A method for detecting abnormal traffic behavior based on cross-validation of multi-source data

Abstract

The present invention provides a method for detecting abnormal traffic behavior based on cross-validation of multi-source data, comprising the following steps: data discretization, collecting the image capture data and ETC transaction data of each ETC gantry, and discretizing the data; data matching, matching the discretized ETC transaction data and the image capture data; calculating the matching degree of the actual license plate; and behavior determination. The present invention provides a method for detecting abnormal traffic behavior based on cross-validation of multi-source data, based on the ETC transaction data recorded by the RSU antenna and the image capture data recorded by the camera, cross-checks the consistency between the license plate issued by the OBU and the actual license plate, has a fast detection rate and high accuracy, and can restore the actual trajectory of the vehicle when a toll dispute occurs, which can effectively avoid the loss of tolls on the highway.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent transportation technology, and in particular relates to an abnormal traffic behavior detection method based on multi-source data cross-validation. Background Art

[0002] The ETC gantry is the only toll collection method on highways. On average, a pair of ETC gantries are installed every 5 to 10 kilometers. There are two main types of data collection equipment on the gantry: one is the RSU antenna, which is responsible for reading and writing the OBU; the other is the camera, which is responsible for taking photos of vehicles and identifying license plates. The two systems run independently and have a strict time synchronization mechanism. The time error between license plate recognition and the generation of ETC transaction records is within 3 seconds.

[0003] As a mainstream toll collection mode for highways, ETC has a higher transaction proportion year by year due to its high traffic efficiency and preferential treatment. Both passenger cars and trucks can apply for it. In order to ensure the accuracy of highway toll collection, the Ministry of Transport requires one ETC vehicle and one label (OBU), and requires that the vehicle's hanging license plate and the license plate in the card are consistent. However, during the ETC processing, the license plate in the OBU card is entered incorrectly by human or negligent means; after the vehicle is traded or transferred, the OBU is not cancelled as required; the OBU is replaced by human violation: such as putting the OBU of vehicle A into vehicle B for use, or putting the OBU of a small car into a large car for use. The above situation will cause the license plate in the OBU card and the actual license plate of the vehicle to be inconsistent when the vehicle is driving on the highway, resulting in inaccurate vehicle billing, and the vehicle's true driving trajectory cannot be restored in the event of a toll dispute, resulting in the loss of highway tolls. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for detecting abnormal traffic behavior based on multi-source data cross-validation. The present invention aims to solve the problem that the transaction license plate in the vehicle OBU card is inconsistent with the actual hanging license plate, making it difficult to restore the real driving trajectory of the vehicle.

[0005] To achieve the above object, the present invention provides a method for detecting abnormal traffic behavior based on multi-source data cross-validation, comprising the following steps:

[0006] S1. Data discretization: Collect the image capture data and ETC transaction data of each ETC gantry and discretize the data;

[0007] S2. Data matching: Match the discretized ETC transaction data and the captured image data;

[0008] S3. Calculate the matching degree of the actual license plate;

[0009] S4. Behavior determination.

[0010] Furthermore, in step S1, the image capture data and ETC transaction data of each ETC gantry are discretized at a granularity of 1 minute.

[0011] Furthermore, in step S2, assuming that the image capture data set is SP and the ETC transaction data set is SE, the matching process is as follows:

[0012] S2.1 Take the intersection of data sets SP and SE, SP∩SE, and then group and aggregate the counts based on the key consisting of the image license plate and the transaction license plate to obtain the result A;

[0013] S2.2 Find the license plate that only appears in SP, that is, find the difference set SP-SE between SP and SE;

[0014] S2.3 Find the license plate that only appears in SE, that is, find the difference set SE-SP between SE and SP;

[0015] S2.4 connects SP-SE and SE-SP according to the key consisting of time period and gantry, and then performs group aggregation counting according to the key consisting of image license plate and transaction license plate to obtain result B;

[0016] S2.5 merges A and B, and then transforms to obtain result C.

[0017] Furthermore, in step S3, the matching degree is calculated as follows:

[0018] S3.1 Group and aggregate the output data C of step S2.5 using the image license plate as a key;

[0019] S3.2 calculates the similarity between the transaction license plate and the image license plate in the group in step S3.1, and then removes the interfering license plate according to the similarity threshold;

[0020] If the similarity between the transaction license plate and the image license plate > the similarity threshold, the transaction license plate matches the image license plate.

