Vehicle tracking method and system combining edge computing and attention mechanism

Abstract

The application provides a vehicle tracking method and system combining edge computing and attention mechanisms, relates to the technical field of intelligent vehicle tracking, and is applied to an edge computing architecture and comprises the following steps: respectively receiving driving image information and geographical position information suitable for indicating that an intelligent vehicle is driving on a road, generating driving feature information according to the driving image information and a preset attention mechanism, converting the driving feature information from an image coordinate system to a map coordinate system in which a preset electronic map is located, so as to obtain a tentative driving point set one, converting the geographical position information from a geographical coordinate system to the map coordinate system, so as to obtain a tentative driving point set two, and generating a continuous driving track on the preset electronic map according to the tentative driving point set one and the tentative driving point set two. The method has the compatibility of the two kinds of information, is convenient for generating a driving track on an electronic map, and is helpful to improving the continuity and accuracy of the driving track.

Description

Technical Field

[0001] This invention relates to the field of intelligent vehicle tracking technology, and more specifically, to a vehicle tracking method and system that combines edge computing and attention mechanisms. Background Technology

[0002] Edge computing architecture is deployed at the network edge, close to the data source, to provide edge intelligence services such as converged networking, computing, storage and applications. Compared with cloud data centers, edge computing architecture has advantages such as high real-time performance, low deployment cost and high security.

[0003] Attention mechanisms are typically incorporated into machine learning models in a modular form. They can automatically learn and calculate the contribution of input data to output data and are widely used in various types of machine learning tasks, such as natural language processing, image recognition, and speech recognition. By introducing attention mechanisms, the focus can be placed on information that is more critical to the current task from a large amount of input information, reducing attention to other information and even filtering out irrelevant information. This can solve the problem of information overload and improve the efficiency and accuracy of task processing.

[0004] Intelligent vehicle tracking methods typically monitor vehicle movement information during driving to generate driving trajectories, exhibiting a trend towards intelligent, in-depth, and diversified technological development.

[0005] Some intelligent vehicle tracking methods use edge computing architecture and / or attention mechanisms to prevent trajectory loss and / or improve tracking accuracy. However, when combining edge computing architecture and attention mechanisms, the use of geographic location information is limited by image information, resulting in poor information compatibility and hindering the optimization of driving trajectories. Summary of the Invention

[0006] The present invention aims to at least partially solve the technical problems in related technologies and provide a vehicle tracking method and system that combines edge computing and attention mechanisms.

[0007] The first aspect of this invention provides a vehicle tracking system method combining edge computing and attention mechanisms, applied in an edge computing architecture, comprising the following steps:

[0008] The system receives driving image information and geographic location information suitable for instructing the intelligent vehicle to travel on the road, respectively.

[0009] Driving feature information is generated based on the driving image information and the preset attention mechanism;

[0010] The driving feature information is transformed from the image coordinate system to the map coordinate system of the preset electronic map to obtain a provisional driving point set one;

[0011] The geographic location information is transformed from the geographic coordinate system to the map coordinate system to obtain the second set of provisional driving points;

[0012] A continuous driving trajectory is generated on the preset electronic map based on the first set of provisional driving points and the second set of provisional driving points.

[0013] Optionally, generating driving feature information based on the driving image information and the preset attention mechanism includes:

[0014] The vehicle image information is preprocessed to obtain multiple inspected vehicle images of the same size;

[0015] Using a pre-trained vehicle detection model, feature detection processing based on a dual-dimensional attention mechanism of channel and space is performed on each of the inspected driving images. Feature fusion processing is then performed on the channel attention feature map and its corresponding spatial attention feature map generated after detection.

[0016] The multiple driving feature points generated through fusion are combined into the driving feature information.

[0017] Optionally, the m-th provisional point in the provisional driving point set is represented as: ,in, This represents the homogeneous transformation matrix from the image coordinate system b to the map coordinate system a. This represents the m-th driving feature point in the image coordinate system b;

[0018] The nth provisional point in the second set of provisional driving points is represented as: ,in, This represents the homogeneous transformation matrix from the geographic coordinate system c to the map coordinate system a. This indicates the nth geographic coordinate point in the geographic coordinate system c.

