A vehicle trajectory recovery method and device with full coverage of road network and full penetration of individuals

Abstract

The present invention discloses a vehicle trajectory recovery method and device with full coverage of road networks and full penetration of individuals. Firstly, an auxiliary trajectory selection module with heuristic constraints is designed to select multiple fine-grained trajectories as auxiliary trajectories; secondly, a spatiotemporal embedding module is designed to embed each point of a coarse-grained trajectory and its corresponding multiple auxiliary trajectories into a dense vector, so as to obtain a spatiotemporal embedding representation sequence of each trajectory; thirdly, a coarse-grained trajectory mobility information extraction module and a fine-grained trajectory mobility information extraction module are designed to respectively encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory and its corresponding multiple auxiliary trajectories into a high-level representation sequence; then a mobility information fusion module is designed to use an attention mechanism between trajectories and an average sampling strategy to fuse the above-mentioned high-level representation sequence; finally, a trajectory generation module is designed to generate trajectory points in an unobservable intersection area, so as to recover a fine-grained vehicle trajectory covering the entire urban road network and penetrating all individual vehicles.

Description

Technical Field

[0001] The present invention relates to the technical field related to vehicle movement trajectory recovery, and in particular to a vehicle trajectory recovery method and device with full coverage of road networks and full penetration of individuals. Background Art

[0002] At present, the work on mobile trajectory recovery mainly includes the following three categories:

[0003] (1) Trajectory generated by the mobile sensing system

[0004] In the past few years, most of the trajectory recovery work is based on the movement trajectory of individuals obtained by mobile sensor devices such as vehicle-mounted GPS and smartphones. These works are to restore sparse, low-sampled trajectories into dense trajectories.

[0005] Work based on GPS trajectories mainly achieves trajectory recovery by capturing individual travel patterns. These works rely heavily on fine-grained GPS trajectories, but due to privacy and other reasons, private car owners do not want their trajectories to be collected and used, so such work cannot be applied to all individual vehicles.

[0006] The research work based on smartphone data focuses on modeling the user's mobility pattern by capturing the correlation between the current trajectory and its historical trajectory. The trajectories of this kind of work are provided by location-based service records, but since location-based service records are very sparse and unevenly distributed, it is difficult to model fine-grained mobility patterns.

[0007] (2) Trajectory generated by static sensing system

[0008] In recent years, fixed sensing infrastructure such as traffic surveillance cameras and toll booths based on contactless automatic identification technology have been widely deployed in most cities, providing a new way to obtain human travel trajectories. In the related trajectory recovery work, some of the research focuses on reconstructing trajectories based on the visual features and mobility information of vehicles obtained by traffic surveillance cameras (such as Tong Panrong, Li Mingqian, Li Mo, et al. Large-scale vehicle trajectory reconstruction with camera sensing network [C]. / / Proc of ACM MobiCom. New York, NY: ACM, 2021: 188–200). However, these works cannot model vehicle movement patterns in areas not covered by the sensing system. Some works have also established fine-grained vehicle mobility models on highways with toll booths (such as Yang Yu, Xie Xiaoyang, Fang Zhihan, et al. Vemo: Enabling transparent vehicular mobility modeling at individual levels with full penetration [J]. IEEE Transactions on Mobile Computing (2020)). This method has excellent performance on highways, but has difficulty handling dynamic traffic conditions in complex urban road networks.

[0009] (3) Trajectories generated by heterogeneous sensing systems

[0010] A heterogeneous sensing system is a sensing system formed by combining different sensing systems. In recent years, some work has combined traffic surveillance camera trajectory data and GPS trajectory data to infer the traffic volume at intersections without traffic surveillance camera coverage (such as Xianfeng Tang, Boqing Gong, Yanwei Yu, Huaxiu Yao, Yandong Li, Haiyong Xie, and Xiaoyu Wang. 2019. Joint modeling of dense and incomplete trajectories for citywide traffic volume inference. In Proc. WWW. 1806–1817.). However, this part of the work only focuses on the overall level of traffic flow information, rather than the individual level of movement trajectory. Some work has also modeled the movement patterns of vehicles with traffic camera trajectory data and GPS trajectory data (such as YangWang, Wuji Chen, Wei Zheng, He Huang, Wen Zhang, and Hengchang Liu. 2017. Tracking hit-and-run vehicle with sparse video surveillance cameras and mobile taxicabs. In Proc. of IEEEICDM. 495–504), but did not fully consider the complementary characteristics between heterogeneous sensing systems, which resulted in limited model performance. Summary of the invention

