Trajectory binding methods, model training methods, devices, equipment and storage media

Abstract

This application provides a trajectory binding method, model training method, apparatus, device, and storage medium, relating to the field of artificial intelligence technology. This application can be applied to fields such as mapping, transportation, and vehicle-mounted systems. The method includes: for a target location point in a target movement trajectory, obtaining the location sequence corresponding to the target location point; based on road network information, obtaining single-point feature information corresponding to multiple location points respectively; determining observation parameters of the target location point based on the single-point feature information corresponding to the multiple location points respectively; the observation parameters of the target location point are used to indicate the probability that the target location point is mapped to at least one candidate road respectively; based on the observation parameters of the target location point, determining the target road mapped to by the target location point from at least one candidate road. The technical solution provided by this application improves the accuracy of trajectory binding.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a trajectory binding method, a model training method, an apparatus, a device, and a storage medium. Background Technology

[0002] In the process of providing services, map applications often need to predict the actual roads that vehicles will travel on, that is, to map the location points on the vehicle's movement trajectory to the corresponding roads. This process can also be called trajectory binding.

[0003] In related technologies, a location point is mapped to the nearest road based solely on its distance from each road. However, due to factors such as poor signal strength, location drift (i.e., a significant discrepancy between the location point and its actual position) often occurs for a single location point, resulting in low accuracy of trajectory mapping. Summary of the Invention

[0004] This application provides a trajectory binding method, a model training method, an apparatus, a device, and a storage medium. The technical solution is as follows:

[0005] According to one aspect of the embodiments of this application, a trajectory binding method is provided, the method comprising:

[0006] For a target location point in the target movement trajectory, obtain the position sequence corresponding to the target location point; wherein, the position sequence includes multiple location points located in the target movement trajectory, and the multiple location points include the target location point and at least one associated location point of the target location point;

[0007] Based on road network information, single-point feature information corresponding to the multiple positioning points is obtained; wherein, the road network information includes multiple roads contained in the road network map and the attribute information of each road;

[0008] Based on the single-point feature information corresponding to the plurality of positioning points, the observation parameters of the target positioning point are determined; wherein, the observation parameters of the target positioning point are used to indicate the probability that the target positioning point is mapped to the at least one candidate road respectively;

[0009] Based on the observation parameters of the target location point, the target road mapped to the target location point is determined from at least one candidate road.

[0010] According to one aspect of the embodiments of this application, a model training method is provided, the method comprising:

[0011] For a first location point in the sample movement trajectory, obtain the position sequence corresponding to the first location point; wherein, the position sequence includes multiple location points located in the sample movement trajectory, and the multiple location points include the first location point and at least one associated location point of the first location point;

[0012] Based on road network information, single-point feature information corresponding to the multiple positioning points is obtained; wherein, the road network information includes multiple sample roads contained in the road network map and attribute information of each sample road, and the multiple sample roads include positive sample roads and negative sample roads corresponding to the first positioning point;

[0013] Based on the single-point feature information corresponding to the multiple positioning points, the observation parameters of the first positioning point are determined through the first network model; wherein, the observation parameters of the first positioning point are used to indicate the probability that the first positioning point is mapped to the multiple sample roads respectively.

[0014] Based on the observation parameters of the first positioning point, the target sample road mapped to the first positioning point is determined from the multiple sample roads;

[0015] Based on the target sample road and the positive and negative sample roads corresponding to the first positioning point, the parameters of the first network model are adjusted.

[0016] According to one aspect of the embodiments of this application, a trajectory binding device is provided, the device comprising:

[0017] The sequence acquisition module is used to acquire a position sequence corresponding to a target positioning point in a target movement trajectory; wherein, the position sequence includes multiple positioning points located in the target movement trajectory, and the multiple positioning points include the target positioning point and at least one associated positioning point of the target positioning point;

[0018] The information acquisition module is used to acquire single-point feature information corresponding to the multiple positioning points based on road network information; wherein, the road network information includes multiple roads contained in the road network map and the attribute information of each road;

[0019] The parameter determination module is used to determine the observation parameters of the target positioning point based on the single-point feature information corresponding to the plurality of positioning points respectively; wherein, the observation parameters of the target positioning point are used to indicate the probability that the target positioning point is mapped to at least one candidate road respectively;

[0020] The road determination module is used to determine the target road mapped to the target positioning point from the at least one candidate road based on the observation parameters of the target positioning point.

[0021] According to one aspect of the embodiments of this application, a model training apparatus is provided, the apparatus comprising:

[0022] The sequence acquisition module is used to acquire a position sequence corresponding to a first positioning point in the sample movement trajectory; wherein, the position sequence includes multiple positioning points located in the sample movement trajectory, and the multiple positioning points include the first positioning point and at least one associated positioning point of the first positioning point.

[0023] The information acquisition module is used to acquire single-point feature information corresponding to the multiple positioning points based on road network information; wherein, the road network information includes multiple sample roads contained in the road network map and attribute information of each sample road, and the multiple sample roads include positive sample roads and negative sample roads corresponding to the first positioning point;

[0024] The parameter determination module is used to determine the observation parameters of the first positioning point based on the single-point feature information corresponding to the multiple positioning points, through a first network model; wherein, the observation parameters of the first positioning point are used to indicate the probability that the first positioning point is mapped to the multiple sample roads respectively.

[0025] The road determination module is used to determine the target sample road to which the first positioning point is mapped from the plurality of sample roads based on the observation parameters of the first positioning point;

[0026] The parameter adjustment module is used to adjust the parameters of the first network model based on the target sample road and the positive and negative sample roads corresponding to the first positioning point.

[0027] According to one aspect of the embodiments of this application, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to implement the above-described trajectory binding method or the above-described model training method.

[0028] According to one aspect of the embodiments of this application, a computer-readable storage medium is provided, wherein at least one instruction, at least one program, code set or instruction set is stored in the computer-readable storage medium, wherein the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the above-described trajectory binding method or the above-described model training method.

[0029] According to one aspect of the embodiments of this application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned trajectory binding method or the aforementioned model training method.

[0030] The technical solutions provided in this application embodiment may have the following beneficial effects:

[0031] By acquiring the location sequence corresponding to the location point, which includes multiple location points (the location point itself and its associated location points), and obtaining the attribute information of each road and the single-point feature information of each location point, the observation parameters of the location point can be generated by referring to the single-point feature information of the location point and its associated location points. Based on the observation parameters, the road corresponding to the location point can be predicted (i.e., mapped to the predicted road). Because the amount of information referenced in the location point mapping process is enriched, the observation parameters of the generated location point are more accurate, thereby improving the accuracy of trajectory binding. Attached Figure Description

[0032] Figure 1 This is a flowchart of a trajectory binding method provided in one embodiment of this application;

[0033] Figure 2 This is a schematic diagram of a positioning point provided in one embodiment of this application;

[0034] Figure 3 This is a schematic diagram of the interval distance provided in one embodiment of this application;

[0035] Figure 4 This is a road network map provided in one embodiment of this application;

[0036] Figure 5 This is a flowchart of a trajectory binding method provided in another embodiment of this application;

[0037] Figure 6 This is a schematic diagram of an NNs model provided in one embodiment of this application;

[0038] Figure 7 This is a schematic diagram of an RNN model provided in one embodiment of this application;

[0039] Figure 8 This is a schematic diagram of an LSTM model provided in one embodiment of this application;

[0040] Figure 9 This is a schematic diagram illustrating the generation of observation values ​​provided in one embodiment of this application;

[0041] Figure 10 This is a schematic diagram of a first network model and a second network model provided in one embodiment of this application;

[0042] Figure 11 This is a flowchart of a model training method provided in one embodiment of this application;

[0043] Figure 12 This is a road network map provided in another embodiment of this application;

[0044] Figure 13 This is a simplified schematic diagram of a road network map provided in one embodiment of this application;

[0045] Figure 14 This is a road network map provided in another embodiment of this application;

[0046] Figure 15 This is a road network map provided in another embodiment of this application;

[0047] Figure 16 This is a block diagram of a trajectory binding device provided in one embodiment of this application;

[0048] Figure 17 This is a block diagram of a model training apparatus provided in one embodiment of this application;

[0049] Figure 18 This is a block diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0050] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of methods consistent with some aspects of this application as detailed in the appended claims.

[0051] First, some terms used in the embodiments of this application will be explained.

[0052] Traffic restrictions: In the road network map, restrictions are placed on whether the corresponding road is passable based on the actual traffic rules.

[0053] Main and auxiliary roads: These correspond to the concepts of main roads and auxiliary roads in actual roads. Main roads and auxiliary roads are usually close to each other.