[0019] Optionally, generating a continuous driving trajectory on the preset electronic map based on the first set of provisional driving points and the second set of provisional driving points includes:

[0020] Based on at least one of the provisional driving point set one and the provisional driving point set two, prefabricated lane lines suitable for indicating the road are determined on the preset electronic map;

[0021] By using a preset linear interpolation model, the tentative driving point set 1 is expanded into an expanded set 1, and the tentative driving point set 1 is expanded into an expanded set 2.

[0022] Point pairing processing is performed on the first expansion set and the second expansion set to obtain a point pair set. The difference between the first expansion set and the point pair set is the point difference set.

[0023] The set of point pairs is calculated using a preset point pair similarity measurement model;

[0024] The point pair set is merged based on the similarity set.

[0025] Based on the prefabricated lane lines, the point difference set and the merged point set are subjected to trajectory smoothing processing to form the driving trajectory.

[0026] Optionally, the point pairing process for the first expansion set and the second expansion set includes:

[0027] Create two new sets, one and the other, both initially empty sets.

[0028] In the expanded set one, determine the kth inspected point one;

[0029] Where k represents a positive integer accumulated from 1 to K, and K represents the length of the extended set;

[0030] Detect whether there exists a second test point in the second expansion set that is suitable for pairing with the kth test point;

[0031] If so, the kth inspected point one and the inspected point two suitable for pairing with it are stored in the newly created set one in point-to-point form;

[0032] If not, the kth inspected point one is stored in the newly created set two as a single point;

[0033] Determine if k is less than K;

[0034] If so, then update the kth inspected point to the next inspected point;

[0035] If not, then the newly created set one is determined as the point-to-point set, and the newly created set two is determined as the point-difference set.

[0036] Optionally, detecting whether there exists a second test point suitable for pairing with the kth test point in the second expansion set includes:

[0037] The detection range is defined with the kth inspected point as the center and a preset distance threshold as the radius.

[0038] When any one of the two test points in the second expansion set is within the detection range, it is determined that the two test points within the detection range are suitable for pairing with the kth test point.

[0039] When each of the two test points in the second expansion set is outside the detection range, there is no test point two suitable for pairing with the kth test point one.

[0040] Optionally, the preset distance similarity calculation model is expressed as:

[0041] ;

[0042] in, Represents Euclidean distance similarity. and Let a and b represent the x-coordinates of the two points in the point pair in the map coordinate system a. and These represent the ordinates of the two points in the point pair in the map coordinate system a.

[0043] Optionally, performing point pair merging processing on the point pair set based on the similarity set includes:

[0044] Determine the i-th Euclidean distance similarity in the similarity set, and determine the i-th pair of points in the pair of points set;

[0045] Where i represents a positive integer accumulated from 1 to I, and I represents the length of either the similarity set or the point pair set;

[0046] A linear weighted merging process is performed on the i-th Euclidean distance similarity and the i-th group of point pairs to obtain the i-th merged point;

[0047] Determine if i is less than I;

[0048] If so, then update the i-th Euclidean distance similarity to the next Euclidean distance similarity, and update the i-th pair of points to the next pair of points;

[0049] If not, then combine I of the merge points to obtain the set of merge points.

[0050] Optionally, the i-th merging point is represented as: ,in, and The i-th pair of points represents the first and second points under inspection in the map coordinate system a, respectively. Let represent the Euclidean distance similarity of the i-th term.

[0051] A second aspect of the present invention provides a vehicle tracking system combining edge computing and attention mechanisms, comprising:

[0052] A first edge computing device is configured to receive driving image information suitable for instructing the intelligent vehicle to drive on the road, generate driving feature information based on the driving image information and a preset attention mechanism, and send the driving feature information.

[0053] The second edge computing device is used to receive the driving feature information and the geographical location information suitable for instructing the intelligent vehicle to drive on the road, and to convert the driving feature information from the image coordinate system to the map coordinate system of the preset electronic map to obtain a provisional driving point set one, and to convert the geographical location information from the geographic coordinate system to the map coordinate system to obtain a provisional driving point set two, and to generate a continuous driving trajectory on the preset electronic map based on the provisional driving point set one and the provisional driving point set two.

[0054] The beneficial effects of the vehicle tracking method and system combining edge computing and attention mechanism described above are as follows: When the attention mechanism is applied to the edge computing architecture, driving image information and geographic location information are used as two types of source information. Considering that driving feature information and geographic location information are in different coordinate systems, after coordinate system transformation, two sets of driving points in the same coordinate system are formed, which has the ability to be compatible with both types of information. This facilitates the generation of driving trajectories on electronic maps and helps to improve the continuity and accuracy of driving trajectories. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the architecture of a vehicle tracking system combining edge computing and attention mechanism according to an embodiment of the present invention;

[0056] Figure 2 This is a flowchart illustrating a vehicle tracking method combining edge computing and attention mechanisms according to an embodiment of the present invention.

