Spatial-temporal trajectory recovery method and device based on context awareness and conditional diffusion model

Abstract

The invention provides a spatio-temporal trajectory recovery method and device based on context awareness and a conditional diffusion model, electronic equipment and a storage medium. The method comprises the steps that sparse trajectory data and multi-source spatio-temporal condition information are acquired; complementing the trajectory through a multi-step denoising process by using a conditional diffusion model; a space-time convolution module is introduced to extract space-time dependence features of the trajectory, and dynamic fusion of multi-source condition information is realized through a cross-modal attention mechanism; a condition reconstruction loss function is combined, and the space-time consistency of the model to track recovery is optimized; a layered rapid denoising mechanism is adopted, so that the calculation efficiency and the real-time performance of the model are improved; and finally generating a high-precision complemented trajectory consistent with the real trajectory. According to the method, the problems of trajectory data sparsity and non-uniform distribution can be effectively solved, the precision and robustness of trajectory recovery are remarkably improved, the method is widely applied to the fields of intelligent transportation, urban planning, personalized travel service and the like, and technical support is provided for related industries.

Description

Technical Field

[0001] The present invention relates to the technical field of trajectory recovery and spatio-temporal data mining, and particularly to a spatio-temporal trajectory recovery method and device based on context awareness and conditional diffusion model. Background Art

[0002] This section aims to provide background or context for the embodiments of the present invention stated in the claims. The description herein is not admitted to be prior art merely by virtue of being included in this section.

[0003] Trajectory recovery technology has important applications in intelligent transportation, urban planning, and personalized travel services. By analyzing and completing sparse trajectory data, more comprehensive and accurate travel predictions and optimization schemes can be provided. However, traditional trajectory recovery methods face various challenges.

[0004] Existing trajectory recovery methods usually rely on statistical models or deep learning techniques, and these methods have limitations in dealing with complex multimodal data. For example, although statistical models have high computational efficiency, it is difficult to capture the non-linear dynamic patterns of trajectory data; although deep learning methods can extract spatio-temporal features, there are still significant deficiencies in accuracy and robustness when facing the non-uniform distribution, sparsity, and context information fusion of trajectory data.

[0005] In recent years, diffusion models have been introduced into the trajectory completion task due to their powerful generation ability, and can recover data conforming to the trajectory distribution from a disordered state by gradually denoising. However, diffusion models still face problems such as low computational efficiency, insufficient conditional control, and poor spatio-temporal consistency in practical applications. For example, in scenarios with large amounts of trajectory data or high real-time requirements, the multi-step denoising process of traditional diffusion models requires a large amount of computational resources. In addition, existing methods are difficult to fully combine multi-source conditional information (such as road networks, points of interest, and weather conditions), resulting in the generated trajectories not matching the actual scenarios.

[0006] To solve the above problems, the present invention proposes a spatio-temporal trajectory recovery method based on context awareness and conditional diffusion model. By introducing a spatio-temporal convolution module, a cross-modal attention mechanism, and a conditional reconstruction loss function, the accuracy and robustness of trajectory completion are significantly improved, and at the same time, the computational efficiency is optimized through a hierarchical fast denoising mechanism. The present invention has broad application prospects in fields such as intelligent transportation and personalized travel services. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to propose a spatio-temporal trajectory recovery method, device, electronic device, storage medium, and program product based on context awareness and conditional diffusion model, which can at least solve some technical problems in the prior art to a certain extent.

[0008] For the above purposes, a first aspect of an exemplary embodiment of the present invention provides a spatio-temporal trajectory recovery method based on context awareness and conditional diffusion models, the method comprising:

[0009] Obtain sparse trajectory data and multi-source spatio-temporal conditional information, combine with the conditional diffusion model, and recover the sparse trajectory through a hierarchical fast denoising mechanism, and preliminarily complete and gradually optimize the trajectory data in stages to generate a high-quality trajectory completion result;

[0010] Use a spatio-temporal convolutional module to extract the spatio-temporal features of the trajectory, and capture the temporal order and spatial distribution law of the trajectory points;

[0011] Combine a cross-modal attention mechanism to dynamically fuse multi-source spatio-temporal conditional information (such as road network, points of interest, and weather conditions, etc.) to enhance the accuracy of trajectory completion;

[0012] Based on the conditional reconstruction loss function, optimize the spatio-temporal consistency of the generated trajectory, and improve the spatial continuity, temporal logical rationality, and matching degree with external conditional information of the trajectory;

[0013] Finally, output a high-precision trajectory completion result that meets multi-source condition constraints, which can be widely applied to fields such as intelligent transportation, urban planning, and personalized travel services.

[0014] Based on the same inventive concept, a second aspect of an exemplary embodiment of the present invention provides a spatio-temporal trajectory recovery device based on context awareness and conditional diffusion models, comprising:

[0015] A spatio-temporal feature extraction module for extracting the time series features and spatial distribution features of trajectory data; a UNet denoising module for completing the missing points in the trajectory by means of step-by-step denoising;

[0016] A conditional denoising process module for optimizing the denoising process of trajectory recovery by combining multi-source spatio-temporal conditional information;

[0017] A training and optimization module for optimizing the accuracy and robustness of the trajectory recovery model based on the conditional reconstruction loss function;

[0018] A hierarchical fast iteration module for iteratively removing the noise of the trajectory data in stages and gradually refining the trajectory completion result;

[0019] A trajectory recovery module for generating and outputting a high-precision completed trajectory that meets multi-source condition constraints.