[0054] Location point: For vehicles, the location information representing the vehicle's current position is obtained by the terminal navigation software through the positioning system carried by the terminal. The location point cannot guarantee that it will accurately reflect the vehicle's true position at all times and in all locations (i.e., the location point can drift).

[0055] Road network map: refers to a directed graph model that uses real-world road information and traffic rules as a background.

[0056] Trajectory binding: In the context of vehicle driving scenarios, this refers to mapping the location points recorded in the vehicle's time series (which may have drift issues) onto the roads in the road network map, that is, mapping them onto the roads where the vehicle is actually driving.

[0057] Track binding accuracy: Each location point is uniquely bound to the most reliable road. If the vehicle is indeed traveling on that road at that time, the binding is considered correct; otherwise, it is considered incorrect. Track binding accuracy refers to the proportion of location points correctly bound to roads. A higher track binding accuracy value is better.

[0058] HMM Model: The HMM (Hidden Markov Model) algorithm is a simple dynamic Bayesian network, a type of Markov chain. Its state cannot be directly observed, but can be observed through a sequence of observation vectors. Each observation vector represents various states through certain probability density distributions, and each observation vector is generated by a sequence of states with a corresponding probability density distribution.

[0059] XGBoost Model: The XGBoost (Extreme Gradient Boosting) algorithm can be considered an engineering implementation of the GBDT (Gradient Boosting Decision Tree) algorithm. However, this algorithm has been optimized and improved in many aspects such as classification regularization, parallel data sampling, and missing value handling strategies, which can achieve faster training and better prediction results.

[0060] NNs Model: NNs (Neural Networks) algorithms are mathematical models that mimic the behavior of animal neural networks to perform distributed parallel information processing. They mainly consist of an input layer, hidden layers, and an output layer. Depending on the complexity of the system, they adjust the weight relationships between a large number of interconnected nodes to achieve the desired fit.

[0061] RNN Model: RNNs (Recurrent Neural Networks) are a type of recurrent neural network that takes sequential data as input, recursively moves along the direction of the sequence, and all nodes (recurrent units) are connected in a chain-like manner. RNN networks add a "memory" component to traditional neural networks; the state at each time step is composed of the current input and the existing memory, making it a deep learning model specifically designed for processing sequential data.

[0062] LSTM Model: LSTM (Long Short Term Memory Networks) is a type of recurrent neural network specifically designed to address the long-term dependency problem inherent in general RNNs. Compared to ordinary RNN models, its main improvement lies in the addition of three gate controllers: the input gate, the output gate, and the forget gate. These gate controllers enhance the control over the weights remembered at different time points and incorporate cross-layer connections to mitigate the vanishing gradient problem.

[0063] Intelligent Vehicle Infrastructure Cooperative Systems (IVICS), or vehicle-road cooperative systems for short, represent a development direction for Intelligent Transportation Systems (ITS). IVICS utilizes advanced wireless communication and next-generation Internet technologies to implement comprehensive, real-time dynamic information interaction between vehicles and infrastructure. Based on the collection and fusion of dynamic traffic information across all times and spaces, it conducts active vehicle safety control and cooperative road management, fully realizing effective collaboration between people, vehicles, and roads to ensure traffic safety, improve traffic efficiency, and ultimately form a safe, efficient, and environmentally friendly road traffic system. This application provides a trajectory binding method that improves the accuracy of trajectory binding, thereby contributing to the optimization of intelligent vehicle-road cooperative systems.

[0064] Track binding is a fundamental service in the mapping industry. The technical solution provided in this application can provide information such as missing roads, traffic rule errors, and real-time road opening or closure for the underlying road network information; it can perform real-time traffic flow statistical analysis to determine road congestion; it can be used for data mining (such as statistical analysis of road speed information, traffic flow information, and turning costs) to provide users with personalized navigation route planning; and it can help establish a map indicator system (such as track binding accuracy, route planning rationality, and determination of whether there is a wrong turn).

[0065] It is also an important source of ground value samples for AI model training;

[0066] It should be noted that the technical solution provided in this application is based on the assumption of traffic rule legality, meaning that the objects (such as vehicles, pedestrians, etc.) are all on passable roads, strictly following traffic rules, and there are no violations of traffic rules or errors in road network information. In other words, the target's movement trajectory must comply with traffic rules. If the object does indeed violate traffic rules by moving against the flow of traffic, and is indeed moving on a certain road, but does not meet traffic rules, the road will still be considered blocked, and the road connection will be considered unsuccessful.

[0067] The method provided in this application can be executed by a computer device, which refers to an electronic device with data computing, processing, and storage capabilities. This computer device can be a terminal such as a PC (Personal Computer), tablet computer, smartphone, wearable device, or intelligent robot; or it can be a server. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0068] The technical solution of this application will be described and illustrated below through several embodiments.

[0069] Please refer to Figure 1 This document illustrates a flowchart of a trajectory binding method provided in an embodiment of this application. In this embodiment, the method is illustrated using the computer device described above. The method may include the following steps (101-104):

[0070] Step 101: For the target location points in the target movement trajectory, obtain the position sequence corresponding to the target location points.

[0071] The location sequence includes multiple positioning points located in the target's movement trajectory, and the multiple positioning points include the target positioning point and at least one associated positioning point of the target positioning point.

[0072] In some embodiments, the target movement trajectory may consist of a series of consecutive points, and the aforementioned positioning points (such as target positioning points or associated positioning points) are points selected from these points. In some embodiments, the target movement trajectory may be fitted as a trajectory line, and the aforementioned positioning points (such as target positioning points or associated positioning points) are points selected from this trajectory line. Optionally, the target movement trajectory may refer to the movement trajectory of any movable organism or object. For example, the movement trajectory of a vehicle can be obtained by installing a device with positioning capabilities on the vehicle; or, for example, a person or animal can record its movement trajectory by carrying a device with positioning capabilities.

[0073] In some embodiments, the multiple positioning points are points corresponding to multiple time points, that is, each positioning point corresponds to a time point, and the time points of each positioning point are different. In other words, there is a temporal order among the multiple positioning points, and there is always a sequential order between any two positioning points. It should be noted that, in the embodiments of this application, two positioning points with different time points but the same location are considered as two different positioning points.

[0074] In some embodiments, the position sequence of the target location point is a sequence of location points selected based on the target location point. This position sequence includes at least one associated location point preceding the target location point and / or at least one associated location point following the target location point. The number of associated points preceding and following the target location point may be the same or different, and this embodiment does not impose specific limitations on this. In some cases, if the target location point is the last location point on the target movement trajectory, since there are no reference location points after the target location point, the associated location points of the target location point only include the location points preceding the target location point. In other cases, if the target location point is the first location point on the target movement trajectory, since there are no reference location points before the target location point, the associated location points of the target location point only include the location points following the target location point.

[0075] In some embodiments, an associated location point refers to a location point that has a first association relationship with a target location point. Optionally, the first association relationship includes at least one of the following:

[0076] (1) There is no passable road junction between the location point and the target location point;

[0077] (2) The number of positioning points located before the target positioning point and separated from the target positioning point is less than or equal to the first threshold;

[0078] (3) The number of positioning points located after the target positioning point and the number of positioning points between the positioning point and the target positioning point is less than or equal to the second threshold.

[0079] (4) The positioning point is located before the target positioning point, and the distance between the positioning point and the target positioning point is less than or equal to the third threshold;

[0080] (5) The location point is located after the target location point, and the distance between the location point and the target location point is less than or equal to the fourth threshold.

[0081] Among them, there is no passable road fork between the positioning point and the target positioning point. It can be specified that there is no road fork between the positioning point and the target positioning point, or there is a road fork between the positioning point and the target positioning point, but due to physical conditions (such as the road fork being blocked so that it is not passable to another road) or traffic rules (such as according to traffic rules, it is not possible to pass through the road fork to another road at the corresponding time of the positioning point).

[0082] In some embodiments, the target location point corresponds to the associated location point of the target candidate road in at least one candidate road, which refers to a location point on the target candidate road that has no passable road forks.

[0083] like Figure 2 As shown, a road fork can refer to an intersection that branches into multiple different roads, such as road fork 13; a road fork can also refer to an intersection with a large corner, such as road fork 14.

