Track prediction method, electronic equipment and storage medium

Abstract

The invention discloses a trajectory prediction method, electronic equipment and a storage medium. The method comprises the following steps: acquiring target trajectory point data of a target object; the target trajectory point data comprises a plurality of trajectory points acquired at different time points; performing feature extraction on the time point data corresponding to the target track point data to obtain time features; and track prediction is carried out on the target track point data based on the time features to obtain a target prediction result, and the target prediction result comprises one or more prediction points of a next track point after the target track point data. According to the scheme, the accuracy of trajectory prediction can be improved.

Description

Technical Field

[0001] This application relates to the field of trajectory data processing, and in particular to a trajectory prediction method, electronic device, and storage medium. Background Technology

[0002] Trajectory prediction can infer the future location of a target object and provide predictive information about its future state. This predictive information can be widely used in object-related planning, resource allocation, decision-making, and other related activities.

[0003] The accuracy of trajectory prediction has become a crucial factor affecting the effectiveness of downstream applications, and how to improve the accuracy of trajectory prediction has become an urgent problem to be solved. Summary of the Invention

[0004] This application provides at least one trajectory prediction method, electronic device, and storage medium.

[0005] This application provides a trajectory prediction method, including: acquiring target trajectory point data of a target object; the target trajectory point data includes multiple trajectory points collected at different time points; extracting features from the time point data corresponding to the target trajectory point data to obtain time features; and performing trajectory prediction on the target trajectory point data based on the time features to obtain a target prediction result, the target prediction result including one or more prediction points for the next trajectory point after the target trajectory point data.

[0006] This application provides an electronic device, including a memory and a processor, wherein the processor is configured to execute program instructions stored in the memory to implement any of the methods described above.

[0007] This application provides a computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, implement any of the methods described above.

[0008] In the above scheme, feature extraction is performed on the time point data corresponding to the target trajectory point data to obtain time features. The distribution of the object trajectory has time characteristics. Combining the time features makes the next trajectory point match the time characteristics of the trajectory, thereby improving the accuracy of trajectory prediction.

[0009] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application. Attached Figure Description

[0010] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.

[0011] Figure 1 This is a flowchart illustrating an embodiment of the trajectory prediction method of this application; Figure 2 This is a flowchart illustrating another embodiment of step S110 of this application; Figure 3 This is a schematic diagram of an embodiment of the trajectory prediction method of this application; Figure 4 This is a schematic diagram of the framework of an embodiment of the electronic device of this application; Figure 5 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application. Detailed Implementation

[0012] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0013] In the following description, specific details such as particular subsystem structures, interfaces, and technologies are presented for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of this application.

[0014] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, "many" in this document means two or more. Moreover, the term "at least one" in this document means any combination of at least two of any one or more of a plurality of objects. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0015] See Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the trajectory prediction method of this application. Specifically, the method may include: Step S110: Obtain the target trajectory point data of the target object.

[0016] The target object can be any type of object whose trajectory is to be tracked. The target trajectory point data can include multiple trajectory points collected at different times. The number of trajectory points included in the target trajectory point data can be set according to the actual application requirements. In some implementation scenarios, the target trajectory point data can include two trajectory points.

[0017] It should be noted that the target trajectory point data can contain multiple trajectory points and corresponding time points, and the trajectory points can be arranged in order of their collection time.

[0018] In some embodiments, a trajectory point can represent the acquisition point where the target object is located at that time. In some implementation scenarios, a trajectory point can be the acquisition point where the target object is located. If the target object is within the acquisition range of an acquisition point, then that acquisition point is used as the trajectory point of the target object.

[0019] In some implementation scenarios, if the images collected by the collection point are identified and a target object is found in the collected image, it can be determined that the target object is within the collection range of the collection point, and the collection point can be identified as a trajectory point of the target object.

[0020] In some implementation scenarios, the target object appears at two different locations within the collection range of a certain collection point, and the corresponding two trajectory points can be the same.

[0021] It should be noted that the target trajectory point data used for prediction above can be the latest trajectory point data of the target object, so as to accurately predict the next trajectory of the target object.

[0022] Step S120: Extract features from the time point data corresponding to the target trajectory point data to obtain time features.

[0023] Among them, the time points corresponding to the trajectory points contained in the target trajectory point data can form time point data, which can represent the information of the target object trajectory in the time dimension.

[0024] Step S130: Perform trajectory prediction on the target trajectory point data based on time features to obtain the target prediction result.

[0025] The target prediction result may include one or more predicted points for the next trajectory point after the target trajectory point data.

[0026] Trajectory prediction aims to obtain the next trajectory point of a target object. A single trajectory prediction can yield one or more possible next trajectory points, i.e., one or more predicted points. In one implementation scenario, the target prediction result may include two predicted points, A and B, indicating that the next trajectory point of the target object can be either A or B.