[0021] If the similarity between the transaction license plate and the image license plate is less than the similarity threshold, the transaction license plate and the image license plate do not match;

[0022] S3.3 calculates the matching degree between the transaction license plate and the image license plate through similarity, and the matching degree is the ratio of the number of matches to the total number of matches.

[0023] Furthermore, in step S3.2, the calculation expression of the similarity S is as follows:

[0024]

[0025] Where maxLen is the maximum length of the transaction license plate and the image license plate string.

[0026] Furthermore, the threshold is a preset value, which is determined according to the actual performance of the license plate recognition system.

[0027] Furthermore, the threshold is 0.76.

[0028] The beneficial effects of the present invention are:

[0029] The present invention provides an abnormal traffic behavior detection method based on multi-source data cross-validation. Based on the ETC transaction data recorded by the RSU antenna and the picture capture data recorded by the camera, the consistency between the license plate issued by the OBU and the actual license plate is cross-checked. The detection rate is fast and the accuracy is high. In addition, the real trajectory of the vehicle can be restored when a toll dispute occurs, which can effectively avoid the loss of tolls on the highway.

[0030] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a method for detecting abnormal traffic behavior based on multi-source data cross-validation according to the present invention;

[0032] Figure 2 This is a schematic diagram of the time-gantry after the data of Example 1 is discretized. DETAILED DESCRIPTION

[0033] In order to make the technical solutions, advantages and purposes of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of this application.

[0034] Example 1

[0035] like Figure 1 As shown, the present invention provides an abnormal traffic behavior detection method based on multi-source data cross-validation, which can detect OBU-issued license plates that are inconsistent with actual license plates. The specific steps are as follows:

[0036] S1. Data discretization;

[0037] S2. Data matching;

[0038] S3. Calculate the matching degree of the actual license plate;

[0039] S4. Behavior determination.

[0040] Step 1: Data discretization

[0041] The image capture data and ETC transaction data collected by each ETC gantry are discretized at a granularity of 1 minute, and the data for each minute is divided into a bucket. The "data block" that stores multiple records in the external storage is called a "bucket".

[0042] For example, the data generated during the time period of 10:00 to 10:01 is placed in one bucket, and the data generated during the time period of 10:01 to 10:02 is placed in another bucket and the data is discretized, such as Figure 2 shown.

[0043] Step 2: Data matching

[0044] Match the ETC transaction data and the image capture data, that is, match which data in the gantry capture data appears in the ETC transaction record. Assume that the image capture data set is SP and the ETC transaction data set is SE. The pseudo code of the data matching algorithm is as follows:

[0045] Input: Image capture dataset SP;

[0046] ETC transaction dataset SE.

[0047] Process 1:

[0048] Take the intersection of the discretized data sets SP and SE, that is, SP∩SE, and then group and aggregate the counts based on the keys consisting of the image license plate and the transaction license plate to obtain the result A.

[0049] The data structure of A is: list [tuple (image license plate - transaction license plate, n)], where the image license plate and the transaction license plate are the same, n is the number of matches, n∈ a positive integer, and the result A is shown in Table 1.

[0050] Table 1

[0051] Picture License Plate - Transaction License Plate Number of Matches Chuan A88845_0 - Chuan A88845_0 24 Chuan E88801_0 - Chuan E88801_0 32 Min C88812_0 - Min C88812_0 65 Chuan R888H8_0 - Chuan R888H8_0 56 Yu D88813_0 - Yu D88813_0 28 Yu C88826_0 - Yu C88826_0 33

[0052] Process 2:

[0053] Find the license plates that only appear in SP, that is, find the difference set SP-SE between SP and SE. The data structure of the resulting data set is: list [tuple (time period-gantry, image license plate)], and the results are shown in Table 2.

