A trajectory classification method, a model training method, a device, an apparatus, and a medium

Abstract

The present disclosure provides a trajectory classification method and device, a model training method and device, and a medium, relates to the technical field of data processing, and in particular to the technical fields of big data, intelligent transportation, high-precision maps, and the like. The specific implementation scheme is as follows: based on target trajectory data to be analyzed, a specified behavior feature of a user to whom the target trajectory data belongs about a waiting-for-a-red-light behavior is generated; based on the specified behavior feature and a trajectory point feature corresponding to the target trajectory data, model input content corresponding to the target trajectory data is determined; the determined model input content is input into a pre-trained trajectory classification model to obtain a trajectory classification result of the target trajectory data; wherein the trajectory classification model is a neural network model trained based on model input content corresponding to sample trajectory data and a label used to represent a trajectory type. It can be seen that, by the present scheme, the accuracy of trajectory classification can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, particularly to the fields of big data, intelligent transportation, and high-precision maps, and specifically to a trajectory classification method, a model training method, a device, an equipment, and a medium. Background Technology

[0002] In the field of intelligent transportation, trajectory classification is used to identify whether the trajectory represented by trajectory data belongs to driving, walking, or cycling, etc. Trajectory data is a sequence of trajectory points with both location and time information.

[0003] In related technologies, trajectory velocity, acceleration, and displacement information are often analyzed from trajectory data and used as input to train an initial trajectory classification model, resulting in a trajectory classification model with a certain classification accuracy, which is then used to classify trajectory data. Summary of the Invention

[0004] This disclosure provides a trajectory classification method, a model training method, an apparatus, a device, and a medium.

[0005] According to one aspect of this disclosure, a trajectory classification method is provided, comprising:

[0006] Based on the target trajectory data to be analyzed, generate specified behavioral features of the user to whom the target trajectory data belongs regarding the behavior of waiting at traffic lights;

[0007] Based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data, the model input content corresponding to the target trajectory data is determined;

[0008] The determined model input content is input into the pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data;

[0009] The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

[0010] According to another aspect of this disclosure, a model training method is provided, comprising:

[0011] Based on the sample trajectory data, generate sample behavioral features of the sample user to which the sample trajectory data belongs, regarding the behavior of waiting at the traffic lights;

[0012] Based on the sample behavior features and the trajectory point features corresponding to the sample trajectory data, the model input content corresponding to the sample trajectory data is determined;

[0013] Input the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, and obtain the trajectory classification result of the sample trajectory data;

[0014] Based on the trajectory classification results of the sample trajectory data and the labels corresponding to the sample trajectory data used to characterize the trajectory type, the model parameters of the trajectory classification model are adjusted to train the trajectory classification model.

[0015] According to another aspect of this disclosure, a trajectory classification apparatus is provided, comprising:

[0016] The generation module is used to generate specified behavioral features of the user to which the target trajectory data belongs, based on the target trajectory data to be analyzed, regarding the behavior of waiting at traffic lights.

[0017] The determination module is used to determine the model input content corresponding to the target trajectory data based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data;

[0018] The classification module is used to input the determined model input content into a pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data;

[0019] The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

[0020] According to another aspect of this disclosure, a model training apparatus is provided, comprising:

[0021] The sample generation module is used to generate sample behavioral features of the sample user to which the sample trajectory data belongs, based on the sample trajectory data, regarding the waiting behavior.

[0022] The sample determination module is used to determine the model input content corresponding to the sample trajectory data based on the sample behavior features and the trajectory point features corresponding to the sample trajectory data.

[0023] The sample classification module is used to input the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, and obtain the trajectory classification result of the sample trajectory data.

[0024] An adjustment module is used to adjust the model parameters of the trajectory classification model based on the trajectory classification results of the sample trajectory data and the labels corresponding to the sample trajectory data that characterize the trajectory type, in order to train the trajectory classification model.

[0025] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the trajectory classification method or the model training method described above.

[0026] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform a trajectory classification method or a model training method according to any of the preceding claims.

[0027] According to another aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the trajectory classification method or the model training method according to any of the preceding claims.

[0028] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0029] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0030] Figure 1 This is a flowchart of a trajectory classification method according to the present disclosure;

[0031] Figure 2 This is a flowchart of step A1 in the trajectory classification method of this disclosure;

[0032] Figure 3 This is a flowchart of a model training method according to the present disclosure.

[0033] Figure 4 This is a schematic diagram of trajectory data according to this disclosure;

[0034] Figure 5 This is a schematic diagram illustrating a feature for determining traffic light behavior using stop line position data according to this disclosure;

[0035] Figure 6 This is a schematic diagram of a road segment sequence related to an intersection, according to this disclosure;

[0036] Figure 7 This is a flowchart of a method for extracting features of lighting behavior according to the present disclosure;

[0037] Figure 8This is a schematic diagram of the structure of a trajectory classification device according to the present disclosure;

[0038] Figure 9 This is a schematic diagram of the structure of a model training device according to the present disclosure;

[0039] Figure 10 This is a block diagram of an electronic device used to implement the methods provided in the embodiments of this disclosure. Detailed Implementation

[0040] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0041] In related technologies, trajectory data such as velocity, acceleration, and displacement are often analyzed and used as input to train an initial trajectory classification model, resulting in a trajectory classification model with a certain level of accuracy for classifying trajectory data. However, trajectory classification models trained using this information often struggle to effectively distinguish between low-speed driving trajectories and high-speed cycling trajectories, making it difficult to improve the accuracy of trajectory classification.

[0042] Based on the above, in order to improve the accuracy of trajectory classification, this disclosure provides a trajectory classification method, a model training method, an apparatus, a device, and a medium.

[0043] Below, we will first introduce a trajectory classification method provided by an embodiment of this disclosure.

[0044] The trajectory classification method provided in this disclosure can be applied to various electronic devices, such as personal computers, servers, and other devices with data processing capabilities. Furthermore, it is understood that the trajectory classification method provided in this disclosure can be implemented through software, hardware, or a combination of both.

[0045] The trajectory classification method provided in this embodiment may include the following steps:

[0046] Based on the target trajectory data to be analyzed, generate specified behavioral features of the user to whom the target trajectory data belongs regarding the behavior of waiting at traffic lights;

[0047] Based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data, the model input content corresponding to the target trajectory data is determined;

[0048] The determined model input content is input into the pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data;

[0049] The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

[0050] In the solution provided in this disclosure, different types of trajectories have different characteristics related to the behavior of waiting at traffic lights. The trajectory classification model is trained based on the model input content corresponding to the sample trajectory data and labels used to represent the trajectory type. That is, the training process of the trajectory classification model incorporates features related to the behavior of waiting at traffic lights. Therefore, compared with existing trajectory classification models trained using trajectory point features, the trajectory classification model trained in this solution has higher classification accuracy. Thus, by determining the model input content corresponding to the target trajectory data and inputting this model input content into the pre-trained trajectory classification model, the trajectory classification model can classify the model input content based on the knowledge learned during model training, thereby obtaining a more accurate trajectory classification result. It is evident that this solution can improve the accuracy of trajectory classification.

[0051] The trajectory classification method provided in the embodiments of this disclosure will now be described in conjunction with the accompanying drawings.

[0052] like Figure 1 As shown, the trajectory classification method provided in this embodiment may include the following steps:

[0053] S101, Based on the target trajectory data to be analyzed, generate the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior;

[0054] In this embodiment, the target trajectory data to be analyzed can be trajectory data pre-stored in a specified storage medium, or trajectory data collected in real time; both are reasonable. In practical applications, the target trajectory data can be a sequence of GPS location points periodically reported by the user's terminal device; in this case, the GPS location points are the trajectory points. The terminal device can be a mobile phone, a wearable electronic device, an in-vehicle terminal of the vehicle being driven or ridden, etc. It should be noted that the method of obtaining the target trajectory data in this embodiment is not limited.