[0057] Figure 3 For corresponding Figure 2 A flowchart illustrating the process of S6 in the diagram;

[0058] Figure 4 For corresponding Figure 3 A flowchart of S63 in the document;

[0059] Figure 5 For corresponding Figure 3 The flowchart of S65 in the document. Detailed Implementation

[0060] Embodiments of the present invention will now be described in detail with reference to the accompanying drawings. When referring to the drawings, unless otherwise indicated, the same reference numerals in different drawings denote the same or similar elements. It should be noted that the embodiments described in the following exemplary embodiments do not represent all embodiments of the present invention. They are merely examples of apparatuses and methods consistent with some aspects of the present invention disclosed in the claims, and the scope of the present invention is not limited thereto. Features in the various embodiments of the present invention can be combined with each other without contradiction.

[0061] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "numerous" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0062] See Figure 1 An embodiment of the present invention provides a vehicle tracking system combining edge computing and attention mechanisms, comprising an edge computing architecture, wherein multiple edge computing devices 110 can communicate wirelessly through a core mesh or through at least one base station 120 to share the information processing load. At least one edge computing device 110 may be an edge computing gateway or an edge server, and at least one base station 120 may be a 5G base station. Any edge computing device 110 can communicate wirelessly with a cloud data center 130 through a core network or with the vehicle monitoring system of an intelligent vehicle 140 through at least one base station 120.

[0063] In some embodiments, the vehicle monitoring system may include, but is not limited to, an RTK device 141, a smartphone 142, and a dashcam 143. The RTK device 141 may be installed on the roof of the smart vehicle 140, the smartphone 142 may be installed inside the smart vehicle 140, and the dashcam 143 may be installed at the front or rear of the smart vehicle 140 to provide a camera for the front of the vehicle and the road in front of it, or to provide a camera for the rear of the vehicle and the road behind it. No limitation is made here.

[0064] In some embodiments, either the RTK device 141 or the smartphone 142 combines geolocation information measurement and camera functions. While the intelligent vehicle is traveling along a road, it can capture video or take photos at a certain frequency as driving image information, and / or locate geographic coordinates at a certain frequency as geolocation information. The smartphone 142 can send the driving image information and the geolocation information respectively. It should be understood that the intelligent vehicle 140 typically has wireless communication capabilities and can also send information; this can be achieved by referring to existing technologies, and will not be elaborated upon here.

[0065] In some embodiments, at least one edge computing device 110 may be an edge computing gateway or an edge server, capable of receiving and processing driving image information of the intelligent vehicle, and / or receiving and processing the geographical location information of the intelligent vehicle on the road. Exemplarily, at least one edge computing device 110 includes a first edge computing device 1101 and a second edge computing device 1102 (both shown as "110" in the figure).

[0066] The first edge computing device 1101 is used to receive driving image information suitable for instructing the intelligent vehicle 140 to drive on the road, generate driving feature information based on the driving image information and a preset attention mechanism, and send the driving feature information.

[0067] In this embodiment of the invention, the first edge computing device 1101 can send driving image information to the core network for the cloud data center 130 to receive and store, and can also send driving feature information to the core network and / or at least one base station 120 for the second edge computing device 1102 to receive and store.

[0068] The second edge computing device 1102 is used to receive driving feature information and geographic location information suitable for instructing the intelligent vehicle 140 to drive on the road, and to convert the driving feature information from the image coordinate system to the map coordinate system of the preset electronic map to obtain a provisional driving point set one, and to convert the geographic location information from the geographic coordinate system to the map coordinate system to obtain a provisional driving point set two, and to generate a continuous driving trajectory on the preset electronic map based on the provisional driving point set one and the provisional driving point set two.

[0069] In this embodiment of the invention, the second edge computing device 1102 can send geographical location information and an electronic map with driving trajectory to the core network for the cloud data center 130 to receive and store. It can also send an electronic map with driving trajectory to at least one base station 120 for the smartphone 142 to receive and display. No limitation is made here.