[0084] exist Figure 2 In the example, if the target location point is G t+4 G needs to be calculated t+4 The actual bound road is the possibility of candidate road 11, which can be followed along from G. t+4 Start selecting location G forward and backward along candidate road 11. t+3 G t+5 G t+6 G t+7 and G t+8 As G t+4 The associated location point corresponding to candidate road 11 (G cannot be selected) t+2 As G t+4 The associated location point corresponding to candidate road 11, because along candidate road 11, G t+2 With G t+4 (There are road forks in the road); if it is necessary to calculate G t+4 The actual bound road is the possibility of candidate road 12, which can be followed along from G. t+4 Start selecting location G forward and backward along candidate road 12. t+3 and G t+5 As G t+4 The associated location point corresponding to candidate road 12 (along candidate road 12, G) t+6 With G t+4 There are road forks between them, therefore G cannot be selected. t+6 As G t+4 Corresponding location point to candidate road 12; G t+7 G t+8 Similarly).

[0085] In some embodiments, the distance between the positioning point and the target positioning point can be specified as the straight-line distance between the positioning point and the target positioning point, such as... Figure 3 The distance L1 between the positioning points 16 and 17 shown.

[0086] In some embodiments, the distance between the location point and the target location point can be specified as the projected distance between the location point and the target location point corresponding to the target candidate road, such as... Figure 3 The distance L2 between the positioning points 16 and 17 shown.

[0087] The first threshold can be 0, 1, 3, 5, 20, 34, etc.; the second threshold can be 0, 2, 10, 15, 30, etc.; the third threshold can be 2 meters, 5 meters, 35 meters, 50 meters, 100 meters, etc.; and the fourth threshold can be 5 meters, 15 meters, 60 meters, 200 meters, etc. It should be noted that the specific values ​​of the first, second, third, and fourth thresholds can be set by relevant technical personnel according to actual conditions. The thresholds (such as the first, second, third, and fourth thresholds) corresponding to different intersections can be the same or different, and this application embodiment does not specifically limit this.

[0088] Step 102: Based on the road network information, obtain the single-point feature information corresponding to multiple positioning points.

[0089] In some embodiments, road network information includes multiple roads contained in the road network map and the attribute information of each road. In some embodiments, the road attribute information includes traffic regulations such as U-turns, traffic lights, and traffic restrictions in real-world roads. Optionally, the attribute information of each road can be virtually represented as a directed graph model in the background to obtain the corresponding road network map, such as the map displayed on the client side of a map application. In some embodiments, the single-point feature information corresponding to each positioning point can be obtained by using the relationships such as distance, angle, and direction between multiple positioning points and roads in the road network map.

[0090] In some embodiments, road network information includes at least one of the following:

[0091] (1) The bifurcation angle of the road forks between roads.

[0092] In some embodiments, the bifurcation angle of a road fork can refer to the angle between the connecting path between roads and the individual roads; it can also refer to the angle by which the actual movement trajectory needs to deflect during the process of moving from one road to another through a road fork. For example... Figure 4 As shown, one can turn from road 19 onto road 20 through road fork 18. The fork angle can be considered as angle 21, angle 22, or both angle 21 and angle 22.

[0093] (2) The average of the absolute values ​​of the vertical distances between at least two location points at the road forks between the roads.

[0094] In some embodiments, the at least two positioning points include at least one positioning point located before the road fork and one positioning point located after the road fork. By calculating the vertical distances between the positioning points located before and after the road fork in the predicted road chain and their respective corresponding roads, and averaging the absolute values ​​of the vertical distances corresponding to each positioning point, the smaller the average value, the higher the probability of accurate prediction.

[0095] (3) Speed ​​characteristics of the road.

[0096] In some embodiments, the speed limits for objects (such as various vehicles, pedestrians, etc.) on different roads may vary; the speeds of objects on different roads may also differ. For example, vehicles generally travel faster on main roads and slower on auxiliary roads. Optionally, the speed characteristics of a road include the average speed of objects on the road. Further, the speed characteristics of a road can be refined to the average speed of objects on the road at different time periods.

[0097] (4) Road grade.

[0098] In some embodiments, roads can be classified by grade. For example, roads can be classified according to the average speed of objects moving within them; the higher the average speed, the higher the grade; and the lower the average speed, the lower the grade. As another example, roads can be classified according to road attributes, such as expressways being higher in grade than national highways, national highways being higher in grade than provincial highways, provincial highways being higher in grade than county roads, and county roads being higher in grade than ordinary roads. Yet another example is that roads can be classified according to traffic / pedestrian flow; the higher the traffic / pedestrian flow, the higher the road's grade; and the lower the traffic / pedestrian flow, the lower the road's grade.

[0099] (5) The types of lanes included in the road and the number of each type of lane.

[0100] A lane is a section of road designed for one vehicle only, in which two or more vehicles cannot travel side-by-side. In some embodiments, a lane includes a fast lane and a slow lane. Road network information may include the total number of lanes in a road, or the number of fast lanes and slow lanes.

[0101] (6) The width of the road.

[0102] The width of a road can refer to the total width of the road, or it can refer to the width corresponding to the two driving directions of the road.

[0103] (7) Lane width information of the road.

[0104] In some embodiments, road network information also includes traffic restriction information, which is information on whether roads and the inter-road traffic is permitted, determined in the road network map based on actual traffic rules.

[0105] In some embodiments, the single-point feature information corresponding to a location point includes feature information of at least one candidate road corresponding to the location point. That is, if a location point corresponds to multiple candidate roads, then the location point has a set of feature information for each candidate road.

[0106] In some embodiments, the single-point feature information corresponding to the positioning point includes at least one of the following:

[0107] (1) The deflection angle of the positioning point relative to the reference direction, and the average deflection angle corresponding to the positioning point, wherein the average deflection angle is the mean of the deflection angles of at least one positioning point relative to the reference direction prior to the specified position.

[0108] Optionally, the reference direction is due north. Of course, the reference direction can also be other directions, such as due south, due east, due west, etc., and this application embodiment does not specifically limit this. In some embodiments, the deflection angle of the positioning point relative to the reference direction refers to the angle of rotation from the reference direction to the driving direction corresponding to the positioning point. The driving direction can refer to the tangent direction of the target's movement trajectory at the positioning point. This deflection angle has a positive or negative sign, used to indicate the direction of deflection. For example, if the reference direction deflects to the driving direction to the left, the deflection angle is positive; if the reference direction deflects to the driving direction to the right, the deflection angle is negative. Of course, the rule for the positive and negative signs of the deflection angle is subjectively set, and those skilled in the art can set it according to actual conditions; this application embodiment does not specifically limit this.

[0109] In an exemplary embodiment, at least one location point is a location point located before the specified site and at a distance less than a seventh threshold.

[0110] In an exemplary embodiment, 20 positioning points within 100 meters of the initial positioning point are selected, and the deflection angles of these 20 positioning points are obtained for each point. The average deflection angle of these 20 positioning points is then calculated (including the sign of the deflection angle). The smaller the difference between the deflection angle of the positioning point relative to the reference direction and the average deflection angle of these 20 positioning points, the lower the likelihood of changing roads; conversely, the larger the difference between the deflection angle of the positioning point relative to the reference direction and the average deflection angle of these 20 positioning points, the higher the likelihood of changing roads.

[0111] (2) Vertical distance of the location point relative to the candidate road.

[0112] The vertical distance is signed. For example, along the direction of movement of the target trajectory, if the location point is on the left side of the candidate road, the sign of the corresponding vertical distance is negative; if the location point is on the right side of the candidate road, the sign of the corresponding vertical distance is positive. Alternatively, along the direction of movement of the target trajectory, if the location point is on the left side of the candidate road, the sign of the corresponding vertical distance is positive; if the location point is on the right side of the candidate road, the sign of the corresponding vertical distance is negative.

[0113] (3) The absolute value of the vertical distance between the location point and the candidate road.

[0114] The absolute value of the vertical distance between the positioning point and the candidate road only includes the distance between the positioning point and the farthest point of the candidate road, and does not include the direction of the positioning point relative to the candidate road.

[0115] (4) The difference between the absolute value of the vertical distance of the location point relative to the candidate road and the average vertical distance, where the average vertical distance is the average of the absolute values ​​of the vertical distances of at least one location point before the specified location relative to the candidate road.

[0116] The explanation of at least one positioning point can be found in the above text, and will not be repeated here.

[0117] In an exemplary embodiment, 15 positioning points within 150 meters of the previous positioning point are selected. The absolute values ​​of the vertical distances of these 15 positioning points relative to the candidate road are obtained, and the mean of these 15 absolute values ​​is calculated. The smaller the difference between the absolute value of the vertical distance of the positioning point relative to the candidate road and the mean of these 15 absolute values, the lower the probability of switching roads; conversely, the larger the difference between the absolute value of the vertical distance of the positioning point relative to the candidate road and the mean of these 15 absolute values, the higher the probability of switching roads.

[0118] (5) Positioning accuracy value of the positioning point.