[0011] In view of the shortcomings that it is currently difficult to recover the fine-grained trajectories of all individual vehicles and ensure that the trajectories involve all intersections in the urban road network using data generated by independent mobile sensing devices or static sensing devices, and that the data based on heterogeneous sensing systems cannot effectively utilize the complementary characteristics between heterogeneous sensing systems and combine them with the travel mode of urban roads to recover the trajectory movement at the individual level, the present invention proposes a fine-grained vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals based on the data generated by the heterogeneous sensing systems. By fusing the fine-grained trajectory data generated by the on-board GPS and the coarse-grained trajectory data generated by the traffic monitoring camera, the fine-grained trajectory with full coverage of the road network and full penetration of individuals is recovered.

[0012] In order to achieve the above object, the present invention provides the following technical solutions:

[0013] A vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals, comprising the following steps:

[0014] S1, taking the coarse-grained trajectory to be restored as the query trajectory and selecting multiple fine-grained trajectories as auxiliary trajectories;

[0015] S2. Use the spatiotemporal embedding method to embed each point of the coarse-grained trajectory and its corresponding auxiliary trajectory into a dense vector, capture the spatiotemporal correlation between different trajectory points, and generate the spatiotemporal embedding representation sequence of the coarse-grained trajectory and the spatiotemporal embedding representation sequence of the auxiliary trajectory;

[0016] S3. Use the inter-intersection attention mechanism to extract the co-occurrence relationship from the spatiotemporal embedding representation sequence of each auxiliary trajectory, and encode the spatiotemporal embedding representation sequence of the auxiliary trajectory into a high-level representation sequence; use another inter-intersection attention mechanism to extract the co-occurrence relationship between observable intersections from the spatiotemporal embedding representation sequence of each coarse-grained trajectory, and encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory into a high-level representation sequence;

[0017] S4, use the inter-track attention mechanism to fuse the high-level representation sequence of the auxiliary track and the high-level representation sequence of the coarse-grained track to obtain each fused representation sequence, and then use the average sampling strategy to aggregate all the fused representation sequences into an aggregated representation sequence;

[0018] S5. Use another inter-trajectory attention mechanism to capture the association information between the i-1th representation in the spatiotemporal embedding representation sequence of the coarse-grained trajectory and the aggregated representation sequence, and generate a representation vector of the missing trajectory points in the missing trajectory of the unobservable intersection area.

[0019] Furthermore, in step S1, the spatial similarity between the given coarse-grained trajectory and the fine-grained trajectory is first calculated, and then the top 100 similar candidate trajectories are obtained by sorting according to the spatial similarity. Then, by comparing the temporal similarity of the coarse-grained trajectory and the fine-grained trajectory at the unobservable intersection, multiple fine-grained trajectories with the highest temporal similarity are selected as auxiliary trajectories. The temporal similarity calculation formula is:

[0020] Sim temporal (C,F j )=1-|UTR(C)-UTR(F j )|

[0021] Among them, UTR(T)=|T un | / |T| is the proportion of the time corresponding to the unobservable part of a coarse-grained trajectory in the total time of the complete trajectory. For trajectory T, |T| represents the total time of the trajectory, and |T un | represents the time corresponding to the unobservable part of the trajectory.

[0022] Furthermore, in step S2, the spatiotemporal embedding method is specifically as follows: first, Skip-Gram is used to generate a spatial embedding of each intersection in the road network, and each timestamp is represented as a quadruple; next, the quadruple is input into a two-layer fully connected neural network and converted into a dense vector of the same dimension as the spatial embedding; finally, for the i-th trajectory point in the trajectory, the spatial and temporal embedding vectors are combined into a spatiotemporal vector.

[0023] Furthermore, in step S3, the high-level representation sequence of the auxiliary trajectory is:

[0024]

[0025] in, is the jth high-level representation sequence obtained for each trajectory point in the jth auxiliary trajectory, is the high-level representation vector of the ith trajectory point in the jth trajectory,

[0026] Furthermore, in step S3, the high-level representation sequence of the coarse-grained trajectory is:

[0027]

[0028] Among them, i represents the intersection attention mechanism applied to the i-th intersection node, For each high-level representation, EMB(C) = {emb1(C), ..., emb n (C)} is the spatiotemporal embedding representation sequence of the coarse-grained trajectory C.

