Multi-source traffic monitoring sensor data calibration method and system

Abstract

The invention relates to the technical field of monitoring, in particular to a multi-source traffic monitoring sensor data calibration method and system. The method comprises the following steps of: recording vehicle track data uploaded by each sensor into a file, calculating a traffic flow context feature vector of each target, matching the targets between adjacent sensors based on the feature vectors, obtaining a large amount of labeled data based on the time and coordinate information of the matched targets, and carrying out real-time labeling on the labeled data. And training in the annotation data to obtain a relative coordinate correction amount between the two sensors. According to the method, the sensor data can be rapidly and automatically processed, a large number of training samples are obtained, the calibration precision is improved, and good practicability and convenience are achieved.

Description

Technical Field

[0001] The present invention relates to the field of monitoring technology, and in particular to a multi-source traffic monitoring sensor data calibration method and system. Background Art

[0002] With the increasing mileage of highway construction in China, major traffic accidents on highways are becoming more and more serious, and operators have a growing demand for the supervision of vehicles on highways. For some key sections of roads, video, radar and other technologies are often used to detect the movement of vehicles. The detection range of any sensor is limited. Therefore, to obtain a complete driving trajectory record of the same vehicle, it is necessary to first fuse the data detected by different sensors on the road. The data structure of the general sensor output is the coordinate data of the vehicle relative to itself. In order to obtain a complete fused driving trajectory of a vehicle, the coordinate calibration technology between sensor data is crucial. Summary of the invention

[0003] In view of the above problems, the present invention proposes the following technical solutions:

[0004] A multi-source traffic monitoring sensor data calibration method is applied to a traffic monitoring system, wherein the traffic monitoring system includes a server and a plurality of sensors, the sensors are sequentially installed on the side of a highway and establish a communication connection with the server;

[0005] The method comprises the following steps:

[0006] S1: The sensor reports the vehicle monitoring data to the server. Each frame of the monitoring data includes a timestamp, a sensor ID, a vehicle ID, a vehicle horizontal coordinate, and a vehicle vertical coordinate.

[0007] The server records the monitoring data of M vehicles, M ≥ 30;

[0008] S2: Construct a feature vector for each vehicle reported by the sensor, expressed as:

[0009]

[0010] in, The feature vector of the nth car representing the ith sensor;

[0011] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-1th car of the i-th sensor;

[0012] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-2th car of the i-th sensor;

[0013] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-3th car of the i-th sensor;

[0014] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-4th car of the i-th sensor;

[0015] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-5th car of the i-th sensor;

[0016] S3: among the vehicles reported by the adjacent sensors, the k pairs of vehicles with the most similar feature vectors are matched, and the matched vehicles are not matched repeatedly, so as to obtain k matching pairs, each matching pair indicating the order of the same vehicle on two adjacent sensors; k is a preset parameter;

[0017] S4: for each matching pair, calculate the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor, and take the average value as the calibration coefficient of the i+1th sensor relative to the ith sensor;

[0018] For a matching pair, the method to calculate the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor is:

[0019] Take the last L trajectory points of the vehicle trajectory of the i-th sensor, each trajectory point is represented by (t, x, y), where t is the time, x and y are the horizontal and vertical coordinates of the vehicle at time t respectively;

[0020] Perform linear fitting on the last L trajectory points of the vehicle trajectory of the i-th sensor to obtain the linear equation system:

[0021]

[0022] in, , , , is the coefficient of the straight line group;

[0023] The corresponding calibration coefficients of the i+1th sensor relative to the ith sensor include the rotation matrix and the offset matrix ;

[0024] Take multiple trajectory points of the vehicle trajectory of the i+1th sensor, each trajectory point is represented as (t, x, y), and fit the following formula:

[0025]

[0026] Get the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor for the corresponding matching pair, that is, , Specific value of

[0027] S5: Calculate the global calibration coefficient of the i-th sensor, that is, the calibration coefficient of the i-th sensor relative to the first sensor;

[0028] Among them, the global calibration coefficients of the i-th sensor include the global rotation matrix and the global offset matrix ;

[0029] The calculation method is:

[0030] When i=2,

[0031]

[0032] When i>2,

[0033]

[0034] The global calibration coefficients of the i-th sensor are obtained, including the global rotation matrix and the global offset matrix ;

[0035] Use the global calibration coefficients to calibrate the corresponding sensor.

[0036] Preferably, the criterion for measuring the similarity of feature vectors is the Euclidean distance.