[0119] In some embodiments, the positioning points obtained by the positioning system may have a certain error; this error range is the positioning accuracy. It can be understood that positioning accuracy should actually be a range; for convenience, the maximum value within this range can be used as the positioning accuracy value to represent the positioning accuracy. For example, if the positioning accuracy is 0–10 meters, then “10 meters” is used as the positioning accuracy value to represent the positioning accuracy.

[0120] Optionally, the positioning accuracy may vary in different areas. For example, some areas are relatively open with less interference to positioning, so the positioning accuracy value may be lower, indicating higher positioning accuracy; while some areas have more obstacles to signal transmission and more severe interference to positioning, so the positioning accuracy value may be higher, indicating lower positioning accuracy.

[0121] Optionally, the positioning accuracy of the area where the positioning point is located is provided by the corresponding positioning system.

[0122] (6) The difference between the absolute value of the vertical distance between the positioning point and the candidate road and the positioning accuracy value of the positioning point.

[0123] It is readily understood that the various distances involved in the embodiments of this application may refer to the map distance on the road network map when the map scale remains unchanged; or they may refer to the actual distance calculated according to the map scale of the road network map. The embodiments of this application do not specifically limit this.

[0124] Step 103: Determine the observation parameters of the target positioning point based on the single-point feature information corresponding to the multiple positioning points.

[0125] Among them, the observation parameters of the target positioning point are used to indicate the probability that the target positioning point is mapped to at least one candidate road.

[0126] In some embodiments, the target location point corresponds to at least one candidate road. Based on the single-point feature information corresponding to multiple location points related to the target location point, the probability of the target location point being mapped to at least one candidate road can be obtained.

[0127] Optionally, the candidate road corresponding to the target location point refers to the road whose shortest distance to the target location point is less than the eighth threshold. The specific value of the eighth threshold can be set by relevant technical personnel according to actual conditions, and this embodiment does not impose a specific limitation on it.

[0128] Step 104: Based on the observation parameters of the target location point, determine the target road to which the target location point is mapped from at least one candidate road.

[0129] In some embodiments, based on the probability that a target location point is mapped to at least one candidate road, the road most likely to be located can be determined from at least one candidate road, and that road can be used as the target road to which the target location point is mapped.

[0130] In summary, the technical solution provided in this application obtains the location sequence corresponding to the positioning point, which includes multiple positioning points (the positioning point and its associated positioning points), and obtains the attribute information of each road and the single-point feature information of each positioning point. This allows for the generation of observation parameters for the positioning point by referring to the single-point feature information of the positioning point and its associated positioning points. Based on these observation parameters, the road corresponding to the positioning point is predicted (i.e., mapped to the predicted road). Because the amount of information referenced during the positioning point mapping process is enriched, the generated observation parameters for the positioning point are more accurate, thereby improving the accuracy of trajectory binding.

[0131] In particular, the technical solution provided in this application significantly improves the accuracy of trajectory binding in traffic restriction judgment scenarios and main / auxiliary road switching judgment scenarios.

[0132] Please refer to Figure 5 This document illustrates a flowchart of a trajectory binding method according to another embodiment of this application. In this embodiment, the method is illustrated using the computer device described above. The method may include the following steps (501-507):

[0133] Step 501: For the target location point in the target movement trajectory, obtain the position sequence corresponding to the target location point.

[0134] This step 501 is related to the above. Figure 1 The content of step 101 in the embodiment is the same or similar, and will not be repeated here.

[0135] Step 502: Based on the road network information, obtain the single-point feature information corresponding to multiple positioning points.

[0136] This step 502 is related to the above. Figure 1 The content of step 102 in the embodiment is the same or similar, and will not be repeated here.

[0137] Step 503: Input the single-point feature information corresponding to multiple positioning points into the first network model with memory function in chronological order to generate the association information of the target positioning points.

[0138] In some embodiments, the association information of the target location point is used to indicate the association relationship between the target location point and at least one associated location point. Since the multiple location points are a positional sequence with a sequential relationship, referring to the features of points preceding and following the target location point helps determine the target road mapped to the target location point. Using a first network model, the single-point feature information corresponding to each of the multiple location points can be extracted to obtain the association information of the target location point. Therefore, the association information is centered on the target location point and integrates the single-point feature information of the target location point and at least one associated point.

[0139] In some embodiments, the first network model is a recurrent neural network, which includes a memory unit and a gate controller. The memory unit is used to record information acquired and generated by the first network model, and the gate controller is used to control the retention of information in the first network model.

[0140] In some embodiments, for the nth iteration of multiple iterations of the first network model, the gate controller processes the memory unit, the feature information corresponding to the nth location point, and the iteration result output of the (n-1)th iteration of the first network model to generate the iteration result output of the nth iteration and update the memory unit, where n is a positive integer greater than 1; wherein, the associated information is the iteration result output of the last iteration of the first network model.

[0141] In some embodiments, the gate controller includes at least one of the following: a forget gate, an input gate, and an output gate. Specifically, for the nth loop, the forget gate is used to determine the forgetting status of the information stored in the memory unit after the (n-1)th loop; the input gate is used to determine the embedding degree of the single-point feature information of the input location point in the nth loop within the first network model; and the output gate is used to determine, from the information generated by the first network model during the nth loop, the information used as the loop result output for the nth loop.

[0142] In some embodiments, the first network model is an LSTM (Long Short Term Memory Networks) model.

[0143] Before introducing the LSTM model, let's first introduce the NNs (Neural Networks) model and the RNN (Recurrent Neural Networks) model.

[0144] like Figure 6 As shown, the NNs model includes an input layer 23, a hidden layer 24, and an output layer 25. The nodes of the input layer 23 are [x1, x2, ..., x...]. mAfter several hidden layers 24, the data finally reaches the output layer 25 and yields the predicted result y (y1 and y2 in the figure). For the input sequence data (w1, w2, ..., w...), ... m Each element corresponds to the position of an input node. As can be seen from the diagram, any two input data points w... i and w j Their effect on the final output is determined by different model parameters. The entire w i and w j There is no connection between the two. Since the input layer 23 and the hidden layer 24 are fully connected, it can be considered that each input node is independent and equal for the intermediate layers. The dependencies (such as the order) between w1 and w2, and between w2 and w3 are not reflected in the network structure at all, and the dependencies between sequential data cannot be learned.

[0145] like Figure 7 As shown, (a) is a simplified diagram of the RNN model, and (b) is the expanded form of (a). (x in the diagram) t-1 x t x t+1 ) is the input sequence vector, h represents the hidden state of the network, and o (as shown in the figure) is the hidden state of the network. t-1 o t o t+1 The symbol () represents the output vector. The input and hidden layers are connected via a parameter matrix U, hidden layers at different time steps are connected via a parameter matrix W, and hidden layers and the output layer are connected via a parameter matrix V. Square 26 in the left figure describes a delayed connection, such as from the hidden state h of the previous time step. t-1 The hidden state h at the current time t The connection between them.

[0146] The forward propagation of an RNN can be expressed as a formula:

[0147] a t =b+W*h t-1 +U*x t

[0148] h t =tanh(a t )

[0149] o t =c+V*h t

[0150]

[0151] Where b and c are constants, W, U, and V are constant matrices, and x t The input at time t represents h.t Let o represent the hidden state at time t. t This represents the output at time t. This represents the normalized predicted probability. It's worth noting that parameter sharing also plays a crucial role in RNN models. From... Figure 7 As can be seen from the formula above, the parameter matrices U, W, and V do not change at each time step; instead, the same set of parameters is used. It is precisely because of this parameter sharing that RNN models can handle sequences of arbitrary length, as long as the elements of the sequence are fed into the network one by one in sequence. The RNN model processes these elements like a pipeline. However, this also leads to problems such as vanishing gradients and the inability to learn long-range knowledge.

[0152] like Figure 8 As shown, the most significant improvement of LSTM model 28 compared to the ordinary RNN model 27 is the addition of three gate controllers: the input gate, the output gate, and the forget gate. The three gate controllers have the same structure, primarily consisting of a sigmoid function (as in the α node of LSTM model 28) and a dot product operation. Since the sigmoid function's value range is [0, 1], the gate controller describes the proportion of information that can pass through. A sigmoid value of 0 indicates that no information can pass through, or it can be understood as forgetting all memories. Conversely, a value of 1 indicates that all information can pass through, completely retaining the memory of that branch.

[0153] For RNN model 27, the state at each time step is composed of the current input and the existing memory. However, the problem is that memory capacity is limited, and early memories decay exponentially. To solve this problem, LSTM model 27 uses the existing short-term memory unit h... t Based on this, add a memory unit c t To maintain long-term memory. From the perspective of network structure design, the improvement of LSTM model 28 over RNN model 27 is mainly reflected in the addition of weight control for memory at different time steps through the gate controller, and the addition of cross-layer connections to reduce the impact of gradient vanishing problem.