[0055] Among these, the behavioral features related to a user waiting at a red light at an intersection are specific behavioral features concerning the waiting behavior. For example, these specific behavioral features may include features such as the waiting distance and the number of times the user waits at a red light. It is understood that, since each trajectory point in the target trajectory data has both location and time information, specific behavioral features concerning the waiting behavior of the user to whom the target trajectory data belongs can be generated based on the target trajectory data.

[0056] Optionally, in one implementation, generating specified behavioral characteristics of the user to whom the target trajectory data belongs regarding their waiting-at-the-light behavior based on the target trajectory data to be analyzed may include steps A1-A2:

[0057] A1. Based on the target trajectory data to be analyzed, identify the target stopping information of the user to which the target trajectory data belongs on the target road segment sequence; wherein, the target road segment sequence is the sequence of road segments related to intersections through which the trajectory represented by the target trajectory data passes, and the target stopping information is stopping information that varies for different modes of transportation;

[0058] In road network data, a sequence of road segments related to an intersection can exist on the road connecting to that intersection. This sequence consists of multiple consecutive road segments, some of which connect to the intersection. In practical applications, the sequence of road segments related to an intersection can be a sequence of road segments preceding the intersection. Different road segments may have different road widths or speed limits, etc. For example,... Figure 6 As shown in the figure, there is a road segment sequence consisting of three road segments labeled link1, link2 and link3. If link1 is a road segment connected to an intersection, then the road segment sequence consisting of the road segments labeled link1, link2 and link3 is a road segment sequence related to the intersection.

[0059] In this implementation, since the user's behavior of waiting for the red light to end before crossing the intersection is the waiting-for-the-light behavior, and different modes of transportation have different characteristics of the waiting-for-the-light behavior, for example, the waiting distance for a user at an intersection is longer when driving than when cycling; therefore, the user's target stopping information on the target road segment sequence can be identified first based on the target trajectory data. The target road segment sequence is the sequence of road segments related to the intersection that the trajectory represented by the target trajectory data passes through.

[0060] For example, road network data can be used to match target trajectory data to obtain various road segments that match the trajectory points in the target trajectory data, i.e., the road segments where each trajectory point is located. This determines the sequence of road segments related to intersections within each segment, which serves as the target road segment sequence. Then, the trajectory points on the target road segment sequence are analyzed to obtain the user's target stopping information on the target road segment sequence. The road network data may include the road segment identifiers and location information of each road segment. It should be noted that the location information of each road segment in the road network data typically does not include the width direction location information of each road segment; that is, the location information of the centerline of each road segment is used as the location information of that road segment. Therefore, a trajectory point matching any road segment can be a trajectory point whose perpendicular distance from the centerline of the road segment is within a preset distance range along the length direction of that road segment.

[0061] Additionally, it should be noted that during the determination of the target road segment sequence, it is also possible to determine whether any trajectory points in the target trajectory data pass through the intersection corresponding to the road segment sequence. If so, the trajectory represented by the target trajectory data passes through the intersection corresponding to that road segment sequence, and this road segment sequence can be identified as the target road segment sequence. If not, the user to whom the target trajectory data belongs has not subsequently passed through the intersection corresponding to that road segment sequence. In this case, the user's stopping behavior on that road segment sequence can be considered unrelated to waiting at traffic lights, and therefore, this road segment sequence can be excluded from being considered a target road segment sequence. This ensures that the target stopping information subsequently identified on the target road segment sequence is information related to the user's waiting at traffic lights.

[0062] For example, in one specific implementation, the target stopping information may include: the stopping distance between the user's stopping location and the target stop line, and / or the number of stops; wherein the target stop line is the intersection stop line on the target road segment sequence.

[0063] In this embodiment, the user's stopping position can be obtained by analyzing the stopping points in the target trajectory data. The target stop line position can be obtained by collecting stop line positions at various intersections using a data collection vehicle, or by analyzing a large amount of historical trajectory data; both are reasonable methods. It should be noted that the methods for determining the stopping distance and the number of stops are described in the following embodiments and will not be repeated here. Furthermore, it should be emphasized that, in one exemplary implementation, the method for determining the intersection stop line on any intersection-related road segment sequence can refer to the method for determining the sample stop line in the following embodiments: The intersection-related road segment sequence is determined as the road segment sequence to be analyzed; multiple stopping points on the road segment sequence to be analyzed are obtained based on historical trajectory data passing through the road segment sequence; the distance from each stopping point to a preset endpoint of the road segment sequence to be analyzed is calculated to obtain multiple queuing distances; these multiple queuing distances are analyzed to obtain the position of the stop line on the road segment sequence to be analyzed.

[0064] It is understandable that different modes of transportation have different stopping distances and number of stops. For example, driving often requires more stopping distances and more stopping times than cycling. Therefore, stopping distance and / or the number of stops can be used as target stopping information.

[0065] It should be noted that, for the sake of clarity in the scheme layout, the method for identifying target parking information will be introduced in the following embodiments, and will not be repeated here.

[0066] A2, based on the identified target stopping information, construct the specified behavioral features of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior.

[0067] For example, in practical applications, the target stopping information can be directly used as a specified behavioral feature, or the target stopping information can be mapped to different preset intervals, and the number of times the target stopping information is mapped to different intervals can be counted. The number of times the target stopping information is mapped to different intervals can then be used as a specified behavioral feature. Both of these are reasonable.

[0068] S102, Based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data, determine the model input content corresponding to the target trajectory data;

[0069] In this embodiment, the trajectory point features corresponding to the target trajectory data may include features such as trajectory velocity, acceleration, and displacement analyzed based on each trajectory point. It is understood that since different modes of transportation have different speed, acceleration, and displacement features, and different modes of transportation have different specified behavioral features regarding waiting at traffic lights, the model input content can be determined based on the trajectory point features corresponding to the target trajectory data and the specified behavioral features.

[0070] For example, in practical applications, specified behavioral features and trajectory point features can be combined as model input. For instance, specified behavioral features and trajectory point features can be constructed into a multi-dimensional vector and used as model input.

[0071] S103, Input the determined model input content into the pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data;

[0072] The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

[0073] In this embodiment, the label representing the trajectory type is used to characterize the ground truth of the trajectory type to which the sample trajectory data belongs. It is understood that since this trajectory classification model is trained based on the model input content and labels corresponding to the sample trajectory data, and the model input content is determined based on the features of the sample trajectory data regarding waiting at traffic lights and the features of trajectory points, and different traffic modes have different features regarding waiting at traffic lights, the trajectory classification model trained in this scheme has higher classification accuracy compared to existing trajectory classification models trained using trajectory point features. Therefore, inputting the determined model input content into the pre-trained trajectory classification model can yield more accurate trajectory classification results, thereby improving the accuracy of trajectory classification.

[0074] It should be noted that, for the sake of clarity in the solution layout, the training method of the trajectory classification model will be introduced in the following examples of model training methods, and will not be repeated here.

[0075] In the solution provided in this disclosure, different types of trajectories have different characteristics related to the behavior of waiting at traffic lights. The trajectory classification model is trained based on the model input content corresponding to the sample trajectory data and labels used to represent the trajectory type. That is, the training process of the trajectory classification model incorporates features related to the behavior of waiting at traffic lights. Therefore, compared with existing trajectory classification models trained using trajectory point features, the trajectory classification model trained in this solution has higher classification accuracy. Thus, by determining the model input content corresponding to the target trajectory data and inputting this model input content into the pre-trained trajectory classification model, the trajectory classification model can classify the model input content based on the knowledge learned during model training, thereby obtaining a more accurate trajectory classification result. It is evident that this solution can improve the accuracy of trajectory classification.

[0076] Alternatively, in another embodiment of this disclosure, in Figure 1 Based on the illustrated embodiments, as Figure 2 As shown, step A1 above, based on the target trajectory data to be analyzed, identifies the target stopping information of the user to which the target trajectory data belongs on the target road segment sequence, which may include steps S201-S202:

[0077] S201, From the trajectory points included in the target trajectory data, determine the target stopping point on the target road segment sequence;

[0078] Understandably, since analyzing the stopping points in the target trajectory data can yield stopping information related to the user's waiting behavior at traffic lights, in order to identify the user's target stopping information on the target road segment sequence, the target stopping points on the target road segment sequence can be determined from the trajectory points included in the target trajectory data.

[0079] Optionally, in one implementation, determining the target stopping point on the target road segment sequence from the trajectory points included in the target trajectory data may include steps B1-B2:

[0080] B1, determine the stopping point among the trajectory points included in the target trajectory data;

[0081] It is understandable that each trajectory point has time information and location information. The stopping points in the trajectory points can be various trajectory points that are in the same location but whose time difference exceeds a preset threshold.

[0082] For example, the method for determining the stopping point in the target trajectory data can be as follows: For each trajectory point in the target trajectory data, identify whether the trajectory point meets preset conditions; wherein, the preset conditions include: a first condition that the distance to a preset number of trajectory points preceding the trajectory point meets a preset distance range, and a second condition that the time difference to a preset number of target trajectory points preceding the trajectory point meets a preset time range; and the trajectory point that meets the preset conditions is determined as the stopping point. For example, the preset distance range can be 5 meters, 10 meters, etc., and the preset time difference can be 5 seconds, 10 seconds, etc. The preset number of trajectory points preceding each trajectory point are the preset number of trajectory points found sequentially before the trajectory point in terms of time, based on the time of the trajectory point.

[0083] B2, based on matching the location information of the stopping point with the road network data, identifies the target stopping point on the target road segment sequence; wherein, the road network data contains the location information of the road segment sequence at each intersection.

[0084] In this implementation, after determining each stopping point in the target trajectory data, the location information of the stopping points can be matched with road network data. The road network data contains the location information of road segment sequences at each intersection. After matching, stopping points located on the road segment sequences at each intersection can be obtained. Then, the stopping points on the road segment sequences at each intersection are identified as the target stopping points.

[0085] It should be noted that the method for determining the target stopping points on the target road segment sequence described above is merely an example. In practical applications, it is also reasonable to first match the road network data with the target trajectory data to determine the trajectory points on each target road segment sequence, and then analyze the stopping points among the trajectory points on the target road segment sequence to obtain the target stopping points.

[0086] S202, Based on the determined target stopping point, identify the target stopping information of the user to whom the target trajectory data belongs on the target road segment sequence.

[0087] Understandably, since the location of the target stopping point can be used to analyze the user's stopping position while waiting at a traffic light, and the user's stopping position can be used to determine information such as the stopping distance and the number of stops, it is possible to analyze the user's target stopping information on the target road segment sequence through the target stopping point.

[0088] Optionally, in one implementation, the method for identifying the stopping distance may include steps C1-C3:

[0089] C1, based on the determined target stopping point, determine the user's stopping position on the target road segment sequence;

[0090] For example, in practical applications, the location of the target stopping point can be directly determined as the stopping position of the user on the target road segment sequence; or, the location of any one of the multiple target stopping points within a preset distance range can be determined as the stopping position corresponding to the multiple target stopping points, which is reasonable.

[0091] For example, in one specific implementation, determining the user's stopping position on the target road segment sequence based on the determined target stopping point in step C1 may include:

[0092] C11, perform cluster analysis on the identified target stopping points to obtain the target cluster point set;

[0093] C12 determines the location of any stopping point in the target cluster as the user's stopping location on the target road segment sequence.

[0094] Understandably, when a user stops at a certain location, the reported GPS location will have a slight offset around that location. Therefore, that location can correspond to multiple target stopping points. If each target stopping point's location is used as the user's actual stopping location, the subsequent calculation of the stopping distance becomes computationally intensive. Therefore, to reduce computational load, we can first perform cluster analysis on the target stopping points to obtain a target cluster set. Then, the location of any stopping point within this target cluster set can be used to determine the user's actual stopping location.

[0095] For example, clustering algorithms such as k-means clustering and FCM (Fuzzy C-Means) clustering can be used to perform cluster analysis on the target stopping points to obtain the target cluster point set.

[0096] C2, calculate the first distance between the stopping position and the preset endpoint of the target road segment, and the second distance between the preset endpoint and the target stop line; wherein, the target road segment is the road segment in which the stopping position is located among the road segments included in the target road segment sequence;

[0097] C3, based on the first distance and the second distance, determines the stopping distance.

[0098] In this embodiment, the preset endpoint of the target road segment can be either the beginning or the end of the target road segment, which is reasonable. It is understandable that since the road network data contains the location information of each road segment, that is, the location of the beginning or end of each road segment, the road network data can be used to calculate the first distance between the stopping position and the preset endpoint of the target road segment, and the second distance between the preset endpoint and the target stop line.

[0099] It's understandable that the location of each road segment in the road network data represents the position of its centerline. However, in practical applications, a user's stopping position can be at a predetermined perpendicular distance from the centerline, or there may be diagonal road segments within the road segment sequence at an intersection. In these cases, directly calculating the distance between the user's stopping position and the target stop line will result in a significant error compared to the actual stopping distance. Therefore, to obtain a more accurate stopping distance, we can calculate a first distance between the stopping position and a predetermined endpoint of the target road segment, and a second distance between the predetermined endpoint and the target stop line. Based on these first and second distances, the stopping distance can then be determined.

[0100] For example, if the preset endpoint of the target road segment is the starting point, then the difference between the first distance and the second distance is the stopping distance. If the preset endpoint of the target road segment is the ending point, then the sum of the first distance and the second distance is the stopping distance. For example, as... Figure 6As shown, if the road segment where the stopping position is located is link3, the first distance between the stopping position and the first point of link3, and the second distance between the first point of the road segment and the target stop line can be calculated, and the difference between the second distance and the first distance can be determined as the stopping distance; or, the first distance between the stopping position and the last point of link3, and the second distance between the last point of the road segment and the target stop line can be calculated, and the sum of the second distance and the first distance can be determined as the stopping distance.

[0101] It should be noted that the above-described method for identifying the stopping distance is merely an example and should not be construed as limiting this disclosure. For instance, in practical applications, the distance between the user's stopping position and the target stopping line can also be directly calculated as the stopping distance, which is also reasonable.

[0102] Optionally, in one implementation, the method for identifying the number of pauses may include steps D1-D2:

[0103] D1. Perform cluster analysis on the target's stopping point to obtain the target's cluster point set;

[0104] It should be noted that step D1 can be implemented in the same way as step C11, and will not be repeated here.

[0105] D2 determines the number of stops based on the number of points in the target cluster.

[0106] It is understandable that, since the target cluster point set is a set of cluster points generated when a user stops on the target road segment sequence, and different target cluster point sets correspond to different stopping positions, the number of the target cluster point set can be counted as the number of stops.

[0107] Furthermore, it's worth mentioning that after obtaining the target cluster point set through cluster analysis in steps C11 and D1, the target cluster point set can be filtered. This filtered set can then be used to determine the stopping location and number of stops more accurately. Specifically, if the perpendicular distance between the location of a target stopping point in any cluster point set and the centerline of the matched road segment exceeds a preset distance threshold, the target stopping point is considered a non-road stopping point, and the target cluster point set is deleted. Alternatively, if no trajectory point belonging to any cluster point set passes through the intersection corresponding to the target road segment sequence in the subsequent time frame, then the user's stopping behavior on the target road segment sequence is unrelated to waiting at traffic lights, and therefore, the target cluster point set can be deleted.

[0108] As can be seen, this scheme can identify the target stopping information of the user to whom the target trajectory data belongs on the target road segment sequence.

[0109] Alternatively, in another embodiment of this disclosure, in Figure 1 Based on the illustrated embodiment, step A2 above, which involves constructing specific behavioral characteristics of the user to whom the target trajectory data belongs, regarding their waiting-at-the-light behavior, based on the identified target stopping information, may include:

[0110] For each type of information identified in the stopping distance and / or number of stops, determine the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior.

[0111] Understandably, since different modes of transportation correspond to different stopping distances and number of stops, it is possible to determine the specific behavioral characteristics of the user to whom the target trajectory data belongs regarding their waiting-at-light behavior for each type of information in the identified stopping distance and / or number of stops.

[0112] Optionally, in one implementation, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior, for the identified stopping distance, may include steps E1-E2:

[0113] E1 determines the number of times the identified stopping distance hits each preset distance interval; wherein, each distance interval is a different stopping distance range set for a specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit.

[0114] In practical applications, the preset distance intervals can be set by relevant personnel based on experience. For example, the specified trajectory type can be driving or cycling, etc. If the specified trajectory type is driving, statistical analysis of historical trajectory data reveals that the stopping distance during cycling is often shorter than that during driving, and when the stopping distance is greater than 40 meters, there is a high probability of driving. When the stopping distance is between 30 and 40 meters, both driving and cycling are possible. However, if the user's stopping distance at traffic lights is repeatedly between 30 and 40 meters, the probability of driving is higher. Therefore, the number of times the stopping distance hits each distance interval can be counted, and the specified behavioral characteristics can be determined based on the number of times each distance interval is hit. In this case, the preset distance intervals can include the distance interval [30, 40] and the distance interval [40, +∞). The probability of belonging to the driving type when the distance interval [40, +∞) is hit is higher than the probability of belonging to the driving type when the distance interval [30, 40] is hit.

[0115] Understandably, in practical applications, the trajectory represented by the target trajectory data can pass through multiple intersections, that is, there are multiple waiting behaviors at multiple intersections. Therefore, it is possible to determine the number of times the identified stopping distance hits each preset distance interval. The more times, the higher the probability that it belongs to the specified trajectory type.

[0116] E2 determines the number of hits in each preset distance interval as the specified behavioral characteristic of the user to whom the target trajectory data belongs regarding the behavior of waiting at traffic lights.

[0117] For example, in practical applications, the number of hits in each preset distance interval can be constructed into a multi-dimensional vector, which serves as a specified behavioral feature of the user to whom the target trajectory data belongs regarding their behavior at the waiting lights.

[0118] Optionally, in one implementation, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior, based on the identified number of stops, may include steps F1-F2:

[0119] F1 determines the number of target road segment sequences with more than a preset threshold number of stops based on the number of stops identified.

[0120] F2 determines the identified quantity as the specified behavioral characteristic of the user to whom the target trajectory data belongs, regarding the behavior of waiting at traffic lights.

[0121] For example, in practical applications, the preset number of times threshold can be set by relevant personnel based on experience. The preset number of times threshold can be 1 time, 2 times, etc.

[0122] Understandably, since the number of times a driver stops at a red light at the same intersection is often higher than the number of times a cyclist stops at a red light, the number of target road segment sequences with a number of stops exceeding a preset threshold can be counted. Based on this determined number, specific behavioral characteristics related to the red light stopping behavior can be identified. For example, if there are 5 target road segment sequences, and the target number of stops on each sequence is 0, 1, 1, 3, and 2 respectively, and the preset threshold is 1, then the number of target road segment sequences with a number of stops exceeding the preset threshold is 2.

[0123] It is understandable that the more target road segment sequences that stop more than a preset threshold, the greater the probability that the trajectory data corresponding to that target trajectory is a driving type. Therefore, the determined number can be used as a specified behavioral feature for waiting at traffic lights.

[0124] As can be seen, this solution can be used to construct specific behavioral characteristics of the user to whom the target trajectory data belongs, regarding their behavior at traffic lights.

[0125] Corresponding to the above method embodiments, this disclosure also provides a model training method, such as... Figure 3 As shown, it includes:

[0126] S301, Based on the sample trajectory data, generate sample behavior features of the sample user to which the sample trajectory data belongs, regarding the behavior of waiting at the traffic lights;

[0127] S302, Based on the sample behavior features and the trajectory point features corresponding to the sample trajectory data, determine the model input content corresponding to the sample trajectory data;

[0128] The implementation methods of steps S301 and S302 can be found in the relevant content of steps S101-S102 above, and will not be repeated here.

[0129] S303, input the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, and obtain the trajectory classification result of the sample trajectory data;

[0130] In this embodiment, the trajectory classification model to be trained can be a classification model based on neural networks such as RNN (Recurrent Neural Network) and CNN (Convolutional Neural Network).

[0131] S304, Based on the trajectory classification results of the sample trajectory data and the labels corresponding to the sample trajectory data used to characterize the trajectory type, the model parameters of the trajectory classification model are adjusted to train the trajectory classification model.

[0132] In this embodiment, the training method of the trajectory classification model can be as follows: For each sample trajectory data, calculate the difference between the trajectory classification result of the sample trajectory data and the label corresponding to the sample trajectory data used to characterize the trajectory type, and use the sum of the differences corresponding to each sample trajectory data as the model loss value. The difference can be calculated by calculating L1 distance or Euclidean distance, etc. Then, determine whether the trajectory classification model has converged based on whether the model loss value is less than a preset threshold. If converged, training ends; otherwise, adjust the model parameters of the trajectory classification model by backpropagation to minimize the model loss value, and return to the step of inputting the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, until the model converges and training ends.

[0133] In the solution provided in this disclosure, the trajectory classification model is trained based on the model input content and labels corresponding to the sample trajectory data. The model input content corresponding to the sample trajectory data is determined based on the sample behavior features and trajectory point features of the sample trajectory data regarding waiting at traffic lights, and different traffic modes have different features regarding waiting at traffic lights. Therefore, compared with the trajectory classification model trained using trajectory point features in the prior art, by adding sample behavior features related to waiting at traffic lights to train the trajectory classification model, the problem of not being able to effectively distinguish between low-speed driving and high-speed cycling trajectories when training using trajectory point features can be solved. This is because the trajectories of low-speed driving and high-speed cycling have similar speed, displacement, and other features. It is evident that through this solution, by introducing sample behavior features related to waiting at traffic lights to train the model, the trained trajectory classification model can effectively distinguish between low-speed driving and high-speed cycling trajectories, thereby improving the classification accuracy of the trajectory classification model.

[0134] Optionally, generating sample behavioral features of the sample user to whom the sample trajectory data belongs, regarding their waiting-at-the-light behavior, based on the sample trajectory data, includes:

[0135] Based on the sample trajectory data, identify the sample stopping information of the sample user to which the sample trajectory data belongs on the sample road segment sequence; wherein, the sample road segment sequence is the sequence of road segments related to intersections through which the trajectory represented by the sample trajectory data passes, and the sample stopping information is stopping information that varies for different modes of transportation;

[0136] Based on the identified sample parking information, sample behavior features of the sample users regarding their waiting behavior are constructed.

[0137] Optionally, the sample stopping information includes: the sample stopping distance between the sample user's sample stopping position and the sample stopping line, and / or the number of sample stopping times; wherein, the sample stopping line is the intersection stopping line on the sample road segment sequence.

[0138] Optionally, in one implementation, the method for determining the position of the sample stop line may include steps G1-G3:

[0139] G1, obtain multiple sample stopping points on the sample road segment sequence determined based on the trajectory points included in the sample trajectory data;

[0140] In this implementation, the method for determining the stopping point of each sample can be similar to the method for determining the stopping point of each target in step S201 above, and will not be repeated here.

[0141] G2, calculate the distance from each sample stopping point to the preset endpoint of the sample road segment sequence to obtain multiple queuing distances;

[0142] In this implementation, the preset endpoint of the sample road segment sequence can be either the first or last point of the sample road segment sequence, which is reasonable. For example, in practical applications, if the last point of the sample road segment sequence is at an intersection, the distance from the sample stopping point to the last point of the sample road segment sequence can be calculated to obtain multiple queuing distances.

[0143] G3 analyzes the multiple queuing distances to obtain the position of the sample stop line.

[0144] Understandably, based on the knowledge that vehicles queue orderly behind the stop line when waiting at intersections, and the assumption that the sample stopping points follow an exponential distribution—meaning that vehicles are more densely packed closer to the intersection—the location of the sample stop line can be determined by analyzing the distribution of sample stopping points, i.e., analyzing multiple queuing distances.

[0145] Optionally, analyzing the multiple queuing distances to obtain the position of the sample stop line may include steps G31-G33:

[0146] G31, sort the multiple queuing distances according to their size to obtain the target sequence;

[0147] G32, determine the queuing distance in the target sequence that is at a preset quantile, and obtain the distance to be utilized;

[0148] G33 determines the position of the sample stop line as the distance from the preset endpoint of the sample road segment sequence to the distance to be utilized.

[0149] It is understandable that, since the stopping points of each sample stop point follow an exponential distribution when waiting at the same intersection, and the queuing distances corresponding to each sample stop point follow a Poisson distribution, the queuing distances can be sorted according to their magnitude to obtain a target sequence. Then, the queuing distances in the target sequence that fall within a preset quantile are taken as the distances to be utilized. The position of the sample stop line is determined by the distance to be utilized from the preset endpoint of the sample road segment sequence. For example, in practical applications, this preset quantile could be the 95th quantile, the 90th quantile, etc.

[0150] It should be noted that, to obtain more accurate sample stop line positions, queuing distances at multiple preset quantiles can be obtained, and then these queuing distances can be averaged. For example, averaging the queuing distances at the 94th, 95th, and 96th quantiles yields the usable distance, thereby reducing errors. Furthermore, after obtaining the sample stop line positions, during subsequent trajectory classification, the sample stop line positions at each intersection obtained from the analysis of the sample trajectory data can be directly used as the target stop line positions for that intersection.

[0151] As can be seen, this solution can determine the location of the sample stop line by analyzing the sample stopping points, which is more cost-effective than collecting the stop line locations at intersections by using a data collection vehicle.

[0152] Optionally, identifying the sample stop information of the sample user to which the sample trajectory data belongs on the sample road segment sequence based on the sample trajectory data includes:

[0153] From the trajectory points included in the sample trajectory data, determine the sample stopping points on the sample road segment sequence;

[0154] Based on the determined sample stopping points, identify the sample stopping information of the user to whom the sample trajectory data belongs on the sample road segment sequence.

[0155] Optionally, the method for identifying the sample dwell distance includes:

[0156] Based on the determined sample stopping points, determine the sample stopping positions of the sample users on the sample road segment sequence;

[0157] Calculate the first sample distance between the sample stopping position and the preset endpoint of the sample road segment, and the second sample distance between the preset endpoint and the sample stop line; wherein, the sample road segment is the road segment in which the sample stopping position is located among the road segments included in the sample road segment sequence;

[0158] The stopping distance is determined based on the first sample distance and the second sample distance.

[0159] Optionally, determining the sample user's sample stopping position on the sample road segment sequence based on the determined sample stopping point includes:

[0160] Cluster analysis is performed on the identified sample stopping points to obtain the sample cluster point set;

[0161] The position of any stopping point in the sample cluster set is determined as the stopping position of the sample user on the sample segment sequence.

[0162] Optionally, the method for identifying the number of times the sample stops includes:

[0163] Cluster analysis was performed on the stopping points of the samples to obtain the sample cluster point set;

[0164] The number of times a sample stops is determined based on the number of cluster points in the sample cluster.

[0165] Optionally, the step of constructing sample behavior features of the sample user regarding their waiting behavior based on the identified sample parking information includes:

[0166] For each type of information identified in the sample dwelling distance and / or sample dwelling frequency, determine the sample user's sample behavior characteristics related to the waiting-at-the-light behavior.

[0167] Optionally, the method for determining the sample user's sample behavior characteristics regarding the waiting-at-the-light behavior, based on the identified sample dwelling distance, includes:

[0168] Determine the number of times each of the identified sample stopping distances hits a preset sample distance interval; wherein, each sample distance interval is a different sample stopping distance range set for the sample trajectory type, and the probability of belonging to the sample trajectory type is different when different sample distance intervals are hit;

[0169] The number of times each preset sample distance interval is hit is determined as the sample user's sample behavior feature related to the waiting light behavior.

[0170] Optionally, the method for determining the sample user's sample behavior characteristics regarding the waiting-at-the-light behavior, based on the identified number of sample stops, includes:

[0171] Based on the number of times the samples stopped, determine the number of sample road segment sequences whose number of stops exceeds a preset threshold.

[0172] The determined quantity is used as the sample behavioral characteristics of the sample users regarding their behavior while waiting for lights.

[0173] Optionally, determining the sample stopping points on the sample road segment sequence from the trajectory points included in the sample trajectory data includes:

[0174] Determine the stopping points among the trajectory points included in the sample trajectory data;

[0175] Based on matching the location information of the stopping points with road network data, sample stopping points on the target road segment sequence are identified; wherein, the road network data contains the location information of the road segment sequence at each intersection.

[0176] To better understand the contents of the embodiments of this disclosure, a specific example will be used for illustration below.

[0177] To improve the accuracy of trajectory classification, this example introduces features related to user waiting-at-light behavior during the training process of the trajectory classification model. This allows the trained model to better distinguish between low-speed driving trajectories and high-speed cycling trajectories. The model training process mainly consists of two stages: the first stage involves mining the locations of stop lines in the road network data using large-scale sample trajectory data; the second stage uses the mined stop line locations to calculate the features of user waiting-at-light behavior represented by the sample trajectory data (corresponding to the sample behavior features mentioned above), and incorporates these waiting-at-light behavior features into the training process of the trajectory classification model, thereby improving the classification performance of the trajectory classification model.

[0178] The implementation steps for the first phase are as follows:

[0179] (1) Obtain trajectory data within a certain time range, and use the stopping point identification strategy and road network matching strategy to process the trajectory data to obtain the distribution data of stopping positions in the trajectory data belonging to a certain intersection link sequence (corresponding to the sample road segment sequence in the above text);

[0180] Among them, trajectory data within a certain time range, such as Figure 4 As shown in the figure, each small square represents a trajectory point.

[0181] (2) Based on the general understanding that vehicles wait at traffic lights in an orderly queue behind the stop line, and the assumption that the queue distance generally follows an exponential distribution, the queue distance of the stopping position and the intersection at a preset quantile in the trajectory data over a period of time is extracted and determined as the approximate position of the stop line.

[0182] (3) Perform smooth aggregation on multiple queuing distances near the preset quantile, such as taking the average of multiple queuing distances, to reduce errors and obtain a more accurate position of the intersection stop line.

[0183] The implementation process for the second phase is as follows:

[0184] like Figure 5 As shown, the stop line location data obtained from the first stage of stop line location mining based on road network data and sample trajectory data is introduced into the training process of the trajectory classification model. The main idea is as follows: the stop line data file is obtained by compiling the road network, that is, the stop line location data is bound to each link (road segment) in the road network data to obtain the stop line data file, which includes the distance from each link to the stop line location; the online module of model training reads the stop line locations in the data file and calculates the features of waiting at traffic lights based on the stop line locations.

[0185] The stop line data file mainly includes the following information: linkid, stopline_link_id, and dist_to_stopline, etc. Figure 6 As shown, `linkid` is the segment identifier of the current road segment, including `link1`, `link2`, and `link3`; `stopline_link_id` is the road segment where the stop line is located in the current road segment sequence consisting of `link1`, `link2`, and `link3`. Whether the trajectory passes through the intersection corresponding to this road segment sequence can be determined by checking if the trajectory subsequently passes through this road segment; `dist_to_stopline` is the distance from the first point of the road segment sequence to the stop line position.

[0186] The following is the calculation approach for the characteristics of the waiting-at-the-light behavior:

[0187] (1) Determine whether a trajectory point is a stopping point and store the identifier of whether the trajectory point is a stopping point in the information carried by the trajectory point.

[0188] The logic for determining the stopping point is as follows: For each trajectory point, look back up to 10 points in time. If the distance between any trajectory point and the current trajectory point is greater than 10m, the trajectory point is not a stopping point; otherwise, continue to determine whether the time difference between any trajectory point and the current trajectory point exceeds 10s. If so, the point is determined to be a stopping point; otherwise, it is not a stopping point.

[0189] (2) Traverse the sequence of sample trajectory points and extract the features of the lighting behavior.

[0190] The process for extracting features of the lighting behavior is as follows: Figure 7 As shown, steps S710-S760 are included:

[0191] S710, traverse the stopping points to find the cluster set of stopping points; that is, traverse the trajectory points with stopping indicators to obtain the cluster set of stopping points.

[0192] S720, delete the cluster point set that stops on non-road; if any trajectory point in the point set has a perpendicular distance greater than 20m from the matching road segment, it is considered to be stopped on non-road and the cluster point set is discarded.

[0193] S730, determine whether the first point in the cluster set matches a road segment sequence related to the intersection; take the road segment that matches the earliest first point in the cluster set in time, and determine whether it is a road segment sequence related to the intersection; if not, return to step S710, if yes, continue to step S740.

[0194] S740, calculate the distance from the starting point to the stop line, and record the road segments belonging to the intersection in the road segment sequence.

[0195] S750, determine whether the subsequent trajectory point passes through the road segment of the intersection; if yes, continue to step S760, otherwise return to step S710.

[0196] S760, confirm that the distance is a valid isolating feature and perform a distance mapping count.

[0197] The process of mapping and counting distances can be as follows: The distance to the traffic light is mapped to several preset distance intervals and statistically analyzed to serve as a feature of the waiting behavior. These distance intervals include: [0-30], [30-40], and [40, +∞), representing indistinguishable, distinguishable, and significantly distinguishable intervals, respectively. That is, if the distance to the traffic light is within 30 meters, it is difficult to distinguish whether the trajectory is a driving or cycling trajectory; the more times the trajectory is mapped to the 30-40 meter interval, the higher the probability of it being a driving trajectory. Conversely, if the distance to the traffic light is greater than 40 meters, it is highly probable that it is a driving trajectory. Additionally, another statistical feature can be calculated: the number of times the user stops more than once at the same intersection.

[0198] After calculating the features related to the above-extracted traffic light behavior, these features can be combined with the trajectory point features used in general trajectory classification models, that is, combined with trajectory point features such as velocity, acceleration, and displacement, to train the trajectory classification model and obtain a trajectory classification model with high accuracy.

[0199] As can be seen, by introducing features related to waiting at traffic lights during the model training process, this scheme enables the trained trajectory classification model to effectively distinguish between low-speed driving trajectories and high-speed cycling trajectories, thereby improving the classification accuracy of the trajectory classification model. Subsequent classification of trajectory data using this trained trajectory classification model can yield highly accurate classification results.

[0200] Corresponding to the above-described trajectory classification method embodiments, this disclosure also provides a trajectory classification device, such as... Figure 8 As shown, it includes:

[0201] The generation module 810 is used to generate specified behavioral features of the user to which the target trajectory data belongs, regarding the waiting-at-the-light behavior, based on the target trajectory data to be analyzed;

[0202] The determination module 820 is used to determine the model input content corresponding to the target trajectory data based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data;

[0203] The classification module 830 is used to input the determined model input content into a pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data;

[0204] The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

[0205] Optionally, the generation module includes:

[0206] The identification submodule is used to identify the target stopping information of the user to which the target trajectory data belongs on the target road segment sequence based on the target trajectory data to be analyzed; wherein, the target road segment sequence is the sequence of road segments related to intersections through which the trajectory represented by the target trajectory data passes, and the target stopping information is stopping information that varies for different modes of transportation;

[0207] A submodule is constructed to build specified behavioral features of the user to whom the target trajectory data belongs, based on the identified target stopping information, regarding the waiting behavior.

[0208] Optionally, the target stopping information includes: the stopping distance between the user's stopping position and the target stop line, and / or the number of stops; wherein the target stop line is the intersection stop line on the target road segment sequence.

[0209] Optionally, the identification submodule information is specifically used for:

[0210] From the trajectory points included in the target trajectory data, determine the target stopping points on the target road segment sequence;

[0211] Based on the determined target stopping point, identify the target stopping information of the user to whom the target trajectory data belongs on the target road segment sequence.

[0212] Optionally, the method for identifying the stopping distance includes:

[0213] Based on the determined target stopping point, determine the user's stopping position on the target road segment sequence;

[0214] Calculate a first distance between the stopping position and a preset endpoint of the target road segment, and a second distance between the preset endpoint and the target stop line; wherein, the target road segment is the road segment in which the stopping position is located among the road segments included in the target road segment sequence;

[0215] The stopping distance is determined based on the first distance and the second distance.

[0216] Optionally, determining the user's stopping position on the target road segment sequence based on the determined target stopping point includes:

[0217] Cluster analysis is performed on the identified target stopping points to obtain the target cluster point set;

[0218] The location of any stopping point in the target cluster set is determined as the stopping position of the user on the target road segment sequence.

[0219] Optionally, the method for identifying the number of pauses includes:

[0220] Cluster analysis is performed on the target stopping points to obtain the target cluster point set;

[0221] The number of stops is determined based on the number of target cluster points.

[0222] Optionally, the construction submodule is specifically used for:

[0223] For each type of information identified in the stopping distance and / or stopping frequency, determine the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior.

[0224] Optionally, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs, regarding the waiting-at-the-light behavior, for the identified stopping distance, includes:

[0225] Determine the number of times each identified stopping distance hits a preset distance interval; wherein, each distance interval is a different stopping distance range set for a specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit;

[0226] The number of times each preset distance interval is hit is determined as a specified behavioral feature of the user to whom the target trajectory data belongs, regarding the behavior of waiting at the traffic lights.

[0227] Optionally, for the identified number of stops, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior includes:

[0228] Based on the number of stops identified, determine the number of target road segment sequences whose number of stops exceeds a preset threshold.

[0229] The determined quantity is identified as a specified behavioral characteristic of the user to whom the target trajectory data belongs, regarding the behavior of waiting at traffic lights.

[0230] Optionally, determining the target stopping point on the target road segment sequence from the trajectory points included in the target trajectory data includes:

[0231] Determine the stopping points among the trajectory points included in the target trajectory data;

[0232] The target stopping point is identified by matching the location information of the stopping point with the road network data; wherein the road network data contains the location information of the road segment sequence at each intersection.

[0233] Corresponding to the above-described model training method embodiments, this disclosure also provides a model training apparatus, such as... Figure 9 As shown, it includes:

[0234] The sample generation module 910 is used to generate sample behavioral features of the sample user to which the sample trajectory data belongs, regarding the behavior of waiting at the traffic lights, based on the sample trajectory data;

[0235] The sample determination module 920 is used to determine the model input content corresponding to the sample trajectory data based on the sample behavior features and the trajectory point features corresponding to the sample trajectory data.

[0236] The sample classification module 930 is used to input the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, and obtain the trajectory classification result of the sample trajectory data.

[0237] The adjustment module 940 is used to adjust the model parameters of the trajectory classification model based on the trajectory classification results of the sample trajectory data and the labels corresponding to the sample trajectory data that characterize the trajectory type, so as to train the trajectory classification model.

[0238] Optionally, the sample generation module includes:

[0239] The sample identification submodule is used to identify the sample stopping information of the sample user to which the sample trajectory data belongs on the sample road segment sequence based on the sample trajectory data; wherein, the sample road segment sequence is the sequence of road segments related to intersections through which the trajectory represented by the sample trajectory data passes, and the sample stopping information is stopping information that varies for different modes of transportation;

[0240] The sample construction submodule is used to construct sample behavior features of the sample user regarding the waiting behavior based on the identified sample parking information.

[0241] Optionally, the sample stopping information includes: the sample stopping distance between the sample user's sample stopping position and the sample stopping line, and / or the number of sample stopping times; wherein, the sample stopping line is the intersection stopping line on the sample road segment sequence.

[0242] Optionally, the method for determining the position of the sample stop line includes:

[0243] Obtain multiple sample stopping points on the sample road segment sequence based on the trajectory points included in the sample trajectory data; calculate the distance from each sample stopping point to a preset endpoint of the sample road segment sequence to obtain multiple queuing distances;

[0244] The position of the sample stop line is obtained by analyzing the multiple queuing distances.

[0245] Optionally, the step of analyzing the plurality of queuing distances to obtain the position of the sample stop line includes:

[0246] The multiple queuing distances are sorted according to their magnitude to obtain the target sequence;

[0247] Determine the queuing distance at a preset quantile in the target sequence to obtain the distance to be utilized;

[0248] The position of the sample stop line is determined as the distance from the preset endpoint of the sample road segment sequence to be utilized.

[0249] Optionally, the sample identification submodule is specifically used for:

[0250] From the trajectory points included in the sample trajectory data, determine the sample stopping points on the sample road segment sequence;

[0251] Based on the determined sample stopping points, identify the sample stopping information of the user to whom the sample trajectory data belongs on the sample road segment sequence.

[0252] Optionally, the method for identifying the sample dwell distance includes:

[0253] Based on the determined sample stopping points, determine the sample stopping positions of the sample users on the sample road segment sequence;

[0254] Calculate the first sample distance between the sample stopping position and the preset endpoint of the sample road segment, and the second sample distance between the preset endpoint and the sample stop line; wherein, the sample road segment is the road segment in which the sample stopping position is located among the road segments included in the sample road segment sequence;

[0255] The stopping distance is determined based on the first sample distance and the second sample distance.

[0256] Optionally, determining the sample user's sample stopping position on the sample road segment sequence based on the determined sample stopping point includes:

[0257] Cluster analysis is performed on the identified sample stopping points to obtain the sample cluster point set;

[0258] The position of any stopping point in the sample cluster set is determined as the stopping position of the sample user on the sample segment sequence.

[0259] Optionally, the method for identifying the number of times the sample stops includes:

[0260] Cluster analysis was performed on the stopping points of the samples to obtain the sample cluster point set;

[0261] The number of times a sample stops is determined based on the number of cluster points in the sample cluster.

[0262] Optionally, the sample construction submodule is specifically used for:

[0263] For each type of information identified in the sample dwelling distance and / or sample dwelling frequency, determine the sample user's sample behavior characteristics related to the waiting-at-the-light behavior.

[0264] Optionally, the method for determining the sample user's sample behavior characteristics regarding the waiting-at-the-light behavior, based on the identified sample dwelling distance, includes:

[0265] Determine the number of times each of the identified sample stopping distances hits a preset sample distance interval; wherein, each sample distance interval is a different sample stopping distance range set for the sample trajectory type, and the probability of belonging to the sample trajectory type is different when different sample distance intervals are hit;

[0266] The number of times each preset sample distance interval is hit is determined as the sample user's sample behavior feature related to the waiting light behavior.

[0267] Optionally, the method for determining the sample user's sample behavior characteristics regarding the waiting-at-the-light behavior, based on the identified number of sample stops, includes:

[0268] Based on the number of times the samples stopped, determine the number of sample road segment sequences whose number of stops exceeds a preset threshold.

[0269] The determined quantity is used as the sample behavioral characteristics of the sample users regarding their behavior while waiting for lights.

[0270] Optionally, determining the sample stopping points on the sample road segment sequence from the trajectory points included in the sample trajectory data includes:

[0271] Determine the stopping points among the trajectory points included in the sample trajectory data;

[0272] Based on matching the location information of the stopping points with road network data, sample stopping points on the target road segment sequence are identified; wherein, the road network data contains the location information of the road segment sequence at each intersection.

[0273] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0274] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0275] An electronic device provided in this disclosure may include:

[0276] At least one processor; and

[0277] A memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of any of the trajectory classification methods or model training methods described above.

[0278] This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the trajectory classification methods or model training methods described above.

[0279] In another embodiment provided in this disclosure, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the steps of any of the trajectory classification methods or model training methods described above.

[0280] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0281] like Figure 10As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded from storage unit 1008 into random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0282] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0283] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as trajectory classification methods or model training methods. For example, in some embodiments, the trajectory classification method or model training method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the trajectory classification method or model training method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform a trajectory classification method or a model training method by any other suitable means (e.g., by means of firmware).

[0284] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0285] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0286] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0287] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0288] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0289] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0290] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0291] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A trajectory classification method, comprising: From the trajectory points included in the target trajectory data, determine the target stopping point on the target road segment sequence; wherein, the target road segment sequence is the sequence of road segments related to intersections traversed by the trajectory represented by the target trajectory data; Based on the determined target stopping point, identify the target stopping information of the user to whom the target trajectory data belongs on the target road segment sequence; wherein, the target stopping information is stopping information that varies for different modes of transportation; the target stopping information includes: the stopping distance between the user's stopping position and the target stop line; wherein, the target stop line is the intersection stop line on the target road segment sequence; Based on the identified target stopping information, a specified behavioral feature of the user to which the target trajectory data belongs regarding the waiting-at-light behavior is constructed; wherein, the specified behavioral feature of the waiting-at-light behavior is different for different modes of transportation; the specified behavioral feature includes: the number of times the stopping distance hits each preset distance interval; each distance interval is a different stopping distance range set for the specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit; Based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data, the model input content corresponding to the target trajectory data is determined; The determined model input content is input into the pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data; The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

2. The method according to claim 1, wherein, The target stopping information also includes: the number of times the target has stopped.

3. The method according to claim 1, wherein, The method for identifying the stopping distance includes: Based on the determined target stopping point, determine the user's stopping position on the target road segment sequence; Calculate a first distance between the stopping position and a preset endpoint of the target road segment, and a second distance between the preset endpoint and the target stop line; wherein, the target road segment is the road segment in which the stopping position is located among the road segments included in the target road segment sequence; The stopping distance is determined based on the first distance and the second distance.

4. The method according to claim 3, wherein, Determining the user's stopping position on the target road segment sequence based on the determined target stopping point includes: Cluster analysis is performed on the identified target stopping points to obtain the target cluster point set; The location of any stopping point in the target cluster set is determined as the stopping position of the user on the target road segment sequence.

5. The method according to claim 2, wherein, The method for identifying the number of pauses includes: Cluster analysis is performed on the target stopping points to obtain the target cluster point set; The number of stops is determined based on the number of target cluster points.

6. The method according to claim 2, wherein, The step of constructing specified behavioral features of the user to whom the target trajectory data belongs, regarding their waiting-at-the-light behavior, based on the identified target stopping information, includes: For each type of information identified in the stopping distance and / or stopping frequency, determine the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior.

7. The method according to claim 6, wherein, Regarding the identified stopping distance, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs, concerning their waiting-at-the-light behavior, includes: Determine the number of times each identified stopping distance hits a preset distance interval; wherein, each distance interval is a different stopping distance range set for a specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit; The number of times each preset distance interval is hit is determined as a specified behavioral feature of the user to whom the target trajectory data belongs, regarding the behavior of waiting at the traffic lights.

8. The method according to claim 6, wherein, Regarding the identified number of stops, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs, concerning their waiting-at-the-light behavior, includes: Based on the number of stops identified, determine the number of target road segment sequences whose number of stops exceeds a preset threshold. The determined quantity is identified as a specified behavioral characteristic of the user to whom the target trajectory data belongs, regarding the behavior of waiting at traffic lights.

9. The method according to any one of claims 1-8, wherein, Determining the target stopping point on the target road segment sequence from the trajectory points included in the target trajectory data includes: Determine the stopping points among the trajectory points included in the target trajectory data; The target stopping point is identified by matching the location information of the stopping point with the road network data; wherein the road network data contains the location information of the road segment sequence at each intersection.

10. A model training method, comprising: From the trajectory points included in the sample trajectory data, determine the sample stopping points on the sample road segment sequence; wherein, the sample road segment sequence is the sequence of road segments related to intersections traversed by the trajectory represented by the sample trajectory data. Based on the determined sample stopping points, identify the sample stopping information of the user to whom the sample trajectory data belongs on the sample road segment sequence; wherein, the sample stopping information is stopping information that varies for different modes of transportation; the sample stopping information includes: the sample stopping distance between the sample user's sample stopping position and the sample stop line; wherein, the sample stop line is the intersection stop line on the sample road segment sequence; Based on the identified sample parking information, sample behavioral features of the sample users regarding their waiting behavior at traffic lights are constructed; wherein, the specified behavioral features regarding waiting behavior differ for different modes of transportation. Based on the sample behavior features and the trajectory point features corresponding to the sample trajectory data, the model input content corresponding to the sample trajectory data is determined; the sample behavior features include: the number of times the stopping distance hits each preset distance interval; each distance interval is a different stopping distance range set for a specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit. Input the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, and obtain the trajectory classification result of the sample trajectory data; Based on the trajectory classification results of the sample trajectory data and the labels corresponding to the sample trajectory data used to characterize the trajectory type, the model parameters of the trajectory classification model are adjusted to train the trajectory classification model.

11. The method according to claim 10, wherein, The sample dwell information also includes: the number of times the sample has been dwelled.

12. The method according to claim 10, wherein, The method for determining the position of the sample stop line includes: Obtain multiple sample stopping points on the sample road segment sequence, determined based on the trajectory points included in the sample trajectory data; Calculate the distance from each sample stopping point to the preset endpoint of the sample road segment sequence to obtain multiple queuing distances; The position of the sample stop line is obtained by analyzing the multiple queuing distances.

13. The method according to claim 12, wherein, The step of analyzing the multiple queuing distances to obtain the position of the sample stop line includes: The multiple queuing distances are sorted according to their magnitude to obtain the target sequence; Determine the queuing distance at a preset quantile in the target sequence to obtain the distance to be utilized; The position of the sample stop line is determined as the distance from the preset endpoint of the sample road segment sequence to be utilized.

14. A trajectory classification device, comprising: The generation module is used to determine the target stopping points on the target road segment sequence from the trajectory points included in the target trajectory data; Based on the determined target stopping point, the target stopping information of the user to whom the target trajectory data belongs is identified on the target road segment sequence; wherein, the target road segment sequence is the sequence of road segments related to intersections through which the trajectory represented by the target trajectory data passes, and the target stopping information is stopping information that varies for different modes of transportation; the target stopping information includes: the stopping distance between the user's stopping position and the target stop line; wherein, the target stop line is the intersection stop line on the target road segment sequence; Based on the identified target stopping information, a specified behavioral feature of the user to which the target trajectory data belongs regarding the waiting-at-light behavior is constructed; wherein, the specified behavioral feature of the waiting-at-light behavior is different for different modes of transportation; the specified behavioral feature includes: the number of times the stopping distance hits each preset distance interval; each distance interval is a different stopping distance range set for the specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit; The determination module is used to determine the model input content corresponding to the target trajectory data based on the specified behavioral features and the trajectory point features corresponding to the target trajectory data; The classification module is used to input the determined model input content into a pre-trained trajectory classification model to obtain the trajectory classification result of the target trajectory data; The trajectory classification model is a neural network model trained based on the model input content corresponding to the sample trajectory data and the labels used to represent the trajectory type.

15. The apparatus according to claim 14, wherein, The target stopping information also includes: the number of times the target has stopped.

16. The apparatus according to claim 14, wherein, The method for identifying the stopping distance includes: Based on the determined target stopping point, determine the user's stopping position on the target road segment sequence; Calculate a first distance between the stopping position and a preset endpoint of the target road segment, and a second distance between the preset endpoint and the target stop line; wherein, the target road segment is the road segment in which the stopping position is located among the road segments included in the target road segment sequence; The stopping distance is determined based on the first distance and the second distance.

17. The apparatus according to claim 16, wherein, Determining the user's stopping position on the target road segment sequence based on the determined target stopping point includes: Cluster analysis is performed on the identified target stopping points to obtain the target cluster point set; The location of any stopping point in the target cluster set is determined as the stopping position of the user on the target road segment sequence.

18. The apparatus according to claim 15, wherein, The method for identifying the number of pauses includes: Cluster analysis is performed on the target stopping points to obtain the target cluster point set; The number of stops is determined based on the number of target cluster points.

19. The apparatus according to claim 15, wherein, Build submodules, specifically for: For each type of information identified in the stopping distance and / or stopping frequency, determine the specified behavioral characteristics of the user to whom the target trajectory data belongs regarding the waiting-at-the-light behavior.

20. The apparatus according to claim 19, wherein, Regarding the identified stopping distance, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs, concerning their waiting-at-the-light behavior, includes: Determine the number of times each identified stopping distance hits a preset distance interval; wherein, each distance interval is a different stopping distance range set for a specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit; The number of times each preset distance interval is hit is determined as a specified behavioral feature of the user to whom the target trajectory data belongs, regarding the behavior of waiting at the traffic lights.

21. The apparatus according to claim 19, wherein, Regarding the identified number of stops, the method for determining the specified behavioral characteristics of the user to whom the target trajectory data belongs, concerning their waiting-at-the-light behavior, includes: Based on the number of stops identified, determine the number of target road segment sequences whose number of stops exceeds a preset threshold. The determined quantity is identified as a specified behavioral characteristic of the user to whom the target trajectory data belongs, regarding the behavior of waiting at traffic lights.

22. The apparatus according to any one of claims 14-21, wherein, Determining the target stopping point on the target road segment sequence from the trajectory points included in the target trajectory data includes: Determine the stopping points among the trajectory points included in the target trajectory data; The target stopping point is identified by matching the location information of the stopping point with the road network data; wherein the road network data contains the location information of the road segment sequence at each intersection.

23. A model training device, comprising: The sample generation module is used to determine the sample stopping points on the sample road segment sequence from the trajectory points included in the sample trajectory data. Based on the determined sample stopping points, the sample stopping information of the user to whom the sample trajectory data belongs is identified on the sample road segment sequence; wherein, the sample road segment sequence is the sequence of road segments related to intersections through which the trajectory represented by the sample trajectory data passes, and the sample stopping information is stopping information that varies for different modes of transportation; the sample stopping information includes: the sample stopping distance between the sample user's sample stopping position and the sample stop line; wherein, the sample stop line is the intersection stop line on the sample road segment sequence; Based on the identified sample parking information, sample behavioral features of the sample users regarding their waiting behavior at traffic lights are constructed; wherein, the specified behavioral features regarding waiting behavior differ for different modes of transportation. The sample determination module is used to determine the model input content corresponding to the sample trajectory data based on the sample behavior features and the trajectory point features corresponding to the sample trajectory data; the sample behavior features include: the number of times the stopping distance hits each preset distance interval; each distance interval is a different stopping distance range set for a specified trajectory type, and the probability of belonging to the specified trajectory type is different when different distance intervals are hit. The sample classification module is used to input the model input content corresponding to the sample trajectory data into the trajectory classification model to be trained, and obtain the trajectory classification result of the sample trajectory data. An adjustment module is used to adjust the model parameters of the trajectory classification model based on the trajectory classification results of the sample trajectory data and the labels corresponding to the sample trajectory data that characterize the trajectory type, in order to train the trajectory classification model.

24. The apparatus according to claim 23, wherein, The sample dwell information also includes: the number of times the sample has been dwelled.

25. The apparatus according to claim 23, wherein, The method for determining the position of the sample stop line includes: Obtain multiple sample stopping points on the sample road segment sequence, determined based on the trajectory points included in the sample trajectory data; Calculate the distance from each sample stopping point to the preset endpoint of the sample road segment sequence to obtain multiple queuing distances; The position of the sample stop line is obtained by analyzing the multiple queuing distances.

26. The apparatus according to claim 25, wherein, The step of analyzing the multiple queuing distances to obtain the position of the sample stop line includes: The multiple queuing distances are sorted according to their magnitude to obtain the target sequence; Determine the queuing distance at a preset quantile in the target sequence to obtain the distance to be utilized; The position of the sample stop line is determined as the distance from the preset endpoint of the sample road segment sequence to be utilized.

27. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the trajectory classification method of any one of claims 1-9, or the model training method of any one of claims 10-13.

28. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the trajectory classification method according to any one of claims 1-9, or the model training method according to any one of claims 10-13.

29. A computer program product comprising a computer program that, when executed by a processor, implements the trajectory classification method according to any one of claims 1-9, or the model training method according to any one of claims 10-13.

Images