[0037] Preferably,

[0038] The specific process of matching vehicles reported by adjacent sensors is as follows:

[0039] The vehicle matching similarity set of the i-th sensor is calculated:

[0040]

[0041] in,

[0042]

[0043] in, represents the Euclidean distance between the feature vector of the nth car of the ith sensor and the feature vector of the mth car of the i+1th sensor;

[0044] Take the 5 smallest vehicle matching similarities of the i-th sensor , the corresponding vehicles are matched, and the matched vehicles are not matched repeatedly, and 5 matching pairs are obtained.

[0045] Preferably, the value of L is 20.

[0046] Preferably,

[0047] The form is or ;

[0048] in, is the rotation angle;

[0049] The form is ;in represents the lateral offset coefficient, Represents the longitudinal offset coefficient.

[0050] The present invention also provides a traffic monitoring system, which uses the above method to calibrate sensor data.

[0051] Beneficial effects: The method provided by the present invention can quickly and automatically process sensor data, obtain a large number of training samples, improve calibration accuracy, and has good practicality and convenience. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 4 is a system structure diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] like Figure 1 As shown, a multi-source traffic monitoring sensor data calibration method is applied to a traffic monitoring system, wherein the traffic monitoring system includes a server and a plurality of sensors, the sensors are sequentially installed on the side of a highway and establish a communication connection with the server;

[0055] The sensor detection direction is generally the direction of the vehicle traveling in the forward direction. The method of the present invention can also be applied to the detection of the direction of the vehicle traveling in the reverse direction, or mixed direction and omnidirectional detection sensors;

[0056] The method comprises the following steps:

[0057] S1: The sensor reports the vehicle monitoring data to the server. Each frame of the monitoring data includes a timestamp, a sensor ID, a vehicle ID, a vehicle horizontal coordinate, a vehicle vertical coordinate, and a vehicle type.

[0058] The server records the monitoring data of M vehicles, where M≥30. In this embodiment, the value of M is 30.

[0059] S2: Construct a feature vector for each vehicle reported by the sensor, expressed as:

[0060]

[0061] in, The feature vector of the nth car representing the ith sensor;

[0062] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-1th car of the i-th sensor;

[0063] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-2th car of the i-th sensor;

[0064] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-3th car of the i-th sensor;

[0065] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-4th car of the i-th sensor;

[0066] represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-5th car of the i-th sensor;

[0067] S3: among the vehicles reported by the adjacent sensors, the k pairs of vehicles with the most similar feature vectors are matched, and the matched vehicles are not matched repeatedly, so as to obtain k matching pairs, each matching pair indicating which of the same vehicles is on two adjacent sensors; in this embodiment, the standard for measuring the similarity of feature vectors is the Euclidean distance; the value of k is 5;

[0068] The specific process of matching vehicles reported by adjacent sensors in this embodiment is as follows:

[0069] The vehicle matching similarity set of the i-th sensor is calculated:

[0070]

[0071] in,

[0072]

[0073] in, represents the Euclidean distance between the feature vector of the nth car of the ith sensor and the feature vector of the mth car of the i+1th sensor;

[0074] Take the 5 smallest vehicle matching similarities of the i-th sensor , match the corresponding vehicles, and the matched vehicles will not be matched again, and 5 matching pairs are obtained;

[0075] S4: for each matching pair, calculate the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor, and take the average value as the calibration coefficient of the i+1th sensor relative to the ith sensor;

[0076] For a matching pair, the method to calculate the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor is:

[0077] Take the last L trajectory points of the vehicle trajectory of the i-th sensor, each trajectory point is represented by (t, x, y), where t is time, x and y are the horizontal coordinate and vertical coordinate of the vehicle at time t respectively; in this embodiment, the value of L is 20;

[0078] Perform linear fitting on the last L trajectory points of the vehicle trajectory of the i-th sensor to obtain the linear equation system:

[0079]

[0080] in, , , , is the coefficient of the straight line group;

[0081] The corresponding calibration coefficients of the i+1th sensor relative to the ith sensor include the rotation matrix and the offset matrix ;

[0082] in The form is or ;

[0083] in, is the rotation angle; for Linear approximation of ;

[0084] The form is ;in represents the lateral offset coefficient, represents the longitudinal offset coefficient;

[0085] Take multiple trajectory points of the vehicle trajectory of the i+1th sensor, each trajectory point is represented as (t, x, y), and fit the following formula:

[0086]

[0087] Get the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor for the corresponding matching pair, that is, , The specific value of .

[0088] Specifically, in this embodiment, The form is ;

[0089] The following set of equations is fitted using multiple trajectory points of the vehicle trajectory of the i+1th sensor:

[0090]

[0091] get , , , that is, the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor for the corresponding matching pair is obtained, that is, , The specific value of .