[0029] Furthermore, step S4 first uses the inter-trajectory attention mechanism to compare the similarity of the co-occurrence relations between the coarse-grained trajectory high-level representation sequence and the j-th auxiliary trajectory high-level representation sequence, and then combines the two to generate a fused representation sequence, and uses an average sampling strategy to aggregate all fused representation sequences into an aggregated representation sequence to aggregate the co-occurrence relations that appear separately.

[0030] Furthermore, step S4 is expressed as follows using the inter-track attention mechanism formula:

[0031]

[0032] in The computational logic of is consistent with the inter-junction attention, but the input is a sequence of high-level representations of the coarse-grained trajectory and the th auxiliary trajectory.

[0033] Furthermore, step S5 uses another formula of the inter-track attention mechanism as follows:

[0034]

[0035] Among them, emb i-1 (C) is the (i-1)th representation in EMB(C), and EMB(AGG) is an aggregate representation sequence.

[0036] On the other hand, the present invention provides a vehicle trajectory recovery device with full coverage of the road network and full penetration of individuals, comprising the following modules to implement any of the above methods:

[0037] An auxiliary trajectory selection module, used to take the coarse-grained trajectory to be restored as a query trajectory and select multiple fine-grained trajectories as auxiliary trajectories;

[0038] The spatiotemporal embedding module is used to embed each point of a given coarse-grained trajectory and its corresponding auxiliary trajectory into a dense vector, capture the spatiotemporal correlation between different trajectory points, and generate a spatiotemporal embedding representation sequence for each trajectory;

[0039] A fine-grained trajectory mobility information extraction module is used to extract co-occurrence relations from the spatiotemporal embedding representation sequence of each auxiliary trajectory using the inter-junction attention mechanism, and encode the spatiotemporal embedding representation sequence of the auxiliary trajectory into a high-level representation sequence;

[0040] A coarse-grained trajectory mobility information extraction module is used to extract the co-occurrence relationship between observable intersections from the spatiotemporal embedding representation sequence of each coarse-grained trajectory using another inter-intersection attention mechanism, and encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory into a high-level representation sequence;

[0041] The mobility information fusion module is used to fuse the high-level representation sequence of the auxiliary trajectory and the high-level representation sequence of the coarse-grained trajectory using the inter-trajectory attention mechanism to obtain each fused representation sequence, and then use the average sampling strategy to aggregate all the fused representation sequences into an aggregated representation sequence;

[0042] The trajectory generation module is used to generate a representation vector of missing trajectory points in the missing trajectory of the unobservable intersection area.

[0043] In addition, the present invention also provides an electronic device, including: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, wherein:

[0044] Memory, used to store computer programs;

[0045] The processor is used to implement any of the above methods when executing the program stored in the memory.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] (1) The vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals proposed in the present invention uses fine-grained GPS trajectory data covering the entire road network and traffic monitoring camera trajectory data that penetrates all individual vehicles. It can recover fine-grained vehicle trajectories that cover the entire urban road network and penetrate all individual vehicles. This is of great significance for understanding the mobility of the city and supports a wide range of downstream applications, such as vehicle tracking, intelligent traffic signal control, congestion prediction, urban planning, etc.

[0048] (2) This method uses the complementary characteristics of coarse-grained and fine-grained trajectories to model and fuse the mobility information of coarse-grained and fine-grained trajectories, thereby recovering the trajectories in the unobservable area. Compared with other methods, the trajectories recovered by this method have the highest accuracy and the smallest distance error.

[0049] (3) This method has good accuracy and small distance error in different time periods, urban areas with different functions, and different traffic surveillance camera coverage rates. This proves that this method is highly robust in recovering trajectories and can cope with complex real-world environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0051] Figure 1 A flow chart of a vehicle trajectory recovery method with full road network coverage and full individual penetration provided by an embodiment of the present invention.

[0052] Figure 2 A functional module diagram of a vehicle trajectory recovery device with full road network coverage and full individual penetration provided by an embodiment of the present invention.