[0054] Table 2

[0055] Time Period - Gantry Picture License Plate 10:00~10:01 - Gantry A Gui AXXXV6_0 10:00~10:01 - Gantry A Yu AXXXU1_0 10:00~10:01 - Gantry A Chuan EXXX59_0 10:00~10:01 - Gantry B E AXXXNE_0 10:00~10:01 - Gantry B Yu AXXXA9_0 10:00~10:01 - Gantry B Yu AXXXJ3_0 10:01~10:02 - Gantry A Yu DXXX80_0 10:01~10:02 - Gantry A Yu AXXXD9_0 10:01~10:02 - Gantry A Chuan SXXXW1_0 10:01~10:02 - Gantry B Yu AXXX7K_0 10:01~10:02 - Gantry B Yu AXXX78_0 10:01~10:02 - Gantry B Yu AXXXF8_0

[0056] Process 3:

[0057] Find the license plates that only appear in SE, that is, find the difference set SE-SP between SE and SP. The data structure of the resulting data set is: list [tuple (time period-gantry, transaction license plate)]; the results are shown in Table III.

[0058] Table 3

[0059] Time Period - Gantry Transaction License Plate 10:00~10:01 - Gantry A Yu AXXXP9_0 10:00~10:01 - Gantry A Yu AXXX15_0 10:00~10:01 - Gantry A Gui DXXX09_0 10:00~10:01 - Gantry B Yu AXXX72_1 10:00~10:01 - Gantry B Xiang AXXXG3_0 10:00~10:01 - Gantry B Yu AXXX14_0 10:01~10:02 - Gantry A Chuan SXXX69_0 10:01~10:02 - Gantry A Chuan AXXX99_0 10:01~10:02 - Gantry A Yu AXXXV1_0 10:01~10:02 - Gantry B Yu AXXXA3_0 10:01~10:02 - Gantry B Yu AXXXX3_0 10:01~10:02 - Gantry B Yu AXXX18_0

[0060] Process 4:

[0061] SP-SE and SE-SP are connected according to the key consisting of time period and gantry, and then grouped and aggregated by the key consisting of image license plate and transaction license plate to obtain result B.

[0062] The data structure of B is: list [tuple (image license plate - transaction license plate, n)], where the image license plate and transaction license plate are different, and n∈ a positive integer; the results are shown in Table 4.

[0063] Table 4

[0064] Picture License Plate - Transaction License Plate Number of Matches Zhe AXXX0U_0 - Zhe CXXX5X_0 55 Yu AXXXS6_0 - E AXXX59_0 93 Zhe AXXX0U_0 - Zhe CXXX5X_0 49 Yu AXXXM3_0 - Yu DXXX99_0 34 Zhe AXXX0U_0 - Yu AXXX13_0 100 Yu AXXXM3_0 - Yu FXXX69_0 75 Yu AXXXM3_0 - Lu BXXXB6_0 66 Yu AXXXS6_0 - Yu FXXX30_0 48 Yu AXXXS6_0 - Yu HXXX21_0 39

[0065] Process 5:

[0066] Merge A and B, and then transform to get the result C. The data structure of C is: list [tuple (image license plate, tuple (transaction license plate, number of matches))], and the results are shown in Table 5.

[0067] Table 5

[0068]

[0069] Output: C, list [tuple (image license plate, tuple (transaction license plate, n))], n∈positive integer. The elements in the list are tuples consisting of image license plate, transaction license plate and matching times, forming a nested tuple object.

[0070] Step 3: Calculate the matching degree of the actual license plate

[0071] 3.1 Based on the output data C in step 2, continue to group and aggregate the data using the image license plate as the key.

[0072] 3.2 Calculate the similarity between the transaction license plate and the image license plate in the grouping in step S3.1, and then remove the interfering license plate based on the similarity threshold.

[0073] The calculation expression of similarity S is as follows:

[0074]

[0075] Where maxLen is the maximum length of the transaction license plate and the image license plate string.

[0076] Levenshtein distance, also known as Levenshtein distance, is a type of edit distance. It refers to the minimum number of edit operations required to convert two strings from one to another.

[0077] Allowed editing operations include:

[0078] (1) Replace one character with another character;

[0079] (2) Insert a character;

[0080] (3) Delete a character.

[0081] For example, the Levenshtein distance for converting the word "kitten" to "sitting" is 3:

[0082] (1) kitten→sitten(k→s);

[0083] (2) sitten→sittin(e→i);

[0084] (3) sitting→sitting (insert g).

[0085] For example: abc and abe, the edit distance is 1, the length is 3, so the similarity is 1-1 / 3=0.666.

[0086] The similarity between the image license plate and the transaction license plate is calculated and compared with the similarity threshold to remove the interfering license plate.

[0087] If the similarity between the transaction license plate and the image license plate is greater than the similarity threshold, the transaction license plate is considered to match the image license plate.