[0092] S5: Calculate the global calibration coefficient of the i-th sensor, that is, the calibration coefficient of the i-th sensor relative to the first sensor;

[0093] Among them, the global calibration coefficients of the i-th sensor include the global rotation matrix and the global offset matrix ;

[0094] The calculation method is:

[0095] When i=2,

[0096]

[0097] When i>2,

[0098]

[0099] The global calibration coefficients of the i-th sensor are obtained, including the global rotation matrix and the global offset matrix ;

[0100] Use the global calibration coefficients to calibrate the corresponding sensor.

[0101] The first sensor is the coordinate reference and does not require calibration, so the situation where i=1 does not exist at this time.

[0102] Finally, the process of calibrating the sensor using the global calibration coefficient belongs to the prior art and will not be described in detail here.

[0103] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A multi-source traffic monitoring sensor data calibration method is applied to a traffic monitoring system, wherein the traffic monitoring system includes a server and multiple sensors, the sensors are sequentially installed on the side of the road and establish a communication connection with the server; characterized in that: The method comprises the following steps: S1: The sensor reports the vehicle monitoring data to the server. Each frame of the monitoring data includes a timestamp, a sensor ID, a vehicle ID, a vehicle horizontal coordinate, and a vehicle vertical coordinate. The server records the monitoring data of M vehicles, M ≥ 30; S2: Construct a feature vector for each vehicle reported by the sensor, expressed as: in, The feature vector of the nth car representing the ith sensor; represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-1th car of the i-th sensor; represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-2th car of the i-th sensor; represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-3th car of the i-th sensor; represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-4th car of the i-th sensor; represents the difference between the first frame timestamp of the nth car and the first frame timestamp of the n-5th car of the i-th sensor; S3: among the vehicles reported by the adjacent sensors, the k pairs of vehicles with the most similar feature vectors are matched, and the matched vehicles are not matched repeatedly, so as to obtain k matching pairs, each matching pair indicating the order of the same vehicle on two adjacent sensors; k is a preset parameter; S4: for each matching pair, calculate the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor, and take the average value as the calibration coefficient of the i+1th sensor relative to the ith sensor; For a matching pair, the method to calculate the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor is: Take the last L trajectory points of the vehicle trajectory of the i-th sensor, each trajectory point is represented by (t, x, y), where t is the time, x and y are the horizontal and vertical coordinates of the vehicle at time t respectively; Perform linear fitting on the last L trajectory points of the vehicle trajectory of the i-th sensor to obtain the linear equation system: in, , , , is the coefficient of the straight line group; The corresponding calibration coefficients of the i+1th sensor relative to the ith sensor include the rotation matrix and the offset matrix ; Take multiple trajectory points of the vehicle trajectory of the i+1th sensor, each trajectory point is represented as (t, x, y), and fit the following formula: Get the corresponding calibration coefficient of the i+1th sensor relative to the ith sensor for the corresponding matching pair, that is, , Specific value of S5: Calculate the global calibration coefficient of the i-th sensor, that is, the calibration coefficient of the i-th sensor relative to the first sensor; Among them, the global calibration coefficients of the i-th sensor include the global rotation matrix and the global offset matrix ; The calculation method is: When i=2, When i>2, The global calibration coefficients of the i-th sensor are obtained, including the global rotation matrix and the global offset matrix ; Use the global calibration coefficients to calibrate the corresponding sensor.

2. A multi-source traffic monitoring sensor data calibration method according to claim 1, characterized in that: The criterion for measuring the similarity of feature vectors is the Euclidean distance.

3. The multi-source traffic monitoring sensor data calibration method according to claim 1, characterized in that: The specific process of matching vehicles reported by adjacent sensors is as follows: The vehicle matching similarity set of the i-th sensor is calculated: in, in, represents the Euclidean distance between the feature vector of the nth car of the ith sensor and the feature vector of the mth car of the i+1th sensor; Take the 5 smallest vehicle matching similarities of the i-th sensor , the corresponding vehicles are matched, and the matched vehicles are not matched repeatedly, and 5 matching pairs are obtained.

4. The multi-source traffic monitoring sensor data calibration method according to claim 1, characterized in that: The value of L is 20.

5. The multi-source traffic monitoring sensor data calibration method according to claim 1, characterized in that: The form is or ; in, is the rotation angle; The form is ;in represents the lateral offset coefficient, Represents the longitudinal offset coefficient.

6. A traffic monitoring system, characterized in that: Sensor data calibration is performed using the method described in any one of claims 1 to 5.

Images