[0070] See Figure 2 Another embodiment of the present invention provides a vehicle tracking method combining edge computing and an attention mechanism, applied in an edge computing architecture. Specifically,

[0071] S1 receives driving image information suitable for instructing intelligent vehicles to drive on the road.

[0072] S2 generates driving feature information based on driving image information and a preset attention mechanism.

[0073] In this embodiment of the invention, when the driving image information is video, it is necessary to extract frames from the video. Each frame has an area suitable for representing the overlap between the intelligent vehicle 140 and the road. The frame extraction frequency can be fixed or set according to the frame rate of the video. For example, the frame rate of the video is usually 30 frames per second, and the frame extraction frequency can be set to 10, 15, 20, 30 frames per second or other values.

[0074] In some embodiments, the preset attention mechanism can be a single-dimensional attention mechanism or a two-dimensional attention mechanism, such as a channel attention mechanism or / and a spatial attention mechanism in a neural network model. The neural network model can be a YOLO model or an R-CNN model.

[0075] S3, transform the driving feature information from the image coordinate system to the map coordinate system of the preset electronic map to obtain a provisional driving point set one.

[0076] S4 receives geographic location information suitable for instructing intelligent vehicles to travel on the road.

[0077] S5 converts the geographic location information from the geographic coordinate system to the map coordinate system to obtain the second set of provisional driving points.

[0078] S6. Generate a continuous driving trajectory on a preset electronic map based on the provisional driving point set one and provisional driving point set two.

[0079] In some embodiments, driving feature information is generated based on driving image information and a preset attention mechanism, specifically including: preprocessing driving image information to obtain multiple driving images of the same size; performing feature detection processing based on a channel and spatial dual-dimensional attention mechanism on each driving image through a pre-trained vehicle detection model; performing feature fusion processing on the channel attention feature map generated after detection and its corresponding spatial attention feature map; and combining the multiple driving feature points generated by fusion into driving feature information.

[0080] In some embodiments, the cloud data center 130 may pre-train and store a machine learning model using a sample set to make it a vehicle detection model, and then distribute it to multiple edge computing devices 110. Alternatively, one edge computing device 110 may pre-train and store the vehicle detection model and then distribute it to other edge computing devices 110. No limitation is imposed here.

[0081] The number of vehicle images to be inspected can be 15, 30, or other values. Taking 5 vehicle images to be inspected as an example, any vehicle image to be inspected can have a line of overlap where the front of the vehicle coincides with the road. The first pixel located in the middle of the line of overlap can be regarded as a vehicle feature point. Alternatively, the second pixel that differs from the first pixel by at least one pixel can be regarded as a vehicle feature point. As long as it can reflect the phase position relationship between the vehicle and the road, there is no restriction.

[0082] It should be noted that the specific methods of vehicle detection model and subsequent feature fusion processing can be adapted to existing technologies related to vehicle detection methods and devices based on attention mechanisms and feature weighted fusion, so as to achieve beneficial effects such as eliminating redundant features, reducing computational load and improving detection accuracy. The innovation of this invention does not lie in this, and will not be elaborated here.

[0083] In some embodiments, the m-th temporary point in the set of temporary driving points is represented as: ,in, This represents the homogeneous transformation matrix from image coordinate system b to map coordinate system a. This represents the m-th driving feature point in the image coordinate system b.

[0084] In other embodiments, the tentatively set of driving points is in sequential form, which can be denoted as { }, Let m represent the m-th provisional point in the map coordinate system a, where m ranges from 1 to M and M represents the number of driving feature points.

[0085] In other embodiments, the nth provisional point in the provisional driving point set two is represented as: ,in, This represents the homogeneous transformation matrix from geographic coordinate system c to map coordinate system a. This represents the nth geographic coordinate point in geographic coordinate system c.

[0086] In other embodiments, the provisional driving point set one and the provisional driving point set two are respectively in sequential form, and can be denoted as { }and{ }, Let m represent the m-th provisional point in the provisional driving point set 1 in map coordinate system a, where m ranges from 1 to M, and M represents the number of driving feature points. Let n represent the nth provisional point in map coordinate system a, where n ranges from 1 to N and N represents the number of geographic coordinate points.

[0087] In this embodiment of the invention, the homogeneous transformation matrix can be used to perform translation transformation on two-dimensional or three-dimensional coordinate points. It is calculated in advance based on at least one point in the image coordinate system b or the geographic coordinate system c and at least one point in the map coordinate system a. The specific calculation method can be implemented using existing technology, and will not be elaborated here.