[0053] Figure 3 A diagram of the structure of a device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0055] First, some basic concepts involved in this method are defined as follows:

[0056] a) Road network structure

[0057] The urban road network is a directed graph and eij ∈ε. represents the set of all intersections in the city, v i represents the i-th intersection in the city; ε represents the set of all roads in the city, e ij Represents from v i Intersection to V j A road section at an intersection corresponds to a directed edge in the graph structure. A collection of observable intersections covered by traffic surveillance cameras and the set of unobservable intersections without traffic surveillance cameras Composition and satisfy and

[0058] b) Coarse-grained trajectory collection

[0059] The coarse-grained trajectory set consists of the trajectories of all individual vehicles captured by traffic surveillance cameras and is represented as Where N is the total number of trajectories of all vehicles. represents the i-th coarse-grained trajectory, where represents the jth observable intersection that the vehicle passes in the current trajectory, represents the time to pass the intersection, and n represents that the i-th trajectory has n trajectory points. The coarse-grained trajectory set is divided into two parts: the ordinary vehicle trajectory set and operational vehicle trajectory collection And meet and Because the coarse-grained trajectories come from traffic surveillance cameras, the set of intersections involved in all coarse-grained trajectories is the observable intersection set

[0060] c) Fine-grained trajectory collection

[0061] The fine-grained trajectory set is collected by the vehicle-mounted GPS deployed on the operating vehicle and matched to the road network, denoted as Where M is the total number of trajectories of all operational vehicles. The i-th fine-grained trajectory is given by Indicates that represents the jth intersection of the vehicle in the current trajectory, represents the time to pass the intersection, and m represents the number of track points of the track. Because the fine-grained track comes from the vehicle GPS, the set of all the intersections involved in all the fine-grained tracks is

[0062] The trajectory of an operating vehicle can be obtained from the vehicle-mounted GPS or from a traffic monitoring camera, and the coarse-grained trajectory of an operating vehicle obtained from the traffic monitoring camera data can find a corresponding fine-grained trajectory.

[0063] d) Trajectory recovery

[0064] Given a coarse-grained set of trajectories and a fine-grained set of trajectories This method needs to recover all coarse-grained common vehicle trajectories At unobservable intersections The missing records are collected to obtain fine-grained vehicle movement trajectories that meet the full coverage of the urban road network and the full penetration of individual vehicles.

[0065] Based on the above, the present invention proposes a vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals, such as Figure 1 As shown, the method includes the following steps: first, an auxiliary trajectory selection module with heuristic constraints is designed to select multiple fine-grained trajectories as auxiliary trajectories; second, a spatiotemporal embedding module is designed to embed each point of the coarse-grained trajectory and its corresponding multiple auxiliary trajectories into a dense vector to obtain a spatiotemporal embedding representation sequence of each trajectory; third, a coarse-grained trajectory mobility information extraction module and a fine-grained trajectory mobility information extraction module are designed to encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory and its corresponding multiple auxiliary trajectories into a high-level representation sequence respectively; then a mobility information fusion module is designed to use the inter-trajectory attention mechanism and the average sampling strategy to fuse the above high-level representation sequence; finally, a trajectory generation module is designed to generate trajectory points in the unobservable intersection area.

[0066] The specific design of this method is introduced as follows:

[0067] a) Auxiliary trajectory selection module

[0068] Trajectory recovery is the process of restoring coarse-grained trajectories to fine-grained trajectories. This method uses the coarse-grained trajectory to be recovered as the query trajectory to retrieve useful fine-grained trajectories as auxiliary trajectories. Specifically, this method first calculates the spatial similarity between the given coarse-grained trajectory and the fine-grained trajectory, and then sorts the top 100 similar candidate trajectories F according to the similarity. top-100 However, even if there are similar trajectory segments in the observable intersection part, the trajectory may have different route choices in the unobservable intersection part. To solve this problem, this method proposes to further select candidate trajectories by comparing their temporal similarities in the unobservable intersection. First, define UTR(T) = |T un| / |T| is the proportion of the time corresponding to the unobservable part of a coarse-grained trajectory in the total time of the complete trajectory. For trajectory T, |T| represents the total time of the trajectory, |T un | represents the time corresponding to the unobservable part in the trajectory. Then, the given coarse-grained trajectory to be restored and the first 100 auxiliary trajectories F are calculated. top-100 The temporal similarity between them is calculated as follows:

[0069] Sim temporal (C,F j )=1-|UTR(C)-UTR(F j )|

[0070] This method selects the k trajectories with the highest temporal similarity among the 100 candidate trajectories that pass spatial screening as auxiliary trajectories.