[0020] Based on the same inventive concept, a third aspect of an exemplary embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program realizes the method described in the first aspect when executed by the processor.

[0021] Based on the same inventive concept, a fourth aspect of the exemplary embodiments of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method described in the first aspect.

[0022] Based on the same inventive concept, a fifth aspect of the exemplary embodiments of the present invention provides a computer program product including computer program instructions that, when run on a computer, cause the computer to execute the method described in the first aspect.

[0023] As can be seen from the above, through steps such as trajectory hierarchical denoising, spatio-temporal feature extraction, conditional information dynamic fusion, and model optimization, the present invention significantly improves the accuracy and robustness of trajectory recovery, overcomes problems such as insufficient processing of trajectory data sparsity, limited utilization of conditional information, and low computational efficiency in the prior art, and provides strong technical support for fields such as intelligent transportation, personalized travel services, and urban planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] To more clearly illustrate the technical solutions in the present invention, the following briefly introduces the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0025] Figure 1 FIG. is a schematic diagram of an application scenario of the spatio-temporal trajectory recovery method based on context awareness and conditional diffusion model provided by an embodiment of the present invention;

[0026] Figure 2 FIG. is a schematic flowchart of the spatio-temporal trajectory recovery method based on context awareness and conditional diffusion model provided by an embodiment of the present invention;

[0027] Figure 3 FIG. is a schematic structural diagram of a spatio-temporal trajectory recovery device based on context awareness and conditional diffusion model provided by an embodiment of the present invention;

[0028] Figure 4 FIG. is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] It can be understood that before using the technical solutions disclosed in the embodiments of the present application, the types, usage scopes, usage scenarios, etc. of the personal information involved in the present application should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0030] For example, when responding to an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested by the user will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, an application program, a server, or a storage medium that performs the operations of the technical solution of this application based on the prompt message.

[0031] As an optional but non-limiting implementation manner, the manner of sending a prompt message to the user in response to receiving an active request from the user may, for example, be in the form of a pop-up window, and the prompt message may be presented in text in the pop-up window. In addition, the pop-up window may also carry selection controls for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0032] It can be understood that the above processes of notifying and obtaining user authorization are merely illustrative and do not limit the implementation manners of this application. Other manners that comply with relevant laws and regulations can also be applied to the implementation manners of this application.

[0033] It can be understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of the corresponding laws, regulations, and related provisions.

[0034] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present invention, rather than limiting the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to convey the scope of the present invention completely to those skilled in the art.

[0035] In this article, it should be understood that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0036] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present invention should have the ordinary meanings understood by those of ordinary skill in the art to which the present invention pertains. The terms "first", "second" and similar words used in the embodiments of the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. Terms such as "upper", "lower", "left", "right" are only used to indicate relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly. The article "a" or "an" before an element does not exclude the existence of multiple such elements.

[0037] Next, with reference to several representative embodiments of the present invention, the principles and spirit of the present invention will be elaborated in detail.

[0038] Existing trajectory recovery technologies often face many challenges when dealing with the sparsity of trajectory data, non-uniform distribution, and insufficient fusion of multi-source conditional information. Uneven time intervals, sparse trajectory points, and insufficient extraction and utilization of context information in trajectory data make the accuracy and robustness of traditional methods for filling in trajectories relatively low. Especially when the trajectory distribution is sparse, the environment is complex, or there are multi-modal information (such as road networks, points of interest, and weather conditions), the effect of the trajectory filling task is significantly affected.

[0039] The inventors of the present invention have found that based on the step-by-step denoising generation mechanism of the conditional diffusion model, sparse trajectories can be effectively filled in by means of hierarchical fast denoising, and optimized by combining multi-source spatio-temporal conditional information, thereby significantly improving the accuracy and consistency of trajectory filling. By introducing a context-aware mechanism, multi-source conditional information (such as road networks, points of interest, and weather conditions) can be dynamically fused, thereby enhancing the accuracy and generalization ability of trajectory filling. In addition, by using spatio-temporal convolutional modules and cross-modal attention mechanisms, the spatio-temporal features of trajectory data and the correlation between different modal data can be fully exploited to ensure that the trajectory filling results meet the requirements of spatio-temporal consistency.

[0040] To overcome the limitations in the prior art, the present invention proposes a new trajectory recovery method based on context awareness and conditional diffusion models. This method first gradually completes the missing points in the sparse trajectory through the hierarchical fast denoising mechanism of the conditional diffusion model, and uses a spatio-temporal convolutional module to extract the time series features and spatial distribution features of the trajectory data; subsequently, it fuses multi-source spatio-temporal conditional information through a cross-modal attention mechanism to dynamically optimize the trajectory completion process; finally, it optimizes the spatial continuity and temporal logical consistency of the trajectory through a conditional reconstruction loss function, and outputs a high-precision trajectory completion result. The method of the present invention has strong practicability and robustness, and has broad application prospects especially in the fields of intelligent transportation, urban planning, and personalized travel services.

