggage feature information obtained at the security check exit to obtain associated information, and then matches the associated information at the security check exit with the associated information at the security check entrance. If there is matching associated information, the security check exit will release the luggage; otherwise, the security check exit will not release the luggage and output a warning prompt signal.
2. According to claim 1, the intelligent security inspection system based on directional code detection is characterized in that: The baggage reclaiming device includes a plurality of baggage reclaiming belts. The central processing unit identifies the reclaiming belt position of the baggage according to the baggage feature information, and sends corresponding baggage position information to the mobile phone communication terminal corresponding to the mobile phone feature code information associated with the baggage feature information according to the associated information obtained at the security check entrance, so as to notify the inspected person of the reclaiming belt position of his / her baggage.
3. According to claim 1, the intelligent security inspection system based on directional code detection is characterized in that: The central processing unit also assigns feature codes to the baggage feature information, and associates the feature codes with the detection patterns of the baggage detection device.
4. The intelligent security inspection system based on directional code detection according to claim 2 is characterized in that: The number of the security check entrances and security check exits is N, where N is an integer greater than 1. A baggage conveyor belt is provided at each security check entrance. The N baggage conveyor belts converge together and pass through the baggage detection device together. The number of the baggage claim belts is greater than N.
5. The intelligent security inspection system based on directional code detection according to claim 4 is characterized in that: The central processing unit executes the following program: Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i Represents the i-th frame image, divides the i-th frame image into s blocks, obtains s sample units, and performs sparse sampling of n sampling points on them; Indicates the s i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s; No. i The description features of all sampling points corresponding to the sample unit are: Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points; Step 2: Model the training set for the example unit and define the training set Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type; according to use Function updates parsing dictionary in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition; according to Update dictionary Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class; If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ]; Calculate the comprehensive dictionary D * =w1D+w2P; Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P; Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s }; ψ(uni i )=[Z(uni1),Z(uni2),...,Z(uni s )]; Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame; From(uni s )={z p :p}; z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ; c j =argmin||X j -D * c j ||+λ||c j ||; c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ; Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
6. An intelligent security inspection method based on directional code detection, characterized in that: The system according to any one of claims 1 to 5 is implemented, comprising the following steps: Step 1: define a video image frame set DY = {(IMG i ,y s )}, i=1,2,3...N,IMG i represents the i-th frame image, divide the i-th frame image into s blocks, obtain s sample units, and perform sparse sampling of n sampling points on them; y si ∈{-1,+1} represents the sth i The label corresponding to the example unit, +1 means the example is relevant, -1 means it is irrelevant, s i =1,2,...s; No. i The description features of all sampling points corresponding to the sample unit are: Among them, X j is the descriptor of the jth sampling point, S j is the position parameter of the jth sampling point, and n is the number of sampling points; Step 2: Model the training set for the example unit and define the training set Step 3: Generate the same dictionary D and the reconstruction dictionary P, and define the initialization dictionary update coefficient X; according to use Function, update coefficients Among them, D g is the g-th dictionary of the same type, P g is the g-th reconstruction dictionary, I is the identity matrix; λ is the predefined update weight; Y g is the training sample matrix, K is the maximum number of iterations; km g is the number of dimensions of the g-th dictionary of the same type; according to use Function updates parsing dictionary in, It represents the remaining matrix after removing the g-th class sample from the sample Y; the superscript T represents transposition; according to Update dictionary Where c is the number of iterations of the dictionary D; NOR() means normalizing the obtained matrix; T g is an intermediate variable; ρ is a predefined coefficient, S g is the predefined auxiliary matrix; X g is the coefficient matrix of the g-th class; If the objective function converges or reaches the maximum number of iterations, the loop ends and the same dictionary D = [D1, D2, ... D K ] and reconstruct the dictionary P = [P1, P2, ... P K ]; Calculate the comprehensive dictionary D * =w1D+w2P; Among them, w1 is the weight of dictionary D, w2 is the weight of dictionary P; Step 4: Calculate the sample package uni s The spatial sparse coding feature ψ(uni s ), construct Trn'={ψ(uni s ), y s }; ψ(uni i )=[Z(uni1),Z(uni2),...,Z(uni s )]; Z(uni s ) is the sparse coding feature of the s-th sub-block in the graphics frame; From(uni s )={z p :p}; z p represents the item with the largest absolute value in the pth row in C, and M represents the dictionary D * The number of code elements in; C = [c1, c2, ..., c n ]∈R M×n ; c j =argmin||X j -D * c j ||+λ||c j ||; c j ∈R M×1 Represents X j The sparse coding coefficients, λ represents the regularization parameter, ||c j || means c j The normalized norm of ; Step 5: Trn' and D * Input classifier SVM(a * ,b * ), input the inspected graphic frame, classify the sub-images of the new inspected graphic frame, if it belongs to a face image, it is input to the face recognition unit, if it belongs to a luggage graphic, it is input to the luggage graphic recognition unit, and the remaining classified graphics are saved and discarded after being saved for a predefined time.