[0071] b) Spatiotemporal Embedding Module

[0072] Given a fine-grained trajectory F = {(v1, t1), ..., (v m , t m )}, this method uses the spatiotemporal embedding module to embed each point (v i , t i ) is embedded into a dense vector to capture the spatiotemporal correlation between different trajectory points. Specifically, this method first uses Skip-Gram to generate a spatial embedding for each intersection in the road network. in d represents the dimension of spatial embedding. However, it is difficult for only the dense vector of spatial embedding to represent the dynamic and periodic movement of the trajectory. Therefore, this method proposes to convert each timestamp t i Represented as a four-tuple, denoted as t i ∈R 7+24+60+60 Next, the quadruple is input into a two-layer fully connected neural network and converted into a dense vector emb of the same dimension as the spatial embedding ti ∈R d Finally, for the i-th trajectory point (v i , t i ), combining the spatial and temporal embedding vectors into a spatiotemporal vector emb i ∈R d As shown below:

[0073] emb i =emb vi +emb ti

[0074] Trajectory F = {(v1, t1), ..., (v m , tm After all the trajectory points in )} pass through the spatiotemporal embedding module, the representation sequence EMB(F) = {emb1(F), …, emb m (F)}.

[0075] c) Fine-grained trajectory mobility information extraction module

[0076] The auxiliary trajectory is embedded in the spatiotemporal embedding module to obtain the spatiotemporal embedding sequence of each auxiliary trajectory, denoted as EMB(F j )={emb1(F j ), …, emb m (F j )}, where j represents the jth fine-grained trajectory selected. This method considers the spatiotemporal dependencies between trajectory points that appear together in the same trajectory and defines these dependencies as co-occurrence relationships. This method uses the intersection attention mechanism to extract the co-occurrence relationship from each auxiliary trajectory F. j The low-level representation sequence EMB(F j ) to extract the co-occurrence relationship. The importance of the co-occurrence relationship between a pair of trajectory points i and u under the h-th head is as follows:

[0077]

[0078]

[0079] in is the transformation matrix, <, > is the dot product function. The weighted co-occurrence relationship of the i-th trajectory point in the trajectory to all points in the trajectory is:

[0080]

[0081] in is another transformation matrix with the same structure as the previous one. Since the multi-head attention mechanism is used to extract co-occurrence relations from multiple angles, the next step is to splice the co-occurrence information extracted from each angle to obtain the integrated co-occurrence relationship as follows:

[0082]

[0083] Where || represents the concatenation operation and H is the number of heads used by the multi-head attention mechanism. The co-occurrence relationship between trajectory point i and all other trajectory points appearing in the same trajectory is expressed as:

[0084]

[0085] Where W 1 ∈R d′×d is the projection matrix that achieves dimension alignment.

[0086] In summary, the complete process of designing the attention mechanism between intersections in this method can be abstracted as follows:

[0087]

[0088] The superscript 1 is used to distinguish the multi-head attention networks in each module. is the high-level representation vector of the i-th trajectory point in the j-th trajectory. Applying the above process to each trajectory point in the j-th auxiliary trajectory can obtain the j-th high-level representation sequence, which is recorded as The high-level representation sequence contains the hidden co-occurrence relationship between observable and unobservable intersections. Since this module uses k auxiliary trajectories, the high-level representation sequence of the auxiliary trajectories is expressed as:

[0089]

[0090] d) Coarse-grained trajectory mobility information extraction module

[0091] Because the observable trajectory points in the trajectory involve the geographical area information of the missing trajectory segments and provide useful mobility information for trajectory recovery, although the coarse-grained trajectory only consists of observable intersections, it is still necessary to extract the co-occurrence relations between these intersections. At the same time, by utilizing the correlation of the co-occurrence relations of observable intersections between different trajectories, useful co-occurrence relations can be obtained from the fine-grained trajectory.