[0088] If the similarity between the transaction license plate and the image license plate is less than the similarity threshold, it is determined that the transaction license plate and the image license plate do not match.

[0089] 3.3 The matching degree between the transaction license plate and the image license plate is calculated by similarity. The matching degree is the ratio of the number of matches to the total number of matches.

[0090] The similarity threshold of this embodiment is 0.76. The threshold needs to be determined in combination with the actual performance of the license plate recognition system. The 0.76 of this embodiment is based on the fact that the current image recognition system recognizes that the error patterns are mainly concentrated in the first provincial abbreviation and the tail license plate color, the middle character 0 is recognized as 8, C is recognized as 0, etc. Usually, at most only two of the above patterns will appear at the same time, so the maximum edit distance is taken as 2, and the final calculated similarity is 0.76.

[0091] The matching degree is calculated as follows:

[0092] Image license plate grouping => group 1, group 2, group 3, ..., group n; n∈positive integer

[0093] for group in all groups:

[0094] for license plate in group

[0095] if calculate similarity(image license plate, transaction license plate) < 0.76

[0096] Generate a list [list[tuple(transaction license plate, transaction license plate matching times, transaction license plate matching degree)]], and the results are shown in Table VI.

[0097] Table 6

[0098]

[0099] Take the first element in each sublist, which is the tuple consisting of the transaction license plate with the highest matching degree, the number of transaction license plate matches, and the transaction license plate matching degree. The results are shown in Table VII.

[0100] Table 7

[0101] Picture License Plate Transaction License Plate Number of Matches Match Degree Yu AXXXM3_0 Yu FXXX69_0 153 0.605 Yu AXXXS6_0 E AXXX59_0 93 0.517 Zhe AXXX0U_0 Yu AXXX13_0 100 0.403

[0102] Step 4: Behavior determination

[0103] The actual example calculated by the algorithm of the present invention is shown in Table 8.

[0104] Table 8

[0105]

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormal traffic behavior based on multi-source data cross-validation, characterized in that: The following steps are involved: S1. Data discretization: Collect the image capture data and ETC transaction data of each ETC gantry and discretize the data; S2. Data matching: Match the discretized ETC transaction data and the captured image data; In step S2, the image capture data set is denoted as SP, and the ETC transaction data set is denoted as SE. The matching process is as follows: S2.1 Take the intersection of data sets SP and SE, SP∩SE, and then group and aggregate the counts based on the key consisting of the image license plate and the transaction license plate to obtain the result A; S2.2 Find the license plate that only appears in SP, that is, find the difference set SP-SE between SP and SE; S2.3 Find the license plate that only appears in SE, that is, find the difference set SE-SP between SE and SP; S2.4 connects SP-SE and SE-SP according to the key consisting of time period and gantry, and then performs group aggregation counting according to the key consisting of image license plate and transaction license plate to obtain result B; S2.5 combines A and B, and then transforms to obtain result C; S3. Calculate the matching degree of the actual license plate; In step S3, the calculation process of the matching degree is as follows: S3.1 Group and aggregate the output data C of step S2.5 using the image license plate as a key; S3.2 calculates the similarity between the transaction license plate and the image license plate in the group in step S3.1, and then removes the interfering license plate according to the similarity threshold; In step S3.2, the calculation expression of the similarity S is as follows: In the formula, maxLen is the maximum length of the transaction license plate and the image license plate string; If the similarity between the transaction license plate and the image license plate > the similarity threshold, the transaction license plate matches the image license plate. If the similarity between the transaction license plate and the image license plate is less than the similarity threshold, the transaction license plate and the image license plate do not match; S3.3 calculates the matching degree between the transaction license plate and the image license plate by similarity, where the matching degree is the ratio of the number of matches to the total number of matches; S4. Behavior determination.

2. The abnormal traffic behavior detection method based on multi-source data cross-validation according to claim 1 is characterized by: In step S1, the image capture data and ETC transaction data of each ETC gantry are discretized at a granularity of 1 minute.

3. The abnormal traffic behavior detection method based on multi-source data cross-validation according to claim 1 is characterized in that: The threshold is a preset value, and the threshold is determined according to the actual performance of the license plate recognition system.

4. The abnormal traffic behavior detection method based on multi-source data cross-validation according to claim 3 is characterized by: The threshold is 0.76.

Images