[0154] Step 504: Based on the association information and road network information of the target location point, the second network model is used to generate the observation parameters of the target location point.

[0155] In some embodiments, after generating the association information of the target location point through the first network model, it is necessary to combine the association information with the road network information. That is, the association information, which integrates the single-point feature information corresponding to multiple location points, is combined with the road network information to obtain the observation parameters of the target location point.

[0156] As described above, the observation parameters of a target location point are used to indicate the probability that the target location point is mapped to at least one candidate road. In some embodiments, the observation parameters include the observation values ​​of at least one candidate road, where the observation values ​​represent the probability that the target location point is mapped to a candidate road. Optionally, the observation values ​​are also referred to as observation probabilities.

[0157] In an exemplary embodiment, such as Figure 9 As shown, in calculating L t,2 and L t,3 When calculating the observed values, an LSTM model was used. This model can handle data point features of variable length and is suitable for L... t,2 The observed values ​​used a total of 29 feature information lists corresponding to i+j+1 positioning points, targeting L. t,3 The observed values ​​used a total of k+m+1 feature information lists corresponding to the location points, totaling 30.

[0158] In an exemplary embodiment, such as Figure 10 As shown, by using the first network model 31 to extract the feature information corresponding to multiple positioning points, the generated h t This refers to the associated information of the target location point, h. t It is in vector form. After vectorizing the road network information, it is compared with h. t After being stitched together, the data is input into the second network model 32, which generates the observation parameters for the target location points. Optionally, the second network model includes at least one fully connected layer.

[0159] Step 505: Obtain the state transition value of at least one candidate road.

[0160] The state transition value indicates the probability of transitioning from the target road to the candidate road from the previous location point of the target location point. Optionally, the state transition value is also called the state transition probability. In some embodiments, the previous location point refers to the location point that precedes and is adjacent to the target location point in the target movement trajectory.

[0161] In some embodiments, a transition probability model is used to determine the probability of transitioning from the previous location point to the target location point along each candidate road, based on the road most likely to be located at the previous location point. This yields the state transition value for each candidate road. A higher state transition value indicates a higher probability of transitioning from the target road to a candidate road. Optionally, the transition probability model is XGBoost (Extreme Gradient Boosting).

[0162] Step 506: Based on the observation values ​​of at least one candidate road and the state transition values ​​of at least one candidate road, determine the observation probability of the target location point corresponding to each candidate road.

[0163] In some embodiments, the observation probability of a candidate road is obtained using a third network model based on the observation values ​​and state transition values ​​of at least one candidate road. The observation probability indicates the likelihood that a target location point will map to a candidate road within a global map, which includes a road network map. Optionally, the third network model is an Hidden Markov Model (HMM).

[0164] Step 507: Determine the target road from at least one candidate road based on the observation probability of the target location point corresponding to each candidate road.

[0165] In some embodiments, based on the observation probability of each location point corresponding to a candidate road, a third network model is used to determine the road mapped to each location point on the target movement trajectory from the perspective of global map optimization, that is, to obtain the target road mapped to the target location point.

[0166] In summary, the technical solution provided in this application involves inputting the single-point feature information corresponding to multiple positioning points into the first network model, extracting it in chronological order, and selectively memorizing the information of previous points stored in the first network model in each loop through the gate controller. This allows the first network model to integrate the single-point feature information corresponding to multiple positioning points in the location sequence based on the chronological order of the multiple positioning points, thereby improving the accuracy of the observation parameters of the target positioning points generated by combining road network information, i.e., improving the accuracy of the observation parameters of a single positioning point, and thus improving the accuracy of trajectory binding.

[0167] In some embodiments, if the target location point meets the first condition, from the above... Figure 1 Step 101 of the embodiment or Figure 5 Step 501 of the embodiment is executed, that is, the trajectory binding method provided in the embodiment of this application is adopted.

[0168] Optionally, the first condition includes that the distance between the target location point and the nearest road fork is less than or equal to a fifth threshold. In some embodiments, from the perspective of trajectory binding efficiency, the trajectory binding method provided in this application embodiment is only applied to location points located near the machine at road forks. Of course, if the performance of the computer equipment allows, or if the efficiency requirement is not high, the trajectory binding method provided in this application embodiment can be applied to every arbitrary location point on the target movement trajectory.

[0169] In some embodiments, if the target location point meets the second condition, the target road mapped to the target location point is determined by comparing the target location point with the previous location point of the target location point.

[0170] Optionally, the second condition includes that the distance between the target location point and the nearest road junction is greater than or equal to a sixth threshold. That is, when the distance to the road junction is far, the trajectory binding method provided in this application embodiment may not be used, but a more efficient but less accurate trajectory binding method may be used instead. Optionally, the sixth threshold is greater than or equal to the fifth threshold.

[0171] In some embodiments, determining the target road mapped to the target location point by comparing the target location point with the previous location point of the target location point includes the following steps:

[0172] 1. Determine the observation probability of the target location point corresponding to each candidate road by using the vertical distance of the target location point relative to each candidate road;

[0173] 2. Obtain the state transition value of at least one candidate road. The state transition value is used to indicate the probability of transitioning from the target road corresponding to the previous positioning point of the target positioning point to the candidate road.

[0174] 3. Based on the observation probabilities of the target location point corresponding to each candidate road and the state transition probabilities of the target location point corresponding to each candidate road, determine the target road to which the target location point is mapped.

[0175] In some embodiments, the shorter the vertical distance between the target location point and the candidate road, the higher the observation probability of the candidate road; the longer the vertical distance between the target location point and the candidate road, the lower the observation probability of the candidate road. Steps 2 and 3 above can refer to the above... Figure 5 The details of steps 505 and 506 in the embodiment will not be repeated here.

[0176] In the above embodiments, the above method is applied only to positioning points that meet the first condition, i.e., only to positioning points whose mapped roads are difficult to determine. Figure 1 Implementation examples or Figure 2 The trajectory binding method provided in the embodiment; for other positioning points where the mapping road is relatively easy to determine (such as positioning points that meet the second condition), other more efficient methods for generating observation parameters can be used, so that accuracy and efficiency can be taken into account in the process of binding the overall trajectory of the target movement trajectory.

[0177] Please refer to Figure 11This document illustrates a flowchart of a model training method provided in an embodiment of this application. In this embodiment, the method is illustrated using the computer device described above. The method may include the following steps (1101-1105):

[0178] Step 1101: For the first positioning point in the sample movement trajectory, obtain the position sequence corresponding to the first positioning point.

[0179] The location sequence includes multiple positioning points located in the sample movement trajectory, and the multiple positioning points include a first positioning point and at least one associated positioning point of the first positioning point.

[0180] Step 1102: Based on the road network information, obtain the single-point feature information corresponding to multiple positioning points.

[0181] The road network information includes multiple sample roads contained in the road network map and the attribute information of each sample road. The multiple sample roads include positive sample roads and negative sample roads corresponding to the first positioning point. Positive sample roads are the roads that actually correspond to the positioning point, and negative sample roads are other roads besides the roads that actually correspond to the positioning point.

[0182] In an exemplary embodiment, such as Figure 12 The road network map shown can be simplified as follows: Figure 13 The simplified map shown, and in Figure 13 The simplified map shown indicates the individual sample roads. For example... Figure 12 As shown, the movement trajectory 33 contains positioning points near road forks, such as positioning point 34; the actual movement path should be: Road 35 – Road 36 – Road 37. Therefore, corresponding to… Figure 13 In the sample, the positive sample roads are roads 38, 39, and 40, and all other roads are negative sample roads.

[0183] Step 1103: Based on the single-point feature information corresponding to multiple positioning points, determine the observation parameters of the first positioning point through the first network model.

[0184] The observation parameters of the first positioning point are used to indicate the probability that the first positioning point is mapped to multiple sample roads respectively.

[0185] In some embodiments, step 1103 further includes the following steps:

[0186] 1. Input the single-point feature information corresponding to multiple positioning points into the first network model in chronological order to generate the association information of the first positioning point. The association information of the first positioning point is used to indicate the relationship between the first positioning point and at least one associated positioning point.

[0187] In some embodiments, the first network model is a recurrent neural network, which includes a memory unit and a gate controller. The memory unit is used to record information acquired and generated by the first network model, and the gate controller is used to control the retention of information in the first network model.

[0188] In some embodiments, for the nth iteration of the first network model's training in the target round, a gate controller processes the memory unit, the feature information corresponding to the nth sample location point, and the loop result output from the (n-1)th iteration of the target round's training to generate the loop result output from the nth iteration and update the memory unit, where n is a positive integer greater than 1. The association information of the first location point obtained from the training in the target round is the loop result output from the last iteration of the first network model's training in the target round.