[0092] First, the given coarse-grained trajectory C is encoded into a spatiotemporal embedding representation sequence through the spatiotemporal embedding module:

[0093] EMB(C)={emb1(C),…,emb n (C)}

[0094] Then, for this low-level representation sequence EMB(C), another intersection attention mechanism is used to extract the co-occurrence relationship between observable intersections to obtain the corresponding high-level representation sequence Each high-level representation obtained is:

[0095]

[0096] Where i indicates that the attention mechanism between intersections is applied to the i-th intersection node. It is obtained by applying the above process to each node of the low-level representation sequence EMB(C), that is,

[0097] e) Mobility information fusion module

[0098] After the above operations, the co-occurrence relationship between the coarse-grained trajectory and the auxiliary trajectory has been extracted as a high-level representation sequence and Next, we compare the co-occurrence relations between observable intersections contained in the coarse-grained representation sequence and the fine-grained representation sequence, and combine useful co-occurrence relations based on their similarities.

[0099] This method proposes an inter-track attention mechanism to compare high-level representation sequences of coarse-grained tracks. and the j-th auxiliary trajectory high-level representation sequence Then they are combined to generate the fusion representation sequence EMB (FUSE j ), which has the same number subscript as the auxiliary track. The formula of the attention mechanism between tracks is as follows:

[0100]

[0101] in The computational logic of is consistent with the attention between intersections, but the input is a high-level representation sequence of the coarse-grained trajectory and the auxiliary trajectory. Each fused representation sequence EMB (FUSE j ) and then use the average sampling strategy to represent all fusions EMB (FUSE j ) is aggregated into an aggregate representation sequence EMB(AGG) to aggregate the co-occurrence relationships that appear separately.

[0102] f) Trajectory generation module

[0103] In order to generate the representation vector emb corresponding to the missing trajectory point in the missing trajectory REC i (REC), which represents the i-th position of the recovered trajectory representation vector sequence EMB(REC), and uses another inter-trajectory attention mechanism to capture the (i-1)th representation in EMB(C) and the aggregate representation sequence EMB(AGG). The formula of the inter-trajectory attention mechanism is as follows:

[0104]

[0105] Generate representation emb i After (REC), it is decoded to obtain the intersection number. Since this method is based on the road network for trajectory recovery, it should be ensured that the decoded intersection number is first-order adjacent to the intersection corresponding to the previous trajectory point. The probability that the coarse-grained trajectory C visits the unobservable intersection l is as follows:

[0106]

[0107] where NBR is the number of first-order neighbors of the intersection at the (i-1)th position of the restored trajectory, emb l is the characterization vector of the first-order neighbor intersection l. The intersection number with the highest probability is taken as the recovery result of the previous missing.

[0108] Corresponding to the vehicle trajectory recovery method with full road network coverage and full individual penetration provided by the above-mentioned embodiment of the present invention, the embodiment of the present invention also provides a vehicle trajectory recovery device with full road network coverage and full individual penetration.

[0109] like Figure 2 As shown, the vehicle trajectory recovery device with full coverage of the road network and full penetration of individuals includes:

[0110] An auxiliary trajectory selection module, used to take the coarse-grained trajectory to be restored as a query trajectory and select multiple fine-grained trajectories as auxiliary trajectories;

[0111] The spatiotemporal embedding module is used to embed each point of a given coarse-grained trajectory and its corresponding auxiliary trajectory into a dense vector, capture the spatiotemporal correlation between different trajectory points, and generate a spatiotemporal embedding representation sequence for each trajectory;

[0112] A fine-grained trajectory mobility information extraction module is used to extract co-occurrence relations from the spatiotemporal embedding representation sequence of each auxiliary trajectory using the inter-junction attention mechanism, and encode the spatiotemporal embedding representation sequence of the auxiliary trajectory into a high-level representation sequence;

[0113] A coarse-grained trajectory mobility information extraction module is used to extract the co-occurrence relationship between observable intersections from the spatiotemporal embedding representation sequence of each coarse-grained trajectory using another inter-intersection attention mechanism, and encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory into a high-level representation sequence;

[0114] The mobility information fusion module is used to fuse the high-level representation sequence of the auxiliary trajectory and the high-level representation sequence of the coarse-grained trajectory using the inter-trajectory attention mechanism to obtain each fused representation sequence, and then use the average sampling strategy to aggregate all the fused representation sequences into an aggregated representation sequence;

[0115] The trajectory generation module is used to generate a representation vector of missing trajectory points in the missing trajectory of the unobservable intersection area.