[0027] It should be noted that the trajectory point represents the collection point where the target object is located, and predicting the next trajectory point means predicting the next collection point where the target object will appear.

[0028] An object's trajectory is time-dependent. For example, travel patterns are closely related to each object's habits and preferences, including preferences on different dates and at different times. Therefore, temporal characteristics help accurately predict an object's trajectory. Extracting the temporal characteristics of existing target trajectory point data for the target object enriches the basis for trajectory prediction, allowing for the prediction of the target object's next trajectory that conforms to these temporal characteristics, thus improving the accuracy of trajectory prediction.

[0029] In some embodiments, the time feature may include multiple sub-time features. For example, it may include at least one of a first sub-time feature and a second sub-time feature. The first sub-time feature may characterize whether the date to which the time point belongs is a weekday or a non-working day. The second sub-time feature may characterize the weekday number of the date to which the time point belongs. Characterizing time information from multiple dimensions allows for a more accurate description of time features, enriches the basis for trajectory prediction, fully utilizes time information from multiple dimensions, and improves the accuracy of trajectory prediction.

[0030] In some implementation scenarios, the time feature can also include a third sub-time feature, which can characterize the time period to which a given time point belongs. The time period can be predefined, representing a smaller unit of time divided from a day. One or more time period division methods can be set, and the corresponding third sub-time feature can be extracted according to each method. For example, a day can be divided into 24 time periods by hour, and the third sub-time feature can represent the time period to which a given time point belongs; and / or, a day can be divided into peak and off-peak time periods, and the third sub-time feature can indicate whether a given time point belongs to a peak or off-peak time period.

[0031] In some implementation scenarios, sub-time features can be extracted for each time point separately, and then the sub-time features of all time points can be merged to obtain the time feature. For example, the first and second sub-time features can be extracted for each time point separately, and then the first and second sub-time features of all time points can be merged to obtain the time feature.

[0032] In some implementation scenarios, all time points can be fused first, and the sub-time features of the fusion result can be extracted separately. The sub-time features can then be fused together to obtain the time features.

[0033] In some embodiments, trajectory prediction can be achieved using a neural network model. The target trajectory point data is used as input data to the model, which can extract features from the input data. This step may include extracting features from time point data to obtain temporal features, and then the model can use these temporal features to ultimately output a prediction result about the next trajectory point.

[0034] In some implementation scenarios, trajectory prediction is achieved using at least one of a target-object model applicable to the target object and a general model applicable to all groups. If only one model is used, the model output can be used as the target prediction result. If two models are used, the outputs of the two models can be combined as the target prediction result.

[0035] In some embodiments, the target object model and the general group model may have the same or different model structures. Both are capable of extracting features from time-to-time point data and using the extracted time features for trajectory prediction.

[0036] In some implementation scenarios, feature extraction is performed on trajectory point data and corresponding time point data to obtain trajectory point features and time features, and these two features are combined to achieve trajectory prediction.

[0037] Specifically, a target object model is trained in advance using the first sample trajectory point data of the target object, which can be used to predict the trajectory of the target object. A general population model is trained in advance using the second sample trajectory point data of multiple sample objects, including the target object. The general population model can be used to predict the trajectory of any sample object. The sample trajectory point data can be preprocessed, including deduplication, trajectory correction, trajectory completion, trajectory fusion, and trajectory segmentation.

[0038] In some implementation scenarios, the target trajectory point data is input into the target object model and the general group model respectively. The former outputs the first prediction result and the latter outputs the second prediction result. The final target prediction result is determined by combining the first and second prediction results. Specifically, one or more prediction points in the target prediction result can be determined by combining the two.

[0039] In a specific application scenario, the first prediction result may include one or more first candidate trajectory points for the next trajectory point and a first confidence level for each first candidate trajectory point. The second prediction result may include one or more second candidate trajectory points for the next trajectory point and a second confidence level for each second candidate trajectory point. One or more trajectory points may be determined by the following steps: based on the first confidence level and the second confidence level, at least one of the first candidate trajectory points and all second candidate trajectory points is selected as the prediction point.

[0040] In a specific application scenario, models typically set a confidence threshold when outputting, and use trajectory points with a confidence level greater than the threshold, along with their confidence levels, as the model output. Generally, the threshold is set to be no less than 1 / 3, so a model can output one or two candidate trajectory points.

[0041] Furthermore, the specific steps for selecting prediction points using the first and second confidence levels can be set according to actual conditions and are not limited here. For example, the probability corresponding to the candidate trajectory points is calculated using the first and second confidence levels, and at least one of them is selected as the prediction point based on the probability.

[0042] In some embodiments, the target prediction result may also include the prediction time corresponding to each prediction point. After obtaining the prediction points, the prediction time period corresponding to the prediction points may be obtained through the following steps: process each prediction point separately, and form a candidate trajectory by combining the latest trajectory point in the target trajectory point data with the prediction point; select a reference sub-trajectory that is the same as the candidate trajectory from the first reference trajectory point data, and determine the prediction time of the corresponding prediction point based on the trajectory duration of the reference sub-trajectory.

[0043] Furthermore, the prediction time corresponding to the prediction point can be a point in time or a time period. The following explanation uses a time period as an example: the trajectory durations of all reference sub-trajectories can be arranged according to their length to form a duration sequence; the upper and lower quartiles of the duration sequence are used as the prediction time periods for the corresponding prediction points.

[0044] The first reference trajectory point data can be historical trajectory point data of the target object, or it can include historical trajectory point data of multiple sample objects. The first reference trajectory point data can be trajectory point data that has undergone preprocessing such as deduplication and segmentation.

[0045] In some implementation scenarios, for the first candidate trajectory point predicted by the target object model, the historical trajectory point data of the target object can be used as the first reference trajectory point data to calculate the corresponding prediction time point or prediction time period. For the second candidate trajectory point predicted by the general population model, the historical trajectory point data of multiple sample objects can be used as the first reference trajectory point data to calculate the corresponding prediction time point or prediction time period.

[0046] In one possible scenario, the first candidate trajectory point predicted by the target object model and the second candidate trajectory point predicted by the general population model may have overlapping trajectory points. For overlapping trajectory points, the historical trajectory point data of the target object and the historical trajectory point data of multiple sample objects can be used as the first reference trajectory point data to calculate two prediction times. Then, the final prediction time can be obtained based on these two prediction times.

[0047] In some embodiments, a target object model is trained in advance using first sample trajectory point data of the target object. This first sample trajectory point data may include trajectory point data of the target object over a historical period. The first sample trajectory point data may be trajectory point data that has undergone preprocessing steps such as deduplication and trajectory segmentation. The first sample trajectory point data may include multiple trajectory segments. The model can learn the correlation between the target object's trajectory and time from the first sample trajectory point data. Then, during the model inference stage, it can use the temporal characteristics of the trajectory to predict the next trajectory point, ensuring that the next trajectory point conforms to the trajectory pattern of the target object under this time characteristic, thus improving trajectory accuracy. The temporal characteristics include sub-features in multiple dimensions, enabling the model to learn the temporal characteristics of the trajectory in different dimensions, to more finely explore the correlation between trajectory and time, fully utilize time information for prediction, and improve accuracy.

[0048] The first sample trajectory point data processed by the target object model in one step has the same format as the target trajectory point data. For example, both can contain two trajectory points and their corresponding time points. The processing steps of the target object model for the first sample trajectory point data can refer to the processing steps for the target trajectory point data in the previous embodiment, and will not be repeated here. The same applies to the general population model.

[0049] See Figure 2 , Figure 2 This is a flowchart illustrating another embodiment of step S110 of this application. Step S110 may include: Step S211: Obtain real-time trajectory point data of the target object.

[0050] Real-time trajectory point data refers to the existing trajectory point data of the target object that can be obtained at the current moment. The real-time trajectory point data is preprocessed, and the latest trajectory point data is selected from the preprocessed data as the target trajectory point data for prediction. Preprocessing may include deduplication.

[0051] It should be noted that all collection points can be pre-classified, and there are correlations between collection points of the same category. Collection point categories can be used to simplify trajectory point data, for example, to deduplicate trajectory point data, so as to improve the efficiency of subsequent processing while ensuring the amount of information contained in the trajectory point data.

[0052] In some embodiments, all collection points within a certain area can be classified.

[0053] Step S212: Based on the category of the collection point corresponding to the real-time trajectory point, perform deduplication processing on the real-time trajectory point data to obtain the deduplicated real-time trajectory point data.

[0054] The category of the collection points can be used to identify duplicate trajectory points in real-time trajectory points. Furthermore, since trajectory point data contains a series of trajectory points arranged in chronological order, multiple consecutive collection points of the same category can be considered duplicate trajectory points and can be simplified.

[0055] Classifying the collection points and merging those of the same category for deduplication can further and more accurately remove duplicate information from the trajectory point data, improve the efficiency of subsequent processing, and prevent the dilution of probability caused by separating multiple similar collection points, which could lead to prediction errors or duplicate predictions, thus effectively improving the prediction results.

[0056] In some embodiments, deduplication may include: determining one or more sub-trajectories to be processed from the real-time trajectory point data, wherein the sub-trajectory points to be processed contain two or more consecutive real-time trajectory points and the collection points corresponding to the real-time trajectory points belong to the same category; for each sub-trajectory to be processed, deleting other real-time trajectory points in the sub-trajectory to be processed except for the first real-time trajectory point and the last real-time trajectory point, to obtain the deduplicated real-time trajectory point data.

[0057] In a specific application scenario, the real-time trajectory point data of the target object includes Where c represents the collection point and t represents time. Collection points c2 and c3 are of the same type. The target object has a total of 4 trajectory points at collection points c2 and c3, which is the sub-trajectory to be processed. Only the first and last real-time trajectory points are retained. The deduplicated real-time trajectory point data is as follows: .

[0058] Step S213: Select the target trajectory point data from the deduplicated real-time trajectory point data.

[0059] In some implementation scenarios, the latest, preset number of consecutive real-time trajectory points can be selected from the real-time trajectory point data as the target trajectory point data.

[0060] In a specific application scenario, the two newest consecutive real-time trajectory points are selected from the real-time trajectory point data as the target trajectory point data.

[0061] In some embodiments, a trajectory segmentation step may be performed before step S130. Specifically, the real-time trajectory point data after deduplication can be segmented to obtain multiple real-time sub-trajectories.

[0062] Trajectory point data contains the trajectory points of an object over a period of time, but the object's trajectory may not be continuous within this time. Therefore, trajectory point data can be segmented to more accurately represent the activity trajectory of the target object.

[0063] In some embodiments, the velocity and time difference between adjacent real-time trajectory point pairs are calculated, and it is determined whether the velocity and time difference meet the segmentation conditions. If the segmentation conditions are met, a segmentation is performed between the two trajectory points contained in this pair of adjacent real-time trajectory point pairs.

[0064] In some implementation scenarios, the segmentation conditions may include that the velocity between adjacent real-time trajectory point pairs is less than a velocity threshold and the time difference is greater than a time difference threshold. If the segmentation conditions are met, it can be considered that the target object stops between these two trajectory points, the trajectory does not have continuity, and segmentation can be performed.

[0065] In a specific application scenario, the real-time trajectory point data of the target object includes Using adjacent trajectory point pairs Let's take an example to illustrate. The time difference can be expressed as t3-t2, and the speed can be expressed as... .like Less than the speed threshold And t3-t2 is less than the time difference threshold. If so, a segmentation can be performed between these two trajectory points. If other adjacent point pairs do not meet the segmentation condition, the trajectory segment can be divided into two sub-trajectories, represented as follows: and .

[0066] In some embodiments, before selecting target trajectory point data, the real-time trajectory point data is divided into multiple real-time sub-trajectories, and the latest preset number of consecutive real-time trajectory points can be selected from the latest real-time sub-trajectories as target trajectory point data.

[0067] In some embodiments, the preprocessing of trajectory point data may further include trajectory correction, trajectory completion, and trajectory fusion. For example, the trajectory point data is first deduplicated, then subjected to trajectory correction, trajectory completion, and trajectory fusion, and finally segmented.

[0068] It should be noted that trajectory correction and trajectory completion can be performed on trajectory point data of a single object. Trajectory fusion can merge trajectory point data of different objects. If trajectory point data of a target object is needed separately, the trajectory points of the target object can be deduplicated before trajectory correction and trajectory completion are performed, and it is determined whether fusion with other trajectories is needed, before trajectory segmentation.

[0069] In some implementation scenarios, after trajectory fusion, the fused trajectory can be preprocessed again for individual objects.

[0070] In real-world scenarios, due to factors such as acquisition angle, built-in view recognition algorithm version, image clarity, weather conditions, and lighting intensity, numerous target objects are often missed or misidentified, resulting in incomplete or erroneous trajectories for each object. Therefore, trajectory correction, completion, and fusion are necessary to restore the correct and complete trajectory as much as possible. Common trajectory completion methods include, but are not limited to, determining whether there were missed shots based on road network data and images from two consecutive acquisition points. Common trajectory correction methods include, but are not limited to, determining whether the distance and time difference between adjacent trajectory points exceeds a speed threshold, or identifying abnormal appearances of the target object within a certain area. Common trajectory fusion methods include, but are not limited to, utilizing the relationships between entity objects to fuse the trajectories of different entity objects, thereby providing more comprehensive information.

[0071] In some embodiments, the category of the collection points needs to be determined before deduplication. Specifically, all collection points can be clustered to obtain the category of each collection point.

[0072] In some implementation scenarios, clustering methods include, but are not limited to, DBSCAN. After clustering, collection points in the same cluster are considered similar collection points, and similar collection points are merged based on the clustering results and treated as collection points in the same cluster for analysis.

[0073] Furthermore, clustering parameters for each data collection point are calculated, and these parameters are used to cluster all data collection points. In some embodiments, the clustering parameters for the data collection points can be calculated based on trajectory point data.

[0074] In some embodiments, each collection point can be used as a target collection point, and each collection point can be processed separately. The target collection points are combined with other collection points to obtain one or more collection point combinations. The correlation parameters of each collection point combination are calculated and used as the clustering parameters of the target collection points.

[0075] In this context, "other collection points" refers to collection points other than the target collection point. A collection point combination consists of two collection points: the target collection point and one other collection point. For example, there are 10 collection points in total, with collection point 1 serving as the target collection point. The remaining 9 collection points are all other collection points. The target collection point can be combined with each of the other collection points, resulting in 9 collection point combinations for collection point 1.

[0076] The association parameters for a collection point combination can include at least one of the following: collection point distance, collection point correlation, and collection point similarity. Collection point distance represents the distance between two collection points in the collection point combination.

[0077] The correlation between collection points represents the degree of correlation between two collection points in a combination of collection points. The probability of this combination of collection points occurring can be used to characterize the correlation between collection points, which is the probability that these two collection points will appear in a fixed combination. The higher the correlation between the two collection points, the more likely they are to appear in such a fixed combination, and the higher the probability of this combination of collection points occurring.

[0078] Collection point similarity represents the similarity between two collection points in this collection point combination. Specifically, the similarity between two collection points can be compared from two perspectives: the previous collection point and the next collection point.

[0079] For each combination of target collection points, calculate its association parameters. Calculating association parameters may include calculating at least one of the following: collection point distance, collection point association degree, and collection point similarity.

[0080] In some embodiments, the step of calculating the distance between collection points may include: calculating the distance between collection points corresponding to the collection point combination based on the position information of two collection points in the collection point combination.

[0081] In a specific application scenario, the road network distance between two collection points is calculated based on the road network; if there is no road network, the straight-line distance is calculated.

[0082] It should be noted that two data collection points are combined into a single data collection point group, and these two data collection points can be in a specific order. This order can represent the movement from one data collection point to the next. For example, for data collection point 1 and data collection point 2, there are two possible combinations: one where data collection point 1 comes first and data collection point 2 comes second, indicating a movement from data collection point 1 to data collection point 2; and another where data collection point 2 comes first and data collection point 1 comes second, indicating a movement from data collection point 2 to data collection point 1. The calculated distance and correlation between the data collection points may differ depending on the combination method.

[0083] Similarly, when a target collection point is combined with other collection points, there are two possible combinations: moving from the target collection point to other collection points, or moving from other collection points to the target collection point. The road network distances calculated by these two methods may be different.

[0084] In some implementation scenarios, when combining a target data collection point with other data collection points, the target data collection point can be designated as the first data collection point, and the other data collection points as the subsequent data collection points. When calculating road network distances, the distance from the target data collection point to the other data collection points is calculated.

[0085] In some implementation scenarios, when combining a target collection point with other collection points, two methods can be used respectively. The distance of the collection point corresponding to the combination of collection points can be obtained by combining the distances calculated under the two methods.

[0086] In some embodiments, the calculation steps of the collection point correlation degree include: counting the first occurrence of the collection point combination in the second reference trajectory point data and the second occurrence of the associated collection point combination, and taking the ratio of the first occurrence to the second occurrence as the collection point correlation degree of the collection point combination.

[0087] Among them, the associated collection point combination is all collection point combinations composed of the target collection point.

[0088] In a specific application scenario, sampling point 1 is used as the target sampling point. It is combined with 9 other sampling points to obtain 9 sets of sampling point combinations. One of the 9 sets of sampling point combinations is used as the sampling point combination currently being calculated. The number of times the sampling point combination currently being calculated appears is the first count, and the total number of times the 9 sets of sampling point combinations appear is the second count.

[0089] It should be noted that when calculating the correlation of collection points, the combination of collection points can be such that the target collection point is the first collection point and the other collection points are the subsequent collection points, and the first and second counts under this combination method are counted.

[0090] In a specific application scenario, sampling point 1 is the target sampling point, and sampling points 2 through 10 are the other sampling points. Sampling point 1, as the prior sampling point, is combined with sampling points 2 through 10, respectively, and represented as 1-2, 1-3, 1-4, 1-5, 1-6, 1-7, 1-8, 1-9, and 1-10. For the sampling point combination 1-2, the sampling point correlation is the ratio of the first occurrence of sampling point combination 1-2 to the second occurrence of all nine combinations. Note that when sampling point 1 is the target sampling point, the combination of sampling point 1 and sampling point 2 is used instead of sampling point combination 2-1, while when sampling point 2 is the target sampling point, sampling point combination 2-1 is used instead of sampling point combination 1-2.

[0091] The second reference trajectory point data can contain trajectory point data from multiple objects. It should be noted that the second reference trajectory point data can be deduplicated and segmented based on the acquisition points. Since the acquisition points have not yet been classified, deduplication can be performed based on the acquisition points during trajectory point preprocessing. This means that two or more consecutive identical trajectory points (i.e., the same acquisition point) are treated as sub-trajectories to be processed, retaining only the first and last trajectory points.

[0092] In a specific application scenario, let's take the trajectory point data of an object as an example. There exists an object with three recent trajectory segments, namely... , , Among them, the object after undergoing The next collection point is The number of times The next collection point is The number of times The number of data point transitions for each combination of data points in the trajectory data of all objects is counted, and then the correlation degree of the data points is calculated, which can also be called the data point transition probability. At that time, then after The next collection point is The probability is 2 / 3, the next sampling point is The probability is 1 / 3.

[0093] The collection point combination includes two collection points: one is the target collection point, and the other is another collection point. In some embodiments, the steps for calculating the collection point similarity include: counting the overlapping collection points between the adjacent collection points of the target collection point and the adjacent collection points of other collection points in the second reference trajectory point data; for each overlapping collection point, calculating the first similarity of the overlapping collection points using the first and second collection point correlation degrees corresponding to the overlapping collection points; calculating the second similarity of adjacent collection points based on the first similarity of each overlapping collection point; and calculating the collection point similarity of the collection point combination based on the second similarity of the previous collection point and the second similarity of the next collection point.

[0094] It should be noted that when calculating the similarity between two data collection points, the previous and next data collection points connected to the two points can be compared separately to determine the similarity between the two data collection points. The previous and next data collection points are considered as adjacent data collection points, and calculating the similarity between two data collection points with respect to adjacent data collection points is equivalent to calculating the similarity between the two data collection points with respect to the previous and next data collection points respectively.

[0095] Among them, the correlation degree of the first collection point corresponding to the overlapping collection point is the correlation degree between the overlapping collection point and the collection point corresponding to the target collection point, and the correlation degree of the second collection point corresponding to the overlapping collection point is the correlation degree between the overlapping collection point and the collection points corresponding to other collection points.

[0096] In some implementation scenarios, taking the previous acquisition point as an example, we find all the previous acquisition points of the target acquisition point in the second reference trajectory point data, as well as all the previous acquisition points of other acquisition points. We then compare these two sets of data to find one or more overlapping acquisition points.

[0097] The correlation between the first and second sampling points is related to the properties of adjacent sampling points. For example, if the previous sampling point is considered an adjacent sampling point, it means that overlapping sampling points are selected from the previous sampling point. The correlation between the first and second sampling points is calculated by taking the overlapping sampling point as the preceding sampling point and the target sampling point as the following sampling point.

[0098] Similarly, "next collection point as adjacent collection point" means selecting overlapping collection points from the next collection point. The first collection point correlation is calculated by taking the target collection point as the preceding collection point and the overlapping collection point as the following collection point. The second collection point correlation is calculated by taking other collection points as the preceding collection points and the overlapping collection point as the following collection point.

[0099] In a specific application scenario, suppose there are two data collection points. and Analyze the similarity between the preceding and following data collection points. For example, The previous collection point was And the probability of data collection point transfer (data collection point correlation) are respectively . The previous collection point was And the probability of data collection point transfer (data collection point correlation) are respectively It can be found that... and There is the same previous collection point and .right In other words, utilizing its associations and The first similarity can be calculated. ,right In other words, utilizing its associations and The first similarity can be calculated. The second similarity of the previous sampling point can be calculated using the first similarity between these two overlapping sampling points, denoted as: .in, This refers to methods for calculating the similarity of multiple similar data collection points, which may include, but are not limited to, direct summation. Similarly, the calculation... and Similarity between two sampling points and the subsequent sampling point Ultimately obtained and Similarity between two sampling points .

[0100] It should be noted that when calculating the similarity of collection points, there is no specific order between the two collection points in the combination. and The similarity between collection points can be used as the collection point... The corresponding associated parameters can also be used as data collection points. The corresponding associated parameters.

[0101] In some embodiments, when calculating the second similarity, different weights can be assigned to different first similarities based on the absolute value of the association degree of the collection points. The larger the absolute value, the greater the probability of the collection point shifting, and a larger weight can be assigned.

[0102] In some embodiments, the association parameters can be preprocessed before clustering. Specifically, if the association degree of the collection points is greater than the association degree threshold, the two collection points can be considered as strongly associated collection points, and the association degree of the collection points can be adjusted to a first preset value, which represents the maximum value of the association degree of the collection points.

[0103] In some embodiments, if the similarity of the collection points is greater than the similarity threshold, the two collection points are considered to be parallel and replaceable collection points, and the similarity of the collection points can be adjusted to a second preset value, which represents the maximum value of the similarity of the collection points.

[0104] In a specific application scenario, the correlation of collection points that are greater than or equal to the correlation threshold is adjusted to 1.

[0105] In a specific application scenario, the similarity of collection points that are greater than or equal to the similarity threshold is adjusted to 1.

[0106] In some embodiments, the device may also display a trajectory prediction page, which displays a map of the area where the collection point is located, and after the prediction point is obtained, the target trajectory point data and one or more prediction points for the next trajectory point can be displayed on the map to intuitively reflect the trajectory of the target object and the target prediction result.

[0107] In some embodiments, the trajectory prediction page provides a search input box. In response to input in the search input box, the page retrieves the input content, searches for objects matching the input content, and displays the information of the found objects, which may include trajectory point data of the object over a period of time. The page can also perform trajectory prediction on the object; specific steps can be found in the relevant steps of the foregoing embodiments and will not be repeated here.

[0108] Furthermore, in response to a click operation on a certain trajectory point, the corresponding time point and the captured image are displayed.

[0109] After obtaining the prediction results for the object, the prediction point can be displayed at the corresponding location on the map, and at least one of the probability of the trajectory point and the prediction time can be displayed.

[0110] In some embodiments, trajectory points of the target object are continuously acquired, the next actual trajectory point of the target object after the target trajectory point data is acquired, and in response to the next actual trajectory point being one of one or more predicted points, other predicted points besides the next actual trajectory point are hidden.

[0111] One possible scenario is that, in response to one or more predicted points being different from the next actual trajectory point, the suspected image of the target object in the relevant acquired images is analyzed; in response to a click operation on the predicted point, the suspected image acquired at that predicted point is displayed; and in response to a selection operation, it is determined whether the suspected image is the image of the target object.

[0112] The relevant acquired images include images acquired by the acquisition points corresponding to each prediction point within the prediction time period.

[0113] After obtaining the next actual trajectory point, this data can be used to update the real-time trajectory point data. Then, the real-time trajectory point data can be preprocessed again, and the updated data can be used for trajectory prediction once more.

[0114] In some embodiments, after a period of time, since the behavior patterns of the object may change, the updated trajectory point data of the object can be reacquired as sample data to retrain the model, so that the model can accurately reflect the current trajectory behavior characteristics of the object.

[0115] See Figure 3 , Figure 3 This is a schematic diagram of an embodiment of the trajectory prediction method of this application.

[0116] Figure 3 This is a map showing a localized area displayed by the device. The map shows existing trajectory points of the target object marked with solid dots, and can also display pre-processed trajectory points. Arrows can be used between trajectory points to indicate the object's movement direction, or numbers can be used to label the trajectory points, with increasing numbers indicating the object's movement direction. Figure 3 The example uses arrows. The two latest trajectory points A and B are used as target trajectory point data, and one or more predicted points for the next trajectory point are displayed as hollow dots, which are points C and D in this example.

[0117] In some implementation scenarios, the probability corresponding to the predicted point and the prediction time required to reach the predicted point from the latest trajectory point can also be displayed. Furthermore, the time point for reaching the predicted point or a countdown can be displayed.

[0118] See Figure 4 , Figure 4 This is a schematic diagram of the framework of an embodiment of the electronic device of this application.

[0119] The electronic device 40 includes a memory 41 and a processor 42. The processor 42 is used to execute program instructions stored in the memory 41 to implement the steps in any of the above-described trajectory prediction method embodiments. In a specific implementation scenario, the electronic device 40 may include, but is not limited to, computer equipment, electrical equipment, microcomputers, desktop computers, and servers. In addition, the electronic device 40 may also include mobile devices such as laptops and tablets, which are not limited here.

[0120] Specifically, processor 42 controls itself and memory 41 to implement the steps in any of the trajectory prediction method embodiments described above. Processor 42 can also be referred to as a CPU (Central Processing Unit). Processor 42 may be an integrated circuit chip with signal processing capabilities. Processor 42 can also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor. Furthermore, processor 42 can be implemented using integrated circuit chips.

[0121] See Figure 5 , Figure 5 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application.

[0122] The computer-readable storage medium 50 provided in this embodiment stores program instructions 51 that can be executed by a processor. When the program instructions 51 are executed by the processor, they are used to implement the steps in any of the above trajectory prediction method embodiments.

[0123] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, units or components may be combined or integrated into another subsystem, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0125] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a server, network device, etc.) or processor to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A trajectory prediction method, characterized in that, The method includes: Acquire target trajectory point data of the target object; the target trajectory point data includes multiple trajectory points collected at different time points; Feature extraction is performed on the time point data corresponding to the target trajectory point data to obtain time features; Based on the time characteristics, trajectory prediction is performed on the target trajectory point data to obtain a target prediction result, which includes one or more predicted points for the next trajectory point after the target trajectory point data.

2. The method according to claim 1, characterized in that, The time feature includes at least one of a first sub-time feature and a second sub-time feature; the first sub-time feature indicates whether the date to which the time point belongs is a weekday or a non-working day, and the second sub-time feature indicates the weekday number of the date to which the time point belongs.

3. The method according to claim 1, characterized in that, Using a target object model applicable to the target object and a general group model that does not distinguish between applicable objects, feature extraction is performed on the time point data corresponding to the target trajectory point data to obtain time features; The step of predicting the trajectory of the target trajectory point data based on the time features yields a first prediction result and a second prediction result; wherein, the target object model is trained using the first sample trajectory point data of the target object, and the group general model is trained using the second sample trajectory point data of multiple sample objects. After performing trajectory prediction on the target trajectory point data based on the time features, the method further includes: The target prediction result is determined based on the first prediction result and the second prediction result; And / or, the target prediction result further includes the prediction time period corresponding to each prediction point; the method further includes: The latest trajectory point in the target trajectory point data is combined with the predicted point to form a candidate trajectory; Arrange the trajectory durations of the reference sub-trajectories that are identical to the candidate trajectory in the first reference trajectory point data according to their time length to form a duration sequence; The upper and lower quartiles of the duration sequence are used as the predicted time periods for the corresponding prediction points.

4. The method according to claim 1, characterized in that, The trajectory point represents the collection point where the target object is located; acquiring the target trajectory point data of the target object includes: Acquire real-time trajectory point data of the target object; the real-time trajectory points in the real-time trajectory point data are arranged in chronological order. Based on the category of the collection point corresponding to the real-time trajectory point, the real-time trajectory point data is deduplicated to obtain the deduplicated real-time trajectory point data. The target trajectory point data is selected from the deduplicated real-time trajectory point data.

5. The method according to claim 4, characterized in that, The real-time trajectory point data is deduplicated based on the category of the collection point corresponding to the real-time trajectory point, resulting in the following deduplicated real-time trajectory point data: One or more sub-trajectories to be processed are determined from the real-time trajectory point data. The sub-trajectories to be processed contain two or more consecutive real-time trajectory points, and the collection points corresponding to the real-time trajectory points belong to the same category. For each of the sub-trajectories to be processed, delete all real-time trajectory points except for the first and last real-time trajectory points to obtain the deduplicated real-time trajectory point data.

6. The method according to claim 4, characterized in that, Before performing deduplication on the real-time trajectory point data based on the category of the collection point corresponding to the real-time trajectory point, and obtaining the deduplicated real-time trajectory point data, the method further includes: Each of the aforementioned collection points is taken as a target collection point. The target collection points are combined with other collection points to obtain one or more collection point combinations. The association parameters of each collection point combination are calculated and used as the clustering parameters of the target collection points. The association parameters of the collection point combination include at least one of collection point distance, collection point association degree, and collection point similarity. The collection point association degree represents the probability of the collection point combination occurring. Based on the clustering parameters of each collection point, clustering is performed on all collection points to determine the category to which each collection point belongs.

7. The method according to claim 6, characterized in that, Calculating the correlation parameters of the collection point combination includes at least one of the following steps: Based on the position information of two of the collection points in the collection point combination, calculate the distance between the collection points corresponding to the collection point combination; The number of times the collection point combination appears for the first time and the number of times the associated collection point combination appears for the second time in the second reference trajectory point data are counted. The ratio of the first number to the second number is used as the collection point correlation degree of the collection point combination. The associated collection point combination is all the collection point combinations composed of the target collection point. In the statistical analysis of the second reference trajectory point data, overlapping acquisition points are identified between adjacent acquisition points of the target acquisition point and adjacent acquisition points of other acquisition points; wherein, the previous acquisition point and the next acquisition point are respectively considered as adjacent acquisition points; for each overlapping acquisition point, a first similarity is calculated using the first acquisition point correlation degree and the second acquisition point correlation degree corresponding to the overlapping acquisition point; the first acquisition point correlation degree is the correlation degree between the overlapping acquisition point and the acquisition point corresponding to the target acquisition point, and the second acquisition point correlation degree is the correlation degree between the overlapping acquisition point and the acquisition points corresponding to other acquisition points; based on the first similarity of each overlapping acquisition point, a second similarity is calculated for the adjacent acquisition points; based on the second similarity of the previous acquisition point and the second similarity of the next acquisition point, the similarity of the acquisition points in the combination of acquisition points is calculated; and / or, Before performing clustering processing on all the collection points based on the clustering parameters of each collection point, the method further includes: The correlation of the collection points that is greater than or equal to the correlation threshold is adjusted to a first preset value, where the first preset value represents the maximum value of the correlation of the collection points. The similarity of the collection points that is greater than or equal to the similarity threshold is adjusted to a second preset value, where the second preset value represents the maximum value of the similarity of the collection points.

8. The method according to claim 1, characterized in that, The target prediction result also includes the prediction time period corresponding to each prediction point; the method further includes: Display the one or more predicted points for the next trajectory point; Obtain the next actual trajectory point of the target object after the target trajectory point data; In response to the next actual trajectory point being one of the one or more predicted points, the predicted points other than the next actual trajectory point are hidden; In response to the fact that one or more predicted points are different from the next actual trajectory point, the suspected image of the target object in the relevant acquired images is analyzed and the suspected image is displayed. In response to a selection operation, it is determined whether the suspected image is an image of the target object; wherein, the relevant acquired images include the acquired images of the acquisition point corresponding to the predicted point within the prediction time period.

9. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores program instructions that, when executed by the processor, implement the method of any one of claims 1 to 8.

10. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by the processor, they implement the method of any one of claims 1 to 8.

Images