[0189] 2. Based on the association information and road network information of the first positioning point, the observation parameters of the first positioning point are generated using the second network model.

[0190] Step 1104: Based on the observation parameters of the first positioning point, determine the target sample road to which the first positioning point is mapped from multiple sample roads.

[0191] Step 1105: Adjust the parameters of the first network model based on the target sample road and the positive and negative sample roads corresponding to the first positioning point.

[0192] Steps 1101 to 1105 above can be referenced from the above. Figure 1 Implementation examples or Figure 5 The details of the embodiments will not be repeated here.

[0193] In summary, the technical solution provided in this application embodiment, during the training of the first network model and the second network model, inputs the first network model with the feature information of the first positioning point and associated positioning points, and the first network model has a memory function, which can extract the feature information of the multiple positioning points according to the sequential order between them. This enables the trained first network model to process position sequences with a certain sequential relationship, and thus the mapping accuracy of the roads mapped to the positioning points obtained by the first network model is high, thereby improving the accuracy of trajectory binding.

[0194] In an exemplary embodiment, such as Figure 14 and Figure 15As shown, the trajectory binding path 41 obtained using the technical solution provided in this application is superior to the trajectory binding path 42 obtained by other solutions; the trajectory binding path 43 obtained using the technical solution provided in this application is superior to the trajectory binding path 44 obtained by other solutions. Experiments show that the trajectory binding path accuracy obtained using the technical solution provided in this application can reach 95%.

[0195] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0196] Please refer to Figure 16 This diagram illustrates a block diagram of a trajectory binding device according to an embodiment of this application. The device has the functionality to implement the above-described trajectory binding method example; this functionality can be implemented in hardware or by hardware executing corresponding software. The device can be the computer device described above, or it can be mounted on a computer device. The device 1600 may include:

[0197] The sequence acquisition module 1610 is used to acquire a position sequence corresponding to a target positioning point in a target movement trajectory; wherein the position sequence includes multiple positioning points located in the target movement trajectory, and the multiple positioning points include the target positioning point and at least one associated positioning point of the target positioning point.

[0198] The information acquisition module 1620 is used to acquire single-point feature information corresponding to the multiple positioning points based on road network information; wherein, the road network information includes multiple roads contained in the road network map and the attribute information of each road.

[0199] The parameter determination module 1630 is used to determine the observation parameters of the target positioning point based on the single-point feature information corresponding to the plurality of positioning points respectively; wherein, the observation parameters of the target positioning point are used to indicate the probability that the target positioning point is mapped to at least one candidate road respectively.

[0200] The road determination module 1640 is used to determine the target road mapped to the target positioning point from the at least one candidate road based on the observation parameters of the target positioning point.

[0201] In an exemplary embodiment, the parameter determination module 1630 is configured to:

[0202] The single-point feature information corresponding to the plurality of positioning points is input into a first network model with memory function in chronological order to generate the association information of the target positioning point. The association information of the target positioning point is used to indicate the association relationship between the target positioning point and the at least one associated positioning point.

[0203] Based on the association information of the target location point and the road network information, the observation parameters of the target location point are generated using a second network model.

[0204] In an exemplary embodiment, the first network model is a recurrent neural network. The first network model includes a memory unit and a gate controller. The memory unit is used to record information acquired and generated by the first network model, and the gate controller is used to control the retention of information in the first network model. The parameter determination module 1630 is used to process the memory unit, the feature information corresponding to the nth location point, and the loop result output by the (n-1)th loop of the first network model through the gate controller for the nth loop, generate the loop result output by the nth loop, and update the memory unit, where n is a positive integer greater than 1; wherein, the associated information is the loop result output by the last loop of the first network model.

[0205] In an exemplary embodiment, the gate controller includes at least one of the following: a forget gate, an input gate, and an output gate; wherein, for the nth loop, the forget gate is used to determine the forgetting status of the information stored in the memory unit after the (n-1)th loop is completed; the input gate is used to determine the embedding degree of the single-point feature information of the location point input in the nth loop in the first network model; and the output gate is used to determine, from the information generated by the first network model during the nth loop, the information used as the loop result output of the nth loop.

[0206] In an exemplary embodiment, the associated location point refers to a location point that has a first association relationship with the target location point, and the first association relationship includes at least one of the following:

[0207] There is no passable road junction between the location point and the target location point;

[0208] The number of positioning points located before the target positioning point and spaced apart from the target positioning point is less than or equal to a first threshold.

[0209] The number of positioning points located after the target positioning point and spaced apart from the target positioning point is less than or equal to the second threshold.

[0210] The positioning point is located before the target positioning point, and the distance between the positioning point and the target positioning point is less than or equal to the third threshold.

[0211] The positioning point is located after the target positioning point, and the distance between the positioning point and the target positioning point is less than or equal to the fourth threshold.

[0212] In an exemplary embodiment, the target location point corresponds to the associated location point of the target candidate road in the at least one candidate road, which refers to a location point on the target candidate road that has no passable road forks with respect to the target location point.

[0213] In an exemplary embodiment, the observation parameters include the observation values ​​of the at least one candidate road; the road determination module 1640 is configured to:

[0214] Obtain the state transition value of the at least one candidate road, the state transition value being used to indicate the probability of transitioning from the target road corresponding to the previous location point of the target location point to the candidate road;

[0215] Based on the observation values ​​of the at least one candidate road and the state transition values ​​of the at least one candidate road, the observation probability of the target location point corresponding to each of the candidate roads is determined. The observation probability is used to indicate the possibility that the target location point is mapped to the candidate road within the scope of the global map. The global map includes the road network map.

[0216] The target road is determined from the at least one candidate road based on the observation probability of the target location point corresponding to each of the candidate roads.

[0217] In an exemplary embodiment, the sequence acquisition module 1610 is further configured to, when the target positioning point meets the first condition, start executing the step of acquiring the position sequence corresponding to the target positioning point from the target positioning point in the target movement trajectory.

[0218] The road determination module 1640 is further configured to determine the target road mapped to the target positioning point by means of the target positioning point and the previous positioning point of the target positioning point when the target positioning point meets the second condition.

[0219] In an exemplary embodiment, the first condition includes the distance between the target location point and the nearest road fork being less than or equal to a fifth threshold.

[0220] In an exemplary embodiment, the second condition includes that the distance between the target location point and the nearest road fork is greater than or equal to a sixth threshold; the road determination module 1640 is configured to:

[0221] The observation probability of the target location point corresponding to each of the candidate roads is determined by the vertical distance of the target location point relative to each of the candidate roads;

[0222] Obtain the state transition value of the at least one candidate road, the state transition value being used to indicate the probability of transitioning from the target road corresponding to the previous location point of the target location point to the candidate road;

[0223] Based on the observation probability of the target location point corresponding to each of the candidate roads, and the state transition probability of the target location point corresponding to each of the candidate roads, the target road mapped to the target location point is determined.

[0224] In an exemplary embodiment, the single-point feature information corresponding to the positioning point includes feature information of the positioning point corresponding to each of the at least one candidate road; the single-point feature information corresponding to the positioning point includes at least one of the following:

[0225] The deflection angle of the positioning point relative to the reference direction, and the average deflection angle corresponding to the positioning point, wherein the average deflection angle refers to the average of the deflection angles of at least one positioning point before the positioning point relative to the reference direction.

[0226] The difference between the absolute value of the vertical distance of the positioning point relative to the candidate road and the average vertical distance, wherein the average vertical distance refers to the average of the absolute values ​​of the vertical distances of at least one positioning point before the positioning point relative to the candidate road.

[0227] The absolute value of the vertical distance between the positioning point and the candidate road;

[0228] The vertical distance of the positioning point relative to the candidate road;

[0229] The difference between the absolute value of the vertical distance of the positioning point relative to the candidate road and the positioning accuracy value of the positioning point;

[0230] The positioning accuracy value of the positioning point.

[0231] In an exemplary embodiment, the road network information includes at least one of the following:

[0232] The bifurcation angle of the road forks between the roads;

[0233] The average of the absolute values ​​of the vertical distances between at least two location points at the road forks between the roads;

[0234] The speed characteristics of the road;

[0235] The grade of the road;

[0236] The road includes lane classes and the number of lanes for each lane class;

[0237] The width of the road;

[0238] The lane width information of the road.