[0116] Corresponding to the vehicle trajectory recovery method with full road network coverage and full individual penetration provided by the above-mentioned embodiment of the present invention, the embodiment of the present invention also provides an electronic device.

[0117] like Figure 3 As shown, the electronic device includes: a processor 301, a communication interface 302, a memory 603 and a communication bus 304, wherein the processor 601, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304.

[0118] Memory 303, used for storing computer programs;

[0119] The processor 301 is used to implement the steps of the vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals provided by the above-mentioned embodiment of the present invention when executing the program stored in the memory 303.

[0120] The communication bus mentioned in the above device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0121] The communication interface is used for communication between the above electronic device and other devices.

[0122] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0123] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0124] In another embodiment provided by the present invention, a computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of the vehicle trajectory recovery method with full road network coverage and full individual penetration provided by any of the above-mentioned embodiments of the present invention are implemented.

[0125] In another embodiment provided by the present invention, a computer program product comprising instructions is also provided. When the computer is run on a computer, the computer executes the steps of any of the vehicle trajectory recovery methods with full road network coverage and full individual penetration provided by the above-mentioned embodiments of the present invention.

[0126] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital terminal equipment line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0127] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0128] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, electronic device embodiment, computer-readable storage medium embodiment, and computer program product embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0129] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals, characterized in that: The following steps are involved: S1, taking the coarse-grained trajectory to be restored as the query trajectory and selecting multiple fine-grained trajectories as auxiliary trajectories; S2. Use the spatiotemporal embedding method to embed each point of the coarse-grained trajectory and its corresponding auxiliary trajectory into a dense vector, capture the spatiotemporal correlation between different trajectory points, and generate the spatiotemporal embedding representation sequence of the coarse-grained trajectory and the spatiotemporal embedding representation sequence of the auxiliary trajectory; S3. Use the inter-intersection attention mechanism to extract the co-occurrence relationship from the spatiotemporal embedding representation sequence of each auxiliary trajectory, and encode the spatiotemporal embedding representation sequence of the auxiliary trajectory into a high-level representation sequence; use another inter-intersection attention mechanism to extract the co-occurrence relationship between observable intersections from the spatiotemporal embedding representation sequence of each coarse-grained trajectory, and encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory into a high-level representation sequence; The high-level representation sequence of the auxiliary trajectory is: in, is the jth high-level representation sequence obtained for each trajectory point in the jth auxiliary trajectory, is the high-level representation vector of the i-th trajectory point in the j-th trajectory, represents the attention mechanism between intersections used, emb i (F j ) represents each auxiliary trajectory F j The space-time vector of EMB(F j ) represents each auxiliary trajectory F j The spatiotemporal embedding of represents the sequence; The high-level representation sequence of the coarse-grained trajectory is: Among them, i represents the intersection attention mechanism applied to the i-th intersection node, For each coarse-grained trajectory C, the high-level representation vector, EMB(C) = {emb1(C),…,emb n (C)} is the spatiotemporal embedding representation sequence of the coarse-grained trajectory C; Indicates another intersection attention mechanism used, emb i (C) represents the spatiotemporal vector of the coarse-grained trajectory C; S4, use the inter-track attention mechanism to fuse the high-level representation sequence of the auxiliary track and the high-level representation sequence of the coarse-grained track to obtain each fused representation sequence, and then use the average sampling strategy to aggregate all the fused representation sequences into an aggregated representation sequence; S5. Use another inter-trajectory attention mechanism to capture the association information between the i-1th representation in the spatiotemporal embedding representation sequence of the coarse-grained trajectory and the aggregated representation sequence, and generate a representation vector of the missing trajectory points in the missing trajectory of the unobservable intersection area.

2. The vehicle trajectory recovery method with full road network coverage and full individual penetration according to claim 1 is characterized in that: In step S1, the spatial similarity between the given coarse-grained trajectory and the fine-grained trajectory is first calculated, and then the top 100 similar candidate trajectories are obtained by sorting according to the spatial similarity. Then, by comparing the temporal similarity of the coarse-grained trajectory and the fine-grained trajectory at the unobservable intersection, multiple fine-grained trajectories with the highest temporal similarity are selected as auxiliary trajectories. The temporal similarity calculation formