Vehicle identification method and system based on fixed camera

Abstract

The invention discloses a vehicle recognition method and system based on fixed cameras, and the method comprises the steps: collecting a vehicle image through a plurality of fixed cameras, obtaining the attributes of a vehicle through image recognition, and obtaining the driving record details of the vehicle; constructing a normal driving time matrix between the cameras, wherein the normal driving time matrix is used for reflecting the normal driving time of the vehicle between any two cameras; based on the normal driving time matrix between the cameras, vehicle number identification of the vehicle is carried out; and based on the train number identification result, counting the attributes of the vehicles of the same train number, and taking the identification result of the attribute with the highest proportion as the final vehicle identification result. According to the method and the system, the accuracy of vehicle identification can be remarkably improved, and effective support is provided for intelligent monitoring and management of the vehicle.

Description

Technical Field

[0001] The present invention relates to the field of vehicle identification, and particularly to a vehicle identification method and system based on a fixed camera. Background Art

[0002] With the continuous development of the Intelligent Transportation System (ITS), road cameras have become one of the important tools for obtaining traffic information. Fixed cameras can capture dynamic information of vehicles, including license plate numbers, vehicle body features, driving trajectories, etc. However, the original data captured by current cameras lacks a unified analysis and identification mechanism, making it difficult to accurately count vehicle attributes and operating status. Especially in scenarios such as logistics monitoring, vehicle classification, and scheduling, the existing technologies have the following deficiencies:

[0003] 1. Isolated analysis of vehicle information: The data between different cameras lacks relevance and cannot form a unified identification result.

[0004] 2. Low accuracy of attribute identification: Existing methods are difficult to stably identify vehicle attributes. Especially in complex environments or with poor image quality, the reliability of the identification results is insufficient.

[0005] 3. Lack of global analysis ability: It is impossible to perform statistical analysis based on the capture information of multiple cameras, and it is difficult to obtain credible vehicle attributes from multiple dimensions. Summary of the Invention

[0006] The present invention provides a vehicle identification method and system based on a fixed camera, aiming to solve the problem of insufficient vehicle information analysis in the existing technology. Through the method and system of the present invention, the accuracy of vehicle identification can be significantly improved, providing effective support for vehicle intelligent monitoring and management.

[0007] The technical solution for achieving the purpose of the present invention is: A vehicle identification method based on a fixed camera, which improves the accuracy of vehicle identification and the robustness of the system by constructing a driving time matrix between cameras and an artificial intelligence model, and specifically includes the steps:

[0008] Collect vehicle images through multiple fixed cameras, obtain vehicle attributes through image recognition, and obtain vehicle driving record details;

[0009] Construct a normal driving time matrix between cameras to reflect the time for a vehicle to drive normally between any two cameras;

[0010] Based on the normal driving time matrix between cameras, perform vehicle trip identification;

[0011] Based on the trip identification result, count the vehicle attributes of the same trip, and the attribute identification result with the highest proportion is the final vehicle identification result.

[0012] Further, the driving record details include the license plate number, camera ID, and camera capture time.

[0013] Further, the vehicle attributes include vehicle type, vehicle size, body color, whether it is loaded with goods, and the category of the loaded goods.

[0014] Further, the normal driving time matrix between cameras is as follows:

[0015]

[0016] where a ij is the normal driving time from the i-th camera to the j-th camera, 1 ≤ i, j ≤ n, and n is the number of cameras.

[0017] Further, the normal driving time a ij from the i-th camera to the j-th camera is calculated using the average value method or the median method.

[0018] Further, the driving time of all vehicles from the i-th camera to the j-th camera within a set time range is obtained from the vehicle driving record details. Using the average value method, the average of all driving times is taken as the normal driving time. Using the median method, the median of all driving times is taken as the normal driving time.

[0019] Further, the vehicle trip identification specifically includes: If the same vehicle drives continuously without stopping, it is regarded as the same trip. If there is a stop between two adjacent cameras, before and after the stop are regarded as different trips.

[0020] Further, the method for judging a stop is as follows:

[0021] According to the normal driving time matrix between cameras, the normal driving time range is determined as [a ij (1 - p), a ij (1 + p)]. If the driving time of a vehicle is within the normal driving time range, it is judged that the vehicle has no stop. If the actual driving time is greater than the upper limit of the normal driving time range, it is judged that the vehicle has a stop on the way, where p is a set coefficient, and a ij is the normal driving time from the i-th camera to the j-th camera.

[0022] Further, the range of the set coefficient p is: 10% ≤ p ≤ 30%.

[0023] A vehicle identification system based on fixed cameras includes:

[0024] A camera capture and collection module that captures vehicle images through multiple fixed cameras and obtains vehicle driving record details;

[0025] A vehicle recognition module obtains vehicle attributes through image recognition, constructs a normal driving time matrix between cameras, which is used to reflect the normal driving time of a vehicle between any two cameras. Based on the normal driving time matrix between cameras, vehicle trip recognition is performed. Based on the trip recognition result, the vehicle attributes of the same trip are statistically analyzed, and the attribute recognition result with the highest proportion is the final vehicle recognition result.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] (1) By constructing a driving matrix, the normal driving time range of a vehicle during the driving process can be obtained, which can be used for abnormal driving detection, congestion analysis, vehicle path tracking, etc.;

[0028] (2) Trip recognition: According to whether there is a stop during driving, a single driving is divided into multiple trips; according to the principle that "the vehicle attributes remain unchanged in the same trip", the vehicle attributes in the same trip can be statistically analyzed; the changes in vehicle attributes in different trips can be analyzed to accurately identify vehicle behavior;

[0029] (3) By combining image recognition, driving matrix, trip recognition and statistical analysis technologies, the accuracy of vehicle recognition results is improved: The model recognition results of each picture are obtained through image recognition, the trips are identified by forming a matrix and trip recognition algorithm, and the model recognition results are statistically analyzed based on the trips, and finally the final vehicle recognition result is obtained. Compared with the model recognition result, the final result has higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a vehicle recognition flow chart.

[0031] Figure 2 It is a schematic structural diagram of a vehicle recognition system. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] The present invention provides a vehicle recognition method and system based on fixed cameras, aiming to achieve the following functions:

[0034] (1) Capture vehicle information and store it as a detailed historical driving record;

[0035] (2) Construct a driving time matrix between cameras to accurately estimate the vehicle running trajectory;

[0036] (3) Use an artificial intelligence model to identify vehicle attributes and perform statistical analysis by vehicle trips, thereby generating reliable vehicle attribute results.

[0037] Combined with Figure 1 , a vehicle recognition method based on fixed cameras provided in this embodiment includes the following processes:

[0038] Step 1, the camera captures the vehicle

[0039] Based on fixed road cameras, capture pictures of the driving vehicles, and at the same time collect information such as the capture time, camera ID, license plate number, etc. The collected data is uploaded to the server in a standardized format for subsequent analysis and processing.

[0040] , Step 2, vehicle driving record details

[0041] Store the capture data of each vehicle (including time, picture, camera position) in the database to form a historical driving record details, including license plate number, camera ID, camera capture time, etc.

[0042] Step 3, the deep learning model identifies the captured picture to obtain the model recognition result of the vehicle.

[0043] Through the deep learning model, analyze the vehicle capture pictures to identify the key attributes of the vehicle, including: vehicle type (such as small car, large truck, etc.), vehicle size, body color, whether loaded with goods, category of goods loaded, etc., to obtain the model recognition result of the vehicle.

[0044] Step 4, driving time matrix between cameras

[0045] Construct a driving time matrix between cameras (hereinafter referred to as "driving matrix"), and the driving matrix can reflect the normal driving time of the vehicle between any two cameras. The normal driving time refers to the time when the vehicle does not stop due to external or internal interference during driving. Based on the driving time matrix, it can be judged whether the vehicle stops between two cameras. For example, a certain muck truck stops to dump muck between two cameras during driving. By comparing the actual driving time of the muck truck between these two cameras with the normal driving time, it can be judged whether the muck truck stops between the cameras.

[0046] Suppose there are n cameras, then the driving matrix is an n-order square matrix composed of n×n numbers a ij as follows:

[0047]

[0048] Among them, a ij is the normal driving time between the i-th camera and the j-th camera. The calculation method can adopt the average value method, the median method, etc. For example: obtain the driving time of all vehicles between the i-th camera and the j-th camera within a certain time range (such as the recent 180 days) from the vehicle driving record details. If the average value method is adopted, take the average value of all driving times as the normal driving time; if the median method is adopted, take the median of all driving times as the normal driving time.

[0049] In actual situations, due to factors such as road conditions and speed differences, the driving time of a vehicle between two cameras may fluctuate. To accommodate these fluctuations, the normal driving time range is set as [a ij (1 - p), a ij (1 + p)], where 0 ≤ p ≤ 100%, generally 10% ≤ p ≤ 30%, for example: p = 20%. If the driving time of a certain vehicle is within the normal driving time range, it can be determined that the vehicle has no stop; if the actual driving time is greater than the upper limit of the normal driving time range, the vehicle has stopped on the way, for example: stopping to dump construction waste midway, traffic congestion, etc.

[0050] Step 5, vehicle trip identification

[0051] If the same vehicle (identified by license plate number) drives continuously without stopping, it is regarded as the same trip. Assume that a vehicle does not stop between the first camera it passes through from the starting point to the m-th camera during a driving process, and stops between the m-th camera and the m + 1-th camera. Then the driving process of the vehicle from the first camera to the m-th camera is recorded as 1 trip. The next trip will start from the m + 1-th camera and end at the camera passed through before the next stop.

[0052] Generally, the attributes of a vehicle remain unchanged within the same trip, such as "whether loaded with goods", "category of goods loaded", etc.

[0053] Step 6, count by trip to obtain the final vehicle identification result

[0054] In step 3, a deep model is used to identify a single image to obtain the model identification result of the vehicle. However, the identification result obtained based on a single image may have large errors due to factors such as camera clarity, angle, and weather, resulting in different identification results for different captured images of the same vehicle.

[0055] Therefore, in the present invention, by using the principle that "in the same train trip, the vehicle attributes remain unchanged", we conduct statistical analysis on the model recognition results in the same train trip, and can obtain more accurate and reliable recognition results. When there are multiple image recognition results for the same attribute of the same vehicle, the recognition result with the highest proportion among them is the final recognition result of this attribute; the final recognition results of each attribute are counted to obtain the final recognition result of the vehicle.

[0056] For example: Vehicle A passes through 3 cameras in the same train trip and 3 captured pictures are obtained. Among them, the model recognition results of 2 pictures for whether Vehicle A "is loaded with goods" are "yes", and the model recognition result of 1 picture for whether Vehicle A "is loaded with goods" is "no". Then, according to the statistical results, it is more credible to consider that this vehicle is "loaded with goods". Therefore, compared with the recognition results of a single camera, the train trip-based statistical recognition can improve the accuracy and reliability of vehicle recognition results.

[0057] Combined with Figure 2 , this embodiment further provides a vehicle recognition system based on fixed cameras, including:

[0058] 1. Camera capture and acquisition module, including:

[0059] · Camera information management: Manage camera information, including camera name, location, type, camera angle, etc.

[0060] · Camera capture settings: Support setting the capture time, number of captured pictures, etc. of the camera.

[0061] · Vehicle captured pictures: Capture pictures of passing vehicles according to the camera capture settings.

[0062] · Capture data acquisition: First, collect and store capture data, including license plate number, capture time, camera, etc. data, to form historical driving record data; second, store the captured pictures and transmit them to the vehicle recognition module.

[0063] 2. Vehicle recognition module, including:

[0064] · Vehicle recognition: The deep learning model performs vehicle attribute recognition on the received vehicle captured pictures, such as: vehicle type (such as small car, large truck, etc.), vehicle size, body color, whether it is loaded with goods, the category of goods loaded, etc.

[0065] ·Train number identification: First, construct a train operation time matrix based on the vehicle operation record details, and set an update period (such as daily, weekly, etc.) to update the data in the train operation matrix. Second, calculate the train number: Construct a train number identification algorithm and calculate the train number according to the principle that "continuous normal driving without stopping of the same vehicle (according to the license plate number) is regarded as the same train number".

[0066] ·Vehicle identification statistics by train number: Statistically analyze the attribute results of vehicles by train number and output the final vehicle identification results. When there are multiple image recognition results for the same attribute of the same vehicle, the recognition result with the highest proportion is the final recognition result for that attribute. Statistically analyze the final recognition results of each attribute as the final recognition result of the vehicle.

[0067] ·Data service: Provide a data service interface for third parties to call the vehicle identification results.

[0068] The present invention uses fixed road cameras to capture vehicles, combines key technologies such as image recognition, train operation matrix, and train number identification to accurately identify vehicle attribute information. Through the method and system of the present invention, the accuracy of vehicle identification can be significantly improved, providing an efficient and reliable solution for traffic management, logistics monitoring, and smart city construction.

[0069] The description of the present invention is given for purposes of illustration and description, and is not intended to be exhaustive or to limit the invention to the disclosed form. Many modifications and variations are obvious to those of ordinary skill in the art. The embodiments are chosen and described in order to best explain the principles of the invention and its practical application, and to enable those of ordinary skill in the art to understand the invention and design various embodiments with various modifications suitable for specific purposes.

Claims

1. A vehicle recognition method based on a fixed camera, characterized in that: Includes steps: Collect vehicle images through multiple fixed cameras, obtain vehicle attributes through image recognition, and obtain vehicle driving record details; Construct a normal driving time matrix between cameras to reflect the normal driving time of a vehicle between any two cameras; Based on the normal driving time matrix between cameras, the vehicle number is identified; Based on the train number recognition results, the vehicle attributes of the same train number are counted, and the attribute recognition result with the highest proportion is the final vehicle recognition result.

2. A vehicle recognition method based on a fixed camera according to claim 1, characterized in that: The driving record details include the license plate number, camera ID and camera capture time.

3. The vehicle recognition method based on a fixed camera according to claim 1, characterized in that: Vehicle attributes include vehicle type, vehicle size, body color, whether it is loaded with cargo, and the type of cargo loaded.

4. The vehicle recognition method based on a fixed camera according to claim 1, characterized in that: The normal driving time matrix between the cameras is: Among them, a ij is the normal driving time between the i-th camera and the j-th camera, 1≤i,j≤n, n is the number of cameras.

5. A vehicle recognition method based on a fixed camera according to claim 4, characterized in that: The normal travel time a between the i-th camera and the j-th camera ij The mean and median methods were used for calculation.

6. A vehicle recognition method based on a fixed camera according to claim 5, characterized in that: The driving time from the i-th camera to the j-th camera of all vehicles within the set time range is obtained from the vehicle driving record details. The average method is used to take the average of all driving times as the normal driving time. The median method is used to take the median of all driving times as the normal driving time.

7. A vehicle recognition method based on a fixed camera according to claim 6, characterized in that: The vehicle number identification specifically includes: if the same vehicle continues to travel normally without stopping, it is regarded as the same vehicle number; if a stop occurs between two adjacent cameras, the vehicles before and after the stop are regarded as different vehicles numbers.

8. The vehicle recognition method based on a fixed camera according to claim 7, characterized in that: The judgment method for staying is: According to the normal driving time matrix between cameras, the normal driving time range is determined as [a ij (1-p),a ij (1+p)], if the driving time of a vehicle is within the normal driving time range, it is judged that the vehicle has no stop; If the actual driving time is greater than the upper limit of the normal driving time range, the vehicle stops on the way, where p is the set coefficient and a ij is the normal travel time between the i-th camera and the j-th camera.

9. The vehicle recognition method based on a fixed camera according to claim 8, characterized in that: The range of setting coefficient p is: 10%≤p≤30%.

10. A vehicle identification system for implementing the method according to any one of claims 1 to 9, characterized in that: include: The camera capture and acquisition module collects vehicle images through multiple fixed cameras and obtains vehicle driving record details; The vehicle identification module obtains vehicle attributes through image recognition and constructs a normal driving time matrix between cameras to reflect the normal driving time of the vehicle between any two cameras. Based on the normal driving time matrix between cameras, the vehicle number is identified. Based on the identification results, the attributes of the same vehicle number are counted, and the attribute identification result with the highest proportion is the final vehicle identification result.

Images