[0239] In summary, the technical solution provided in this application obtains the location sequence corresponding to the positioning point, which includes multiple positioning points (the positioning point and its associated positioning points), and obtains the attribute information of each road and the single-point feature information of each positioning point. This allows for the generation of observation parameters for the positioning point by referring to the single-point feature information of the positioning point and its associated positioning points. Based on these observation parameters, the road corresponding to the positioning point is predicted (i.e., mapped to the predicted road). Because the amount of information referenced during the positioning point mapping process is enriched, the generated observation parameters for the positioning point are more accurate, thereby improving the accuracy of trajectory binding.

[0240] Please refer to Figure 17 This diagram illustrates a block diagram of a model training apparatus according to an embodiment of this application. The apparatus has the functionality to implement the model training method example described above; this functionality can be implemented in hardware or by hardware executing corresponding software. The apparatus can be the computer device described above, or it can be mounted on a computer device. The apparatus 1700 may include:

[0241] The sequence acquisition module 1710 is used to acquire a position sequence corresponding to a first positioning point in the sample movement trajectory; wherein, the position sequence includes multiple positioning points located in the sample movement trajectory, and the multiple positioning points include the first positioning point and at least one associated positioning point of the first positioning point.

[0242] The information acquisition module 1720 is used to acquire single-point feature information corresponding to the multiple positioning points based on road network information; wherein, the road network information includes multiple sample roads contained in the road network map and attribute information of each sample road, and the multiple sample roads include positive sample roads and negative sample roads corresponding to the first positioning point.

[0243] The parameter determination module 1730 is used to determine the observation parameters of the first positioning point based on the single-point feature information corresponding to the multiple positioning points through a first network model; wherein, the observation parameters of the first positioning point are used to indicate the probability that the first positioning point is mapped to the multiple sample roads respectively.

[0244] The road determination module 1740 is used to determine the target sample road to which the first positioning point is mapped from the plurality of sample roads based on the observation parameters of the first positioning point.

[0245] The parameter adjustment module 1750 is used to adjust the parameters of the first network model based on the target sample road and the positive sample road and negative sample road corresponding to the first positioning point.

[0246] In an exemplary embodiment, the parameter determination module 1730 is configured to:

[0247] The single-point feature information corresponding to the plurality of positioning points is input into the first network model in chronological order to generate the association information of the first positioning point. The association information of the first positioning point is used to indicate the relationship between the first positioning point and the at least one associated positioning point.

[0248] Based on the association information of the first location point and the road network information, the observation parameters of the first location point are generated using the second network model.

[0249] In an exemplary embodiment, the first network model is a recurrent neural network, which includes a memory unit and a gate controller. The memory unit is used to record information acquired and generated by the first network model, and the gate controller is used to control the retention of information in the first network model. The parameter determination module 1730 is used to process the memory unit, the feature information corresponding to the nth sample location point, and the loop result output by the (n-1)th loop of the training of the first network model in the target round, through the gate controller, to generate the loop result output by the nth loop and update the memory unit, where n is a positive integer greater than 1. The association information of the first location point obtained from the training of the target round is the loop result output by the last loop of the training process of the first network model in the target round.

[0250] In summary, the technical solution provided in this application involves inputting the single-point feature information corresponding to multiple positioning points into the first network model, extracting it in chronological order, and selectively memorizing the information of previous points stored in the first network model in each loop through the gate controller. This allows the first network model to integrate the single-point feature information corresponding to multiple positioning points in the location sequence based on the chronological order of the multiple positioning points, thereby improving the accuracy of the observation parameters of the target positioning points generated by combining road network information, i.e., improving the accuracy of the observation parameters of a single positioning point, and thus improving the accuracy of trajectory binding.

[0251] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0252] Please refer to Figure 18 This diagram illustrates the structural block diagram of a computer device according to an embodiment of this application. The computer device is used to implement the trajectory binding method or model training method provided in the above embodiments. Specifically:

[0253] The computer device 1800 includes a CPU (Central Processing Unit) 1801, a system memory 1804 including RAM (Random Access Memory) 1802 and ROM (Read-Only Memory) 1803, and a system bus 1805 connecting the system memory 1804 and the central processing unit 1801. The computer device 1800 also includes a basic I / O (Input / Output) system 1806 that facilitates information transfer between various components within the computer, and a mass storage device 1807 for storing the operating system 1813, application programs 1814, and other program modules 1815.

[0254] The basic input / output system 1806 includes a display 1808 for displaying information and an input device 1809 for user input, such as a mouse or keyboard. Both the display 1808 and the input device 1809 are connected to the central processing unit 1801 via an input / output controller 1810 connected to the system bus 1805. The basic input / output system 1806 may also include the input / output controller 1810 for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller 1810 also provides output to a display screen, printer, or other types of output devices.

[0255] The mass storage device 1807 is connected to the central processing unit 1801 via a mass storage controller (not shown) connected to the system bus 1805. The mass storage device 1807 and its associated computer-readable media provide non-volatile storage for the computer device 1800. That is, the mass storage device 1807 may include computer-readable media (not shown) such as a hard disk or a CD-ROM (Compact Disc Read-Only Memory) drive.

[0256] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include RAM, ROM, and EPROM (Erasable Programmable ROM).

[0257] Read-Only Memory (EEPROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other solid-state storage, CD-ROM, DVD (Digital Video Disc), or other optical storage, magnetic tape cassette, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will understand that the computer storage media are not limited to the above-mentioned types. The system memory 1804 and mass storage device 1807 described above can be collectively referred to as memory.

[0258] According to various embodiments of this application, the computer device 1800 can also be connected to a remote computer on a network, such as the Internet. That is, the computer device 1800 can be connected to a network 1812 via a network interface unit 1811 connected to the system bus 1805, or the network interface unit 1811 can be used to connect to other types of networks or remote computer systems (not shown).

[0259] In an exemplary embodiment, a computer-readable storage medium is also provided, the storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set, when executed by a processor, implements the above-described trajectory binding method or model training method.

[0260] Optionally, the computer-readable storage medium may include: ROM (Read Only Memory), RAM (Random Access Memory), SSD (Solid State Drives), or optical disc, etc. The random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0261] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. The processor of the terminal reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the terminal to perform the aforementioned trajectory binding method or model training method.

[0262] It is understood that in the specific implementation of this application, user information and other related data are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and maps. For example, the collection of driving data and the acquisition of positioning information involved in this application are all carried out with the user's permission and in compliance with relevant laws, regulations and standards.

[0263] Additionally, "multiple" as used in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0264] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A trajectory binding method, characterized in that, The method includes: For a target location point in the target movement trajectory, obtain the position sequence corresponding to the target location point; wherein, the position sequence includes multiple location points located in the target movement trajectory, and the multiple location points include the target location point and at least one associated location point of the target location point; Based on road network information, single-point feature information corresponding to the multiple positioning points is obtained; wherein, the road network information includes multiple roads contained in the road network map and the attribute information of each road; Based on the single-point feature information corresponding to the plurality of positioning points, the observation parameters of the target positioning point are determined; wherein, the observation parameters of the target positioning point are used to indicate the probability that the target positioning point is mapped to at least one candidate road respectively; Based on the observation parameters of the target location point, the target road mapped to the target location point is determined from the at least one candidate road.

2. The method according to claim 1, characterized in that, The step of determining the observation parameters of the target positioning point based on the single-point feature information corresponding to the plurality of positioning points includes: The single-point feature information corresponding to the plurality of positioning points is input into a first network model with memory function in chronological order to generate the association information of the target positioning point. The association information of the target positioning point is used to indicate the association relationship between the target positioning point and the at least one associated positioning point. Based on the association information of the target location point and the road network information, the observation parameters of the target location point are generated using a second network model.

3. The method according to claim 2, characterized in that, The first network model is a recurrent neural network. The first network model includes a memory unit and a gate controller. The memory unit is used to record the information acquired and generated by the first network model, and the gate controller is used to control the retention of information in the first network model. The step of inputting the single-point feature information corresponding to the plurality of positioning points into a first network model with memory function in chronological order to generate the association information of the target positioning points includes: For the nth iteration of the multiple iterations of the first network model, the gate controller processes the memory unit, the feature information corresponding to the nth location point, and the loop result output by the (n-1)th iteration of the first network model to generate the loop result output by the nth iteration and update the memory unit, where n is a positive integer greater than 1. The associated information is the loop result of the last loop output of the first network model.

4. The method according to claim 3, characterized in that, The gate controller includes at least one of the following: a forget gate, an input gate, and an output gate; Specifically, for the nth iteration, the forget gate is used to determine the forgetting status of the information stored in the memory unit after the (n-1)th iteration; the input gate is used to determine the embedding degree of the single-point feature information of the location point input in the nth iteration in the first network model; and the output gate is used to determine the information used as the loop result output of the nth iteration from the information generated by the first network model during the nth iteration.