[0088] By using a homogeneous transformation matrix, driving feature information and geographical location information can be transformed into the same coordinate system, balancing transformation efficiency and accuracy.

[0089] See Figure 3 The generation of continuous driving trajectories on a preset electronic map based on provisional driving point set one and provisional driving point set two specifically includes:

[0090] S61, based on at least one of the provisional driving point set one and the provisional driving point set two, determine a prefabricated lane line suitable for indicating the road on a preset electronic map.

[0091] S62, by using a preset linear interpolation model, expand the temporary driving point set 1 into expanded set 1, and expand the temporary driving point set 1 into expanded set 2;

[0092] S63, perform point pairing processing on expanded set 1 and expanded set 2 to obtain point pair set, and the difference between expanded set 1 and point pair set is the point difference set;

[0093] S64, calculates the point pair set through a preset point pair similarity calculation model;

[0094] S65, perform point-to-point merging on the point-to-point set based on the calculated similarity set;

[0095] S66, based on the prefabricated lane lines, performs trajectory smoothing on the point difference set and the merged point set to form the driving trajectory.

[0096] In this embodiment of the invention, at least one of the provisional driving point set one and the provisional driving point set two is distributed along the prefabricated lane line. The required prefabricated lane line can pass through the most points in the provisional driving point set one, or / and pass through the most points in the provisional driving point set two.

[0097] In this embodiment of the invention, the preset linear interpolation model can be expressed as:

[0098] ;

[0099] in, This indicates that interpolation needs to be performed between two adjacent points. and Expansion points between, and In order to represent the previous point and the last point The x-coordinate in map coordinate system a and In order to represent the previous point and the last point In the map coordinate system a, the vertical coordinate r ranges from 1 to R, where R equals M-1, or R equals N-1.

[0100] In this embodiment of the invention, each point deviating from the prefabricated lane line can be searched in the point difference set and the point union set, and each point deviating from the prefabricated lane line can be projected onto the prefabricated lane line. Then, trajectory smoothing is performed so that the segments of the prefabricated lane line and their passing points form a driving trajectory.

[0101] After point set expansion processing, the expanded set has more points than the corresponding provisional driving point set to improve the point pairing success rate. After point pair merging processing, the point set is more accurate and concise than the point pair set, breaking through the limitation of driving trajectory being restricted by image information and difficult to further refine.

[0102] Optionally, see Figure 4 S63 includes S631 to S638.

[0103] S631, respectively set up two new sets, one and the other, which are initially empty sets;

[0104] S632, determine the k-th inspected point 1 in the extended set 1, where k represents a positive integer accumulated from 1 to K, and K represents the length of the extended set 1;

[0105] S633, check if there exists a second test point in the second expansion set that is suitable to be paired with the kth test point. If yes, execute S634; otherwise, execute S635.

[0106] S634, store the k-th inspected point one and the inspected point two suitable for pairing with it in the newly created set one in point-to-point form;

[0107] S635, store the kth inspected point one in the newly created set two in single-point form;

[0108] S636, determine if k is less than K. If yes, execute S637; otherwise, execute S638.

[0109] S637, after setting k=k+1, return to S632 so that the kth inspected point is updated to the next inspected point;

[0110] S638, designate the newly created set 1 as a point-to-point set, and designate the newly created set 2 as a point-difference set.

[0111] In this embodiment of the invention, the newly created set can be presented as... An array stores a pair of points in each row, or a newly created set can be presented as an array. An array stores a pair of points in each column; a new set can be a one-dimensional array, storing one point in each row or column.

[0112] During the traversal of the expansion set, two newly created point sets are used to classify all points in the expansion set to prevent errors or omissions, which helps to ensure the reliability and stability of point pairing processing.

[0113] Optionally, S633 includes:

[0114] Using the kth inspected point as the center and a preset distance threshold as the radius, the detection range is delineated. For example, the preset distance threshold can be 0.1 cm or 0.5 cm, etc.

[0115] When any one of the test points 2 in the expanded set 2 is within the detection range, it is determined that the test point 2 within the detection range is suitable to be paired with the kth test point 1.

[0116] When each of the two test points in the extended set is outside the detection range, there is no test point two suitable for pairing with the k-th test point one.

[0117] By utilizing a circular detection range, the paired or individual attributes of each inspected point can be detected, which helps to balance detection efficiency and accuracy.

[0118] Optionally, the preset distance similarity calculation model is represented as follows:

[0119] ;

[0120] in, Represents Euclidean distance similarity. and Let a and b represent the x-coordinates of the two points in the point pair in the map coordinate system a. and These represent the ordinates of the two points in the point pair in the map coordinate system a, which helps to balance the efficiency and accuracy of distance similarity calculation.

[0121] Optionally, see Figure 5 S65 includes S651 to S655:

[0122] S651, determine the i-th Euclidean distance similarity in the similarity set, and determine the i-th pair of points in the pair set, where i represents a positive integer accumulated from 1 to I, and I represents the length of either the similarity set or the pair set;

[0123] S652, perform linear weighted merging on the i-th Euclidean distance similarity and the i-th pair of points to obtain the i-th merged point;

[0124] S653, determine if i is less than I. If yes, execute S654; otherwise, execute S655.

[0125] S654, let i=i+1 and return to S651, so that the i-th Euclidean distance similarity is updated to the next Euclidean distance similarity and the i-th pair of points is updated to the next pair of points;

[0126] S655, combine I merge points to obtain a set of merge points.

[0127] During the synchronous traversal of the similarity set and the point pair set, a linear weighted merging method is used to merge all point pairs in the point pair set to prevent omissions and errors, which helps to ensure the reliability, stability and accuracy of the point pair merging process.

[0128] Optionally, the i-th merging point is represented as: ,in, and Let i represent the i-th pair of points in map coordinate system a, specifically the first and second points being checked. This represents the similarity of the i-th Euclidean distance, ensuring the efficiency and accuracy of merging each pair of points.

[0129] Using the vehicle tracking system and method that combines edge computing and attention mechanism described above, when the attention mechanism is applied to the edge computing architecture, driving image information and geographic location information are used as two types of source information. Considering that driving feature information and geographic location information are in different coordinate systems, after coordinate system transformation, two sets of driving points are formed in the same coordinate system. This has the ability to be compatible with two types of source information, which makes it easier to generate driving trajectories on electronic maps and helps to improve the continuity and accuracy of driving trajectories.

[0130] Generally, the computer instructions used to implement the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself which is temporarily propagating.

[0131] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0132] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks based on TensorFlow, PyTorch, etc., can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet using an Internet service provider).

[0133] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1.A vehicle tracking method combining edge computing with attention mechanism, characterized in that, Applied in an edge computing architecture, comprising the following steps: Respectively receiving driving image information and geographical position information suitable for indicating that an intelligent vehicle is driving on a road; Generating driving feature information according to the driving image information and a preset attention mechanism; Converting the driving feature information from an image coordinate system to a map coordinate system in which a preset electronic map is located, to obtain a tentative driving point set one; Converting the geographical position information from a geographical coordinate system to the map coordinate system, to obtain a tentative driving point set two; Generating a continuous driving trajectory on the preset electronic map according to the tentative driving point set one and the tentative driving point set two; Wherein, generating a continuous driving trajectory on the preset electronic map according to the tentative driving point set one and the tentative driving point set two comprises: According to at least one of the tentative driving point set one and the tentative driving point set two, determining a prefabricated lane line suitable for indicating the road on the preset electronic map; Through a preset linear interpolation model, expanding the tentative driving point set one into an expansion set one, and expanding the tentative driving point set one into an expansion set two; Point pairing processing is performed on the expansion set one and the expansion set two to obtain a point pair set, and the difference between the expansion set one and the point pair set is a point difference set; The point pair set is calculated through a preset point pair similarity calculation model; According to the calculated similarity set, the point pair set is subjected to point pair merging processing; According to the prefabricated lane line, the point difference set and the merged point set generated after merging are subjected to trajectory smoothing processing to form the driving trajectory. 2.The vehicle tracking method of claim 1, wherein, Generating driving feature information according to the driving image information and a preset attention mechanism comprises: Preprocessing the driving image information to obtain multiple sizes of the same size of the driving image; Through a pre-trained vehicle detection model, each of the driving image is subjected to feature detection processing based on a channel and spatial two-dimensional attention mechanism, and the channel attention feature map and the spatial attention feature map corresponding thereto generated after detection are subjected to feature fusion processing; The multiple driving feature points generated after fusion are combined into the driving feature information. 3.The vehicle tracking method of claim 1, wherein, The mth tentative point in the tentative set of driving points is denoted as wherein, denotes a homogeneous transformation matrix from the image coordinate system b to the map coordinate system a, denotes the mth driving feature point of the driving feature information in the image coordinate system b; The nth tentative point two in the tentative set of driving points two is expressed as: wherein, denotes a homogeneous transformation matrix from the geographic coordinate system c to the map coordinate system a, denotes the nth geographic coordinate point of the geographic position information in the geographic coordinate system c. 4.The vehicle tracking method of claim 3, wherein, The point pairing processing of the expansion set one and the expansion set two comprises: Respectively setting newly created set one and newly created set two, both initially empty; Determining the kth inspected point one in the expansion set one; Wherein, k represents a positive integer accumulated from 1 to K, and K represents the length to which the expansion set one belongs; Detecting whether there is an inspected point two suitable for pairing with the kth inspected point one in the expansion set two; If yes, the kth inspected point one and the inspected point two suitable for pairing therewith are stored in the newly created set one in the form of a point pair; If not, the kth inspected point one is stored in the newly created set two in the form of a single point; Judging whether k is less than K; If yes, the kth inspected point one is updated to the next inspected point one; If not, the newly created set one is determined as the point pair set, and the newly created set two is determined as the point difference set. 5.The vehicle tracking method of claim 4, wherein, Detecting whether there is an inspected point two suitable for pairing with the kth inspected point one in the expansion set two comprises: A detection range is circled with the kth detection point one as a center and a preset distance threshold as a radius; When any of the second detection points in the extended set two is in the detection range, it is determined that the second detection point in the detection range is suitable for pairing with the kth detection point one; When each of the second detection points in the extended set two is out of the detection range, there is no second detection point suitable for pairing with the kth detection point one. 6.The vehicle tracking method of claim 4, wherein, The preset point pair similarity calculation model is represented as: ； wherein, denotes the Euclidean distance similarity, and denote the horizontal coordinates of the two points of the point pair, respectively, in the map coordinate system a, and denote the vertical coordinates of the two points of the point pair, respectively, in the map coordinate system a. 7.The vehicle tracking method of claim 4, wherein, The point pair set is processed for point pair merging according to the similarity set, which includes: An i th Euclidean distance similarity is determined in the similarity set, and an i th point pair is determined in the point pair set; Wherein, i represents a positive integer accumulated from 1 to I, I represents the length of any one of the similarity set and the point pair set; The i th Euclidean distance similarity and the i th point pair are processed for linear weighted merging to obtain an i th merged point; It is judged whether i is less than I; If yes, the i th Euclidean distance similarity is updated as a next Euclidean distance similarity, and the i th point pair is updated as a next point pair; If no, I merged points are combined to obtain the merged point set. 8.The vehicle tracking method of claim 7, wherein, The i-th said merging point is represented as: wherein, and the i-th said point pair in said map coordinate system a, represents the i-th said Euclidean distance similarity. 9.A vehicle tracking system combining edge computing with attention mechanism, characterized in that, It includes: A first edge computing device is used to receive driving image information suitable for indicating that an intelligent vehicle drives on a road, generate driving feature information according to the driving image information and a preset attention mechanism, and send the driving feature information; A second edge computing device is used to respectively receive the driving feature information and geographic location information suitable for indicating that the intelligent vehicle drives on the road, convert the driving feature information from an image coordinate system to a map coordinate system in which a preset electronic map is located to obtain a tentative driving point set one, convert the geographic location information from a geographic coordinate system to the map coordinate system to obtain a tentative driving point set two, and generate a continuous driving trajectory on the preset electronic map according to the tentative driving point set one and the tentative driving point set two; Wherein, generating a continuous driving trajectory on the preset electronic map according to the tentative driving point set one and the tentative driving point set two includes: According to at least one of the tentative driving point set one and the tentative driving point set two, a prefabricated lane line suitable for indicating the road is determined on the preset electronic map; The tentative driving point set one is expanded into an expanded set one and the tentative driving point set one is expanded into an expanded set two through a preset linear interpolation model; Point pairing processing is performed on the expanded set one and the expanded set two to obtain a point pair set, and the difference between the expanded set one and the point pair set is a point difference set; The point pair set is calculated through a preset point pair similarity calculation model; Point pair merging processing is performed on the point pair set according to the calculated similarity set; According to the prefabricated lane line, trajectory smoothing processing is performed on the point difference set and a merged point set generated after merging to form the driving trajectory.

Images