[0041] After introducing the basic principle of the present invention, the various non-limiting embodiments of the present invention will be specifically introduced below.

[0042] Reference Figure 1 , which is a schematic diagram of an application scenario of the spatio-temporal trajectory recovery method based on context awareness and conditional diffusion models provided by an exemplary embodiment of the present invention. In this application scenario, it includes a sparse trajectory data collection device 101, a trajectory processing server 102, a data storage system 103, and a user device 104. Among them, the sparse trajectory data collection device 101, the trajectory processing server 102, the data storage system 103, and the user device 104 can all be connected through a wired or wireless trajectory data communication network to achieve data interaction.

[0043] The sparse trajectory data collection device 101 can be a device near the user side for collecting trajectory point data, including but not limited to smartphones, in-vehicle GPS devices, smart wearable devices, tablets, or other electronic devices with positioning functions. These devices provide sparse input data for trajectory recovery by collecting real-time trajectory points. The collected trajectory data can include timestamps, location points, and other attribute information related to the trajectory.

[0044] Both the trajectory processing server 102 and the data storage system 103 can be independent physical servers, or a server cluster or distributed system composed of multiple physical servers, or a cloud computing device providing cloud services. The server 102 is responsible for preprocessing, hierarchical denoising, spatio-temporal feature extraction, and trajectory completion of the collected trajectory data. In the method of the present invention, the server 102 uses a conditional diffusion model to gradually denoise the trajectory data, and combines a cross-modal attention mechanism to dynamically fuse multi-source spatio-temporal conditional information, such as road networks, points of interest, and weather conditions, so as to generate a high-precision trajectory completion result.

[0045] The data storage system 103 is used to store multi-source conditional data required for trajectory recovery and related historical trajectory information. Its functions include providing road network information (such as road topology, lane information, and speed limit information), point-of-interest information (such as store locations, gas station locations), and weather information (such as rainfall, temperature, etc.). At the same time, a large amount of sample data for training the trajectory completion model is also stored in the data storage system 103. These data can include sparse trajectory points and their completed trajectory results, which are used to improve the learning ability and completion accuracy of the model.

[0046] The user device 104 can be a smartphone, a tablet computer, or other electronic devices that support multimedia interaction. Through the device 104, the user can receive the completed trajectory results sent by the trajectory processing server 102 in real time and view or analyze the results according to needs. For example, the user can use the completed trajectory for travel planning, traffic flow optimization, or personalized travel services.

[0047] In this embodiment, the sparse trajectory data acquisition device 101 sends the trajectory data to the trajectory processing server 102 through the trajectory data communication network. The server 102 combines the multi-source spatio-temporal conditional information provided by the data storage system 103 to complete the trajectory data, generates a high-precision trajectory completion result, and sends it to the user device 104 for visual display or service invocation. Through the above cooperation, the spatio-temporal trajectory recovery method provided by the present invention can effectively address the sparsity and non-uniform distribution problems of trajectory data, providing technical support for applications such as intelligent transportation, urban planning, and personalized travel services.

[0048] It should be noted that the above application scenarios are only shown for the convenience of understanding the spirit and principle of the present invention, and the embodiments of the present invention are not limited in this regard. On the contrary, the embodiments of the present invention can be applied to any applicable scenario.

[0049] Reference Figure 2 , the trajectory recovery method based on context awareness and conditional diffusion model is applied to the trajectory processing server, and this method includes the following steps:

[0050] Step S210, collect sparse trajectory data and related multi-source spatio-temporal conditional information.

[0051] In this embodiment, the sparse trajectory data is collected by trajectory sampling devices (such as in-vehicle GPS, smartphones, etc.) and contains a series of discontinuous trajectory point data. Each trajectory point consists of a timestamp and spatial position coordinates, usually represented as where T i =(x i ,y i ,t i ) represents the longitude and latitude coordinates of the trajectory point (x i, y i ), and the timestamp t i . Due to the sampling interval and environmental limitations, these trajectory points usually have problems of sparsity and uneven distribution.

[0052] To improve the accuracy of trajectory completion, the present invention combines multi-source spatio-temporal condition information while collecting sparse trajectory data including road network information, point-of-interest information, and weather condition information. Specifically, the road network information provides the road topology, lane information, and speed limit information around the trajectory points; the point-of-interest information includes the location, type, and distance from the trajectory points of the facilities near the trajectory points; the weather condition information describes the meteorological information at the moment when the trajectory points are located, such as temperature, rainfall, and wind speed.

[0053] Through the integration of the above data, an input set is constructed This set contains both the spatio-temporal information of the sparse trajectory points and the external multi-source condition information, providing comprehensive data support for the subsequent trajectory denoising and completion processes. In this step, the joint collection of sparse trajectory data and multi-source condition information can effectively solve the problem of data isolation and lay a solid foundation for subsequent modeling and trajectory completion.

[0054] Step S220: Use the spatio-temporal convolution module to extract features from the trajectory data and fuse spatio-temporal control information.

[0055] Trajectory spatio-temporal feature extraction

[0056] Trajectory data has significant temporal and spatial dependencies. To capture the spatio-temporal features of the trajectory data, this step first preprocesses the trajectory data and rasterizes it into discrete grid cells. Subsequently, convolutional operations are used to extract multi-level spatio-temporal features from the trajectory grid data. One-dimensional convolution is used to capture the short-term dynamic features in the time dimension, and two-dimensional convolution is used to capture the spatial distribution patterns. Through this convolution mechanism, the local and global features of the trajectory points can be obtained, providing a basis for feature fusion.

[0057] Multi-source feature fusion of cross-modal attention mechanism

[0058] To fully explore the dynamic association between the trajectory data and the external environmental data, this step introduces a cross-modal attention mechanism to combine the trajectory features with the environmental features. This mechanism can capture the interaction relationship between the two types of features, thereby realizing the effective fusion of multi-source features.

[0059] The core formula of the cross-modal attention mechanism is as follows:

[0060]

[0061] Among them, Q T , KB , V E respectively represent the query of the trajectory, the key of the environment, and the value of the environment. Q E , K T , V T respectively represent the query of the environment, the key of the trajectory, and the value of the trajectory. Through this cross-modal attention mechanism, the model can learn the interdependent relationship between trajectory data and environmental information.

[0062] The cross-modal attention mechanism also includes a self-attention mechanism for capturing the relationships within each modality. Specifically, the self-attention mechanism calculates the correlation within the trajectory information or environmental information through the following formula:

[0063]

[0064] These self-attention modules help the model further understand the spatio-temporal dependence relationship of the trajectory and the environment.

[0065] Feature Fusion and Unified Representation Generation

[0066] To further capture the temporal information in the trajectory data, the features fused by the cross-modal attention mechanism are input into a bidirectional long short-term memory network (BiLSTM). BiLSTM can capture the temporal dependence of the trajectory data from both the forward and backward directions simultaneously. This process helps to strengthen the model's understanding of the time context and provides more comprehensive temporal information for the subsequent denoising process. Finally, these features will be further processed by a multi-layer perceptron (MLP) to generate conditional information for input to the subsequent UNet denoising process.

[0067] Step S230: Use the spatio-temporal convolution module to extract spatio-temporal features from the trajectory data and fuse spatio-temporal control information;

[0068] In this embodiment, the spatio-temporal convolution module is used to extract features from multi-source spatio-temporal control information (including trajectory time series, road network, meteorological data, and point-of-interest information) and optimize the denoising process of the trajectory data through an accurate control mechanism. This module combines multi-source conditional input, two-dimensional convolution operations, and residual connections, thus significantly improving the accuracy and consistency of trajectory completion.

[0069] The spatio-temporal convolution module starts with the input feature map and first extracts hidden state features through a series of two-dimensional convolution operations The specific operations are as follows:

[0070] X hid = Conv2D(X in )

[0071] where Conv2D represents the two-dimensional convolution operation, Lin is the spatial resolution of the feature map, C in and C hid are the number of channels of the input and hidden states respectively.

[0072] Next, the spatio-temporal control information is dynamically adjusted by fusing the positional embedding of the diffusion step number n and the attention mechanism of multi-source control information. Specifically, the multi-source conditional information {C 1 , C2, ..., Ck} is fused through the attention mechanism, and the calculation formula is:

[0073] X′hid = Xhid + FCCond(PE(n) + Attention(C 1 , ..., C k ))

[0074] where FCCond(·) is the fully connected layer; PE(n) represents the positional embedding of the diffusion step number n; Attention(·) represents the attention fusion operation on the multi-source control information.

[0075] The hidden state feature X′hid after the above adjustment is further processed through a two-layer two-dimensional convolutional network, and combined with the activation function and residual connection to generate the output feature. The calculation process is as follows:

[0076] Xout = Conv2D(σ(Conv2D(X′hid)))

[0077] X′out = X out + ResConv(X in )

[0078] where σ(·) represents the activation function; ResConv(·) represents the two-dimensional convolutional operation of the residual connection.

[0079] The final output feature is passed to the subsequent module in the denoiser. For the downsampling module, usually set C out = 2·C in ; for the upsampling module, then set

[0080] Step S240, use the conditional diffusion model combined with the hierarchical fast denoising mechanism to perform preliminary denoising and fine completion on the sparse trajectory;

[0081] In this embodiment, the conditional diffusion model is used to denoise and complete the sparse trajectory data, and the hierarchical fast denoising mechanism is combined to achieve the efficiency and accuracy of the trajectory completion process. The design of the hierarchical fast denoising mechanism is divided into two stages: coarse-grained denoising and refined completion, which can gradually optimize the spatial position and temporal correlation of the trajectory points.

[0082] In the conditional diffusion model, the denoising process is based on generating a noise distribution by perturbing the trajectory point data through the forward diffusion process. The mathematical expression of the forward diffusion is:

[0083]

[0084] where, represents the true trajectory. α t is a time scheduling parameter related to the diffusion step t; ∈ is Gaussian noise.

[0085] During the reverse denoising process, the conditional diffusion model gradually reconstructs the distribution of trajectory points through multi-source conditional information (including road network, points of interest, weather, etc.). The formula for the reverse process is:

[0086]

[0087] where, is the multi-source conditional information, and μ θ and Σ θ are the mean and variance predicted by the model, respectively.

[0088] To further optimize the efficiency and accuracy of the diffusion model, this step introduces a hierarchical fast denoising mechanism, which gradually restores the true distribution of trajectory points through two-stage processing:

[0089] Low-level fast coarse-grained denoising:

[0090] In the low-level stage, the input trajectory features are mapped to a low-dimensional space through a simplified representation. For example, a large convolutional kernel or a linear transformation is designed to reduce the trajectory feature dimension while capturing the global structure and main noise features in the trajectory data. The simplified trajectory feature is represented as:

[0091] X low =Conv2Dlarge(X input )

[0092] where, Conv2Dlarge is a two-dimensional convolution operation with a large convolutional kernel, and X input is the input trajectory feature.

[0093] Based on the low-dimensional feature space, a fast denoising algorithm (such as an efficient Gaussian filtering variant) is used to preliminarily remove the significant noise in the trajectory data:

[0094] X′ low =X low -∈ low

[0095] where, ∈ lowRepresents the noise removed during the low-level fast coarse-grained denoising process. In this stage, combined with GPU parallel computing technology, the trajectory data is divided into multiple sub-regions for fast parallel processing, thus significantly improving the processing efficiency.

[0096] High-level fine-tuning denoising:

[0097] After the low-level denoising is completed, the denoising result is passed to the high level for fine-tuning. In this stage, the adaptive adjustment mechanism dynamically optimizes the denoising parameters according to the characteristics of the remaining noise. Specifically, the neural network module analyzes the low-level denoising result to predict the appropriate denoising parameters:

[0098] θ adaptive = Predictor(X′ low )

[0099] where Predictor is the parameter prediction module, and θ adaptive is the adaptive denoising parameter.

[0100] High-level fine-tuning denoising further optimizes the spatio-temporal distribution of trajectory points through the conditional diffusion model, eliminates the subtle noise remaining after low-level denoising, and restores the trajectory details:

[0101] X′ high = X′ low - ∈ high (θ adaptive )

[0102] where ∈ high represents the high-level noise prediction function.

[0103] The hierarchical fast denoising mechanism further optimizes the result through multiple iterations. In each iteration, the result generated by the low-level fast denoising is used as the input for the high-level fine-tuning denoising. After being optimized by the high-level denoising, it is returned to the low level for a new round of coarse-grained processing. This iterative process can gradually optimize the denoising result, enabling the model to approach the optimal solution in fewer diffusion steps. The denoising formula during the iterative process is as follows:

[0104] X t-1 = X t - ∈ low (X t ) - ∈ high (X t , θ adaptive )

[0105] By reasonably designing the number of iterations and hierarchical parameters, the model reaches a convergent state of the denoising effect after a small number of iterations.

[0106] Step S250: Optimize the trajectory recovery model based on the conditional reconstruction loss function to ensure the rationality of the generated trajectory under spatiotemporal consistency and multi-source condition constraints;

[0107] In this embodiment, by designing the conditional reconstruction loss function, the model optimizes the denoising process to ensure that the generated trajectory is consistent with the real trajectory in space and time. The design of the loss function combines the noise matching loss and the conditional reconstruction loss, and through joint optimization, the denoising ability of the model and the accuracy of trajectory recovery are improved.

[0108] Calculation of conditional reconstruction loss

[0109] The conditional reconstruction loss is used to measure the difference between the recovered trajectory and the real trajectory to ensure the spatiotemporal consistency of the trajectory data. This loss function is defined by the following formula:

[0110] L recon =||T true -T decoder || 2

[0111] where T true represents the real trajectory; T decoder represents the recovered trajectory generated by the conditional decoder.

[0112] By minimizing L recon , the model can effectively utilize multi-source condition information to ensure that the recovered trajectory is consistent in spatial structure and time series and conforms to the constraints of the actual environment.

[0113] Calculation of noise matching loss

[0114] The noise matching loss is used to optimize the model's ability to predict noise and ensure that the model can accurately restore trajectory data from the noise in the diffusion state. Its calculation formula is as follows:

[0115] L n =||∈-∈ θ (X n ,n,STcond)|| 2

[0116] where ∈ represents the real noise, ∈ θ (X n ,n,STcond) represents the noise predicted by the model; X n represents the trajectory data at diffusion step n; STcond represents multi-source spatio-temporal condition information.

[0117] By minimizing L n , the model can improve the accuracy of noise prediction and provide higher-quality input for the denoising process.

[0118] Design of the combined loss function

[0119] To achieve the global optimization of the denoising process, the model adopts a combined loss function L total , which combines the noise matching loss and the conditional reconstruction loss. The expression of the combined loss function is:

[0120] L total (θ) = L n (θ) + λL recon (θ)

[0121] where λ is a hyperparameter used to balance the weights of the noise matching loss and the conditional reconstruction loss.

[0122] By minimizing L total , the model can achieve a balance between noise removal and trajectory completion, thereby generating trajectory data with higher spatio-temporal consistency and multi-source condition adaptability.

[0123] Step S260: Generate a high-precision trajectory completion result, perform consistency verification, and then output trajectory data that meets the requirements of intelligent transportation and personalized travel services.

[0124] In this embodiment, the trajectory completion result generated by the trajectory recovery model undergoes a series of post-processing operations to ensure the spatial and temporal consistency of the trajectory data, and finally outputs high-precision trajectory data that meets the actual application requirements.

[0125] Restoration of rasterized trajectories

[0126] During the trajectory completion process, the trajectory points generated by the model are usually represented in a rasterized form, that is, the trajectory points are mapped to discrete raster cells in space. To restore the continuity of the trajectory, this step restores the rasterized representation to the coordinates of the actual trajectory points.

[0127] Through the mapping relationship between the trajectory points and the raster cells, the raster identifier of each trajectory point is converted into the real spatial coordinates. The restored trajectory points have a continuous spatial distribution, can accurately reflect the user's movement path, and provide real and reliable trajectory data for subsequent analysis and applications.

[0128] Time series restoration

[0129] The trajectory completion result not only requires spatial continuity but also needs to ensure the correctness of the time series. Since there may be time jumps or disorder in the completion process, this step corrects the time information of the generated trajectory points.

[0130] Time series recovery First, sort the timestamps of all trajectory points to ensure that the trajectory points are arranged in chronological order. For detected time jumps or reverse order cases, correct these problems by adjusting the timestamps so that the chronological order of the trajectory points conforms to the actual travel pattern, generating a complete and logically coherent time series trajectory.

[0131] Output and storage of trajectory data

[0132] After completing the trajectory recovery and time series correction, the final trajectory data needs to be output and stored. The recovered trajectory data is usually stored in the form of a time series, including the spatial coordinates (such as longitude and latitude) and timestamp information of each trajectory point.

[0133] According to the requirements of different applications, the trajectory data can be stored in different formats. For example, in map applications, the trajectory data can be saved in formats such as GeoJSON and Shapefile; in trajectory analysis, the trajectory data can be stored as a time series dataset. During the storage process, relevant metadata can also be generated for each trajectory, such as the start time, end time, and length of the trajectory. These metadata are of great value for subsequent analysis, query, and visualization operations.

[0134] Reference Figure 3 , which is a schematic structural diagram of the trajectory recovery device provided by an exemplary embodiment of the present invention. The device includes a spatio-temporal feature extraction module 310, a UNet denoising module 320, a conditional denoising process module 330, a hierarchical iterative denoising module 340, a training and optimization module 350, and a trajectory recovery output module 360.

[0135] The spatio-temporal feature extraction module 310 is configured to extract the spatio-temporal features of the trajectory from the input sparse trajectory data and multi-source spatio-temporal condition information (such as road network, points of interest, and weather conditions). Specifically, this module uses a spatio-temporal convolutional network (ST-Conv) to deeply model the local time series and spatial distribution characteristics of the trajectory points, captures the local dynamic changes of the trajectory through convolutional kernels, and further extracts the global time dependence relationship by combining bidirectional LSTM. The spatio-temporal features generated by the spatio-temporal feature extraction module 310 not only include the spatial positions of the trajectory points, but also comprehensively include their dynamic change patterns and context information, providing accurate feature inputs for the subsequent denoising modules.

[0136] The UNet denoising module 320 is configured to initially complete the sparse trajectory data in a hierarchical and step-by-step denoising manner. This module adopts the classic UNet architecture and extracts different-scale features of the trajectory data through multiple layers of encoding and decoding processes. During the encoding process, the sparse trajectory data is compressed and deep features are extracted; during the decoding process, these features are gradually restored to the completed trajectory. Through the skip connection technology, the UNet denoising module can retain the detailed features of the trajectory during the denoising process. In addition, this module embeds conditional information (such as road topology and point-of-interest features) to ensure that the completed trajectory conforms to the external environmental information.

[0137] The conditional denoising process module 330 is one of the key modules in the entire trajectory recovery system and is configured to dynamically combine multi-source spatio-temporal conditional information to optimize the denoising process. Specifically, this module assigns weights to conditional information of different modalities (such as road network, point of interest, and weather) through a cross-modal attention mechanism, and dynamically adjusts the contribution ratio of each condition. For example, when there is a complex road structure near the trajectory, the module will increase the weight of the road network information to improve the accuracy of trajectory completion. The conditional denoising process module 330 also maintains the spatio-temporal consistency of the trajectory during the denoising process by introducing a conditional reconstruction loss function, ensuring that the completed trajectory matches the real-scene conditions.

[0138] The hierarchical iterative denoising module 340 is configured to perform multiple rounds of optimization on the trajectory data through a coarse-to-fine hierarchical strategy. Specifically, in the initial stage, a fast algorithm is used to remove most of the noise in the trajectory, significantly improving the denoising efficiency; subsequently, through refined iteration, the module locally optimizes the remaining noise points and gradually adjusts the spatio-temporal distribution of the trajectory. In each round of iteration, the hierarchical iterative denoising module 340 combines the multi-source spatio-temporal conditional information provided by the conditional denoising process module 330 to further optimize the spatio-temporal features of the trajectory, ensuring that the finally generated trajectory has high spatio-temporal consistency.

[0139] The training and optimization module 350 is configured to optimize each step of the trajectory recovery by minimizing the joint loss function during the model training phase. The loss function includes the denoising error loss and the conditional consistency loss, which are used to optimize the denoising process and the utilization efficiency of the conditional information respectively. To improve the adaptability of the model in different scenarios, the training and optimization module 350 adopts a data augmentation strategy, such as randomly sparsifying, translating, or rotating the trajectory data to simulate the trajectory data distribution in various complex environments. In addition, this module optimizes the computational efficiency of the model through a distributed training method, ensuring that the system can process large-scale trajectory data.

[0140] The trajectory recovery output module 360 is configured to convert the denoised trajectory data into a directly applicable sequence of trajectory points. Before output, this module performs time logic correction and spatial consistency optimization on the trajectory, such as correcting abnormal intervals between timestamps or sudden offsets between trajectory points. The optimized sequence of trajectory points is generated as a complete trajectory completion result and output in a standardized format for direct application in scenarios such as path planning, traffic management, or personalized travel services in intelligent transportation.

[0141] Figure 4 FIG. shows a more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. Among them, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other inside the device through the bus 1050.

[0142] The processor 1010 can be implemented in ways such as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0143] The memory 1020 can be implemented in forms such as a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 1020 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1020 and called and executed by the processor 1010.

[0144] The input / output interface 1030 is used to connect to the input / output module to achieve information input and output. The input / output module can be configured as a component in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Among them, the input device can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device can include a display, a speaker, a vibrator, an indicator light, etc.

[0145] The communication interface 1040 is used to connect to a communication module (not shown in the figure) to achieve communication and interaction between this device and other devices. The communication module can communicate through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0146] The bus 1050 includes a path for transmitting information between various components of the device (such as the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0147] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device may also only include the components necessary to implement the solution of the embodiments of this specification, and do not necessarily include all the components shown in the figure.

[0148] The electronic device of the above embodiment is used to implement the corresponding mapless relocalization method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0149] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the mapless relocalization method as described in any of the above embodiments.

[0150] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0151] The above non-transitory computer-readable storage medium can be any available medium or data storage device accessible by a computer, including but not limited to magnetic memories (such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical memories (such as CDs, DVDs, BDs, HVDs, etc.), and semiconductor memories (such as ROMs, EPROMs, EEPROMs, non-volatile memories (NAND FLASH), solid-state drives (SSD)), etc.

[0152] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the mapless repositioning method described in any one of the above exemplary method embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0153] Based on the same inventive concept, corresponding to the mapless repositioning method described in any of the above embodiments, the present invention also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of the computer to cause the computer and / or the processor to execute the mapless repositioning method. Corresponding to the execution subjects of the respective steps in the respective embodiments of the mapless repositioning method, the processors executing the corresponding steps can belong to the corresponding execution subjects.

[0154] The computer program product of the above embodiments is used to cause the computer and / or the processor to execute the mapless repositioning method described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0155] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, a method, or a computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software, generally referred to herein as "circuit", "module", or "system". In addition, in some embodiments, the present invention can also be implemented in the form of a computer program product in one or more computer-readable media, which contain computer-readable program codes.

[0156] Any combination of one or more computer-readable media may be employed. The computer-readable media may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium may include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program which can be used by or in connection with an instruction execution system, apparatus, or device.

[0157] A computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which computer-readable program code is carried. Such a propagated data signal may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0158] The program code contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0159] The computer program code for performing the operations of the present invention may be written in one or more programming languages or combinations thereof, including object-oriented programming languages such as Python, Smalltalk, C++, and also including conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network (including a local area network (LAN) or a wide area network (WAN)), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0160] It should be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program instructions executed by the computer or other programmable data processing apparatus create means for implementing the functions / operations specified in the blocks of the flowchart and / or block diagram.

[0161] These computer program instructions can also be stored in a computer-readable medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable medium produce an article of manufacture including instruction means for implementing the functions / operations specified in the blocks of the flowchart and / or block diagram.

[0162] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process such that the instructions executed on the computer or other programmable apparatus provide a process for implementing the functions / operations specified in the blocks of the flowchart and / or block diagram.

[0163] In addition, although the operations of the method of the present invention are depicted in the drawings in a particular order, this is not required or implied to perform the operations in that particular order, or to perform all of the illustrated operations to achieve the desired result. On the contrary, the steps depicted in the flowchart may be performed in a different order. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step and performed, and / or one step may be decomposed into multiple steps and performed.

[0164] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present application. Each block in the flowchart or block diagram may represent a module, a segment of code, or a portion thereof that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than noted in the figures. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functionality involved. It should also be noted that each block in the block diagram or flowchart, and combinations of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0165] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0166] Those of ordinary skill in the art should understand that: the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the present application (including the claims) is limited to these examples; under the concept of the present application, the technical features between the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of the present application as described above, and for the sake of brevity, they are not provided in detail.

[0167] In addition, for simplicity of description and discussion, and in order not to make the embodiments of the present application difficult to understand, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Further, the device may be shown in block diagram form in order to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of such block diagram devices are highly dependent on the platform on which the embodiments of the present application are to be implemented (i.e., these details should be fully within the understanding of those skilled in the art). In cases where specific details (such as circuits) are set forth to describe exemplary embodiments of the present application, it will be apparent to those skilled in the art that the embodiments of the present application can be implemented without these specific details or with variations of these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0168] Although the present application has been described in connection with specific embodiments of the present application, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art based on the foregoing description. For example, other memory architectures (such as dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0169] The embodiments of the present application are intended to cover all such alternatives, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the embodiments of the present application shall be included within the protection scope of the present application.