5. The method according to claim 1, characterized in that, The associated location point refers to a location point that has a first association relationship with the target location point, and the first association relationship includes at least one of the following: There is no passable road junction between the location point and the target location point; The number of positioning points located before the target positioning point and spaced apart from the target positioning point is less than or equal to a first threshold. The number of positioning points located after the target positioning point and spaced apart from the target positioning point is less than or equal to the second threshold. The positioning point is located before the target positioning point, and the distance between the positioning point and the target positioning point is less than or equal to the third threshold. The positioning point is located after the target positioning point, and the distance between the positioning point and the target positioning point is less than or equal to the fourth threshold.

6. The method according to claim 5, characterized in that, The target location point corresponds to the associated location point of the target candidate road in the at least one candidate road, which is a location point on the target candidate road that has no passable road forks with respect to the target location point.

7. The method according to claim 1, characterized in that, The observation parameters include the observation values ​​of the at least one candidate road; The step of determining the target road mapped to the target location point from the at least one candidate road based on the observation parameters of the target location point includes: Obtain the state transition value of the at least one candidate road, the state transition value being used to indicate the probability of transitioning from the target road corresponding to the previous location point of the target location point to the candidate road; Based on the observation values ​​of the at least one candidate road and the state transition values ​​of the at least one candidate road, the observation probability of the target location point corresponding to each of the candidate roads is determined. The observation probability is used to indicate the possibility that the target location point is mapped to the candidate road within the scope of the global map. The global map includes the road network map. The target road is determined from the at least one candidate road based on the observation probability of the target location point corresponding to each of the candidate roads.

8. The method according to claim 1, characterized in that The method further includes: If the target location point meets the first condition, the step of obtaining the position sequence corresponding to the target location point from the target location point in the target movement trajectory shall be executed. If the target location point meets the second condition, the target road mapped to the target location point is determined by comparing the target location point with the previous location point of the target location point.

9. The method according to claim 8, characterized in that, The first condition includes that the distance between the target location point and the nearest road fork is less than or equal to a fifth threshold.

10. The method according to claim 8, characterized in that, The second condition includes that the distance between the target location point and the nearest road fork is greater than or equal to the sixth threshold. Determining the target road mapped to the target location point by comparing the target location point with the previous location point of the target location point includes: The observation probability of the target location point corresponding to each of the candidate roads is determined by the vertical distance of the target location point relative to each of the candidate roads; Obtain the state transition value of the at least one candidate road, the state transition value being used to indicate the probability of transitioning from the target road corresponding to the previous location point of the target location point to the candidate road; Based on the observation probability of the target location point corresponding to each of the candidate roads, and the state transition probability of the target location point corresponding to each of the candidate roads, the target road mapped to the target location point is determined.

11. The method according to any one of claims 1 to 10, characterized in that, The single-point feature information corresponding to the positioning point includes feature information of the positioning point corresponding to each of the at least one candidate road; the single-point feature information corresponding to the positioning point includes at least one of the following: The deflection angle of the positioning point relative to the reference direction, and the average deflection angle corresponding to the positioning point, wherein the average deflection angle refers to the average of the deflection angles of at least one positioning point before the positioning point relative to the reference direction. The difference between the absolute value of the vertical distance of the positioning point relative to the candidate road and the average vertical distance, wherein the average vertical distance refers to the average of the absolute values ​​of the vertical distances of at least one positioning point before the positioning point relative to the candidate road. The absolute value of the vertical distance between the positioning point and the candidate road; The vertical distance of the positioning point relative to the candidate road; The difference between the absolute value of the vertical distance of the positioning point relative to the candidate road and the positioning accuracy value of the positioning point; The positioning accuracy value of the positioning point.

12. The method according to any one of claims 1 to 10, characterized in that, The road network information includes at least one of the following: The bifurcation angle of the road forks between the roads; The average of the absolute values ​​of the vertical distances between at least two location points at the road forks between the roads; The speed characteristics of the road; The grade of the road; The road includes lane classes and the number of lanes for each lane class; The width of the road; The lane width information of the road.

13. A model training method, characterized in that, The method includes: For a first location point in the sample movement trajectory, obtain the position sequence corresponding to the first location point; wherein, the position sequence includes multiple location points located in the sample movement trajectory, and the multiple location points include the first location point and at least one associated location point of the first location point; Based on road network information, single-point feature information corresponding to the multiple positioning points is obtained; wherein, the road network information includes multiple sample roads contained in the road network map and attribute information of each sample road, and the multiple sample roads include positive sample roads and negative sample roads corresponding to the first positioning point; Based on the single-point feature information corresponding to the multiple positioning points, the observation parameters of the first positioning point are determined through the first network model; wherein, the observation parameters of the first positioning point are used to indicate the probability that the first positioning point is mapped to the multiple sample roads respectively. Based on the observation parameters of the first positioning point, the target sample road mapped to the first positioning point is determined from the multiple sample roads; Based on the target sample road and the positive and negative sample roads corresponding to the first positioning point, the parameters of the first network model are adjusted.

14. The method according to claim 13, characterized in that, The step of determining the observation parameters of the first positioning point based on the single-point feature information corresponding to the plurality of positioning points through the first network model includes: The single-point feature information corresponding to the plurality of positioning points is input into the first network model in chronological order to generate the association information of the first positioning point. The association information of the first positioning point is used to indicate the relationship between the first positioning point and the at least one associated positioning point. Based on the association information of the first location point and the road network information, the observation parameters of the first location point are generated using the second network model.

15. The method according to claim 14, characterized in that, The first network model is a recurrent neural network. The first network model includes a memory unit and a gate controller. The memory unit is used to record the information acquired and generated by the first network model, and the gate controller is used to control the retention of information in the first network model. The step of inputting the single-point feature information corresponding to the plurality of positioning points into the first network model in chronological order to generate the association information of the first positioning points includes: For the nth iteration of the training of the first network model in the target round, the gate controller processes the memory unit, the feature information corresponding to the nth sample location point, and the loop result output of the (n-1)th iteration of the training of the target round to generate the loop result output of the nth iteration and update the memory unit, where n is a positive integer greater than 1. The association information of the first localization point obtained from the training of the target round is the loop result output by the first network model in the last loop of the training process of the target round.

16. A trajectory binding device, characterized in that, The device includes: The sequence acquisition module is used to acquire a position sequence corresponding to a target positioning point in a target movement trajectory; wherein, the position sequence includes multiple positioning points located in the target movement trajectory, and the multiple positioning points include the target positioning point and at least one associated positioning point of the target positioning point; The information acquisition module is used to acquire single-point feature information corresponding to the multiple positioning points based on road network information; wherein, the road network information includes multiple roads contained in the road network map and the attribute information of each road; The parameter determination module is used to determine the observation parameters of the target positioning point based on the single-point feature information corresponding to the plurality of positioning points respectively; wherein, the observation parameters of the target positioning point are used to indicate the probability that the target positioning point is mapped to at least one candidate road respectively; The road determination module is used to determine the target road mapped to the target positioning point from the at least one candidate road based on the observation parameters of the target positioning point.

17. A model training device, characterized in that, The device includes: The sequence acquisition module is used to acquire a position sequence corresponding to a first positioning point in the sample movement trajectory; wherein, the position sequence includes multiple positioning points located in the sample movement trajectory, and the multiple positioning points include the first positioning point and at least one associated positioning point of the first positioning point. The information acquisition module is used to acquire single-point feature information corresponding to the multiple positioning points based on road network information; wherein, the road network information includes multiple sample roads contained in the road network map and attribute information of each sample road, and the multiple sample roads include positive sample roads and negative sample roads corresponding to the first positioning point; The parameter determination module is used to determine the observation parameters of the first positioning point based on the single-point feature information corresponding to the multiple positioning points, through a first network model; wherein, the observation parameters of the first positioning point are used to indicate the probability that the first positioning point is mapped to the multiple sample roads respectively. The road determination module is used to determine the target sample road to which the first positioning point is mapped from the plurality of sample roads based on the observation parameters of the first positioning point; The parameter adjustment module is used to adjust the parameters of the first network model based on the target sample road and the positive and negative sample roads corresponding to the first positioning point.

18. A computer device, characterized in that, The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the trajectory binding method according to any one of claims 1 to 12, or to implement the model training method according to any one of claims 13 to 15.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the trajectory binding method according to any one of claims 1 to 12, or to implement the model training method according to any one of claims 13 to 15.

20. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium. The processor reads and executes the computer instructions from the computer-readable storage medium to implement the trajectory binding method according to any one of claims 1 to 12, or the model training method according to any one of claims 13 to 15.

Images