[0170] Although the spirit and principles of the present invention have been described with reference to several specific embodiments, it should be understood that the present invention is not limited to the specific embodiments disclosed, and the division of each aspect does not mean that the features in these aspects cannot be combined for benefit. Such division is only for convenience of expression. The present invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. The scope of the appended claims is to be accorded the broadest interpretation so as to encompass all such modifications and equivalent structures and functions.

Claims

1. A spatiotemporal trajectory recovery method based on context awareness and conditional diffusion model, characterized in that: include: Acquire sparse trajectory data and multi-source spatiotemporal condition information; The conditional diffusion model is used to complete the sparse trajectory data through a hierarchical fast denoising mechanism; Use the spatiotemporal convolution module to extract the spatiotemporal features of trajectory data; Dynamically integrate multi-source spatiotemporal condition information based on cross-modal attention mechanism; The conditional reconstruction loss function is used to optimize the spatiotemporal consistency of trajectory recovery; Finally, a high-precision completion trajectory that meets multi-source constraints is generated.

2. The method according to claim 1, characterized in that The obtaining of sparse trajectory data and multi-source spatiotemporal condition information includes: Get a sparse trajectory point sequence of trajectory data; Obtain road network information, including road topology, lane information, and speed limit information; Get information about points of interest, including their location, type, and distance from the track; Get information on weather conditions including temperature, precipitation, and visibility.

3. The method according to claim 1, characterized in that The hierarchical fast denoising mechanism includes: Low-level fast coarse-grained denoising: In the initial stage, the feature compression method is used to map the input trajectory features to a low-dimensional space through large convolution kernel convolution or linear transformation processing, and then a fast denoising algorithm such as an efficient Gaussian filter variant is used to remove significant noise; High-level fine-tuning denoising: After low-level denoising is completed, the results are passed to the high-level. Through the adaptive mechanism, according to the residual noise characteristics, the neural network module analyzes and predicts the adaptive denoising parameters, and then uses the conditional diffusion model to optimize the trajectory spatiotemporal distribution, eliminate subtle noise, refine the completion results, and balance efficiency and accuracy. Iterative optimization: This mechanism is iterated multiple times, and each time the low-level denoising result is used as the high-level input. After the high-level optimization, it returns to the low-level for a new round of processing. By reasonably designing the number of iterations and level parameters, the model converges after a small number of iterations, and finally finely optimizes the trajectory data and outputs a high-precision completed trajectory.

4. The method according to claim 1, characterized in that The spatiotemporal convolution module extracts the spatiotemporal features of the trajectory in the following way: Design a specific convolution kernel size and step size to capture the local time series characteristics and spatial distribution characteristics of trajectory data; By combining local features with global features through multi-layer convolutional networks, the model's ability to understand dynamic changes in trajectories is improved; The consistency of time order is maintained in the convolution operation to ensure that the extracted features can reflect the temporal correlation and spatial distribution of trajectory points.

5. The method according to claim 1, characterized in that The cross-modal attention mechanism fuses multi-source spatiotemporal condition information in the following way: Feature encoding of multi-source spatiotemporal condition information to generate corresponding embedding vectors; Dynamically assign modal feature weights, adjust weights according to the importance of conditional information, and enhance the contribution of key conditions to trajectory completion; Integrate feature information from different modalities and strengthen the correlation between trajectory points and multi-source information through a cross-modal attention mechanism; Maintaining spatiotemporal consistency during the fusion process improves the accuracy and robustness of trajectory completion.

6. The method according to claim 1, characterized in that The conditional reconstruction loss function optimizes the spatiotemporal consistency of trajectory recovery in the following way: The spatial continuity of the trajectory points is integrated to optimize the smoothness of the generated trajectory and ensure that the trajectory points are reasonably distributed in space; Consider the time sequence of trajectory points, optimize the time logic consistency of generated trajectories, and avoid time jumps of trajectory points; Fusion of multi-source spatiotemporal information to ensure the consistency of generated trajectories with external information such as road networks, points of interest, and weather conditions; Minimize the difference between the generated trajectory and the true trajectory to improve the accuracy and robustness of trajectory completion.

7. A spatiotemporal trajectory restoration device based on context perception and conditional diffusion model, characterized in that: include: The spatiotemporal feature extraction module is used to extract the spatiotemporal features of trajectory data and capture the temporal sequence and spatial distribution characteristics of trajectory points; UNet denoising module, used to recover missing points in sparse trajectories by progressive denoising; The conditional denoising process module is used to optimize the denoising process by combining multi-source spatiotemporal condition information to ensure that the generated trajectory meets spatiotemporal consistency; The training and optimization module is used to optimize the trajectory recovery model based on the conditional reconstruction loss function to improve the accuracy and robustness of the generated trajectory; A hierarchical fast iteration module is used to remove noise from trajectory data in stages and gradually refine the trajectory completion results; The trajectory recovery module is used to generate high-precision completion trajectories that meet the constraints of multi-source spatiotemporal conditions and output the final results.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The method comprises computer program instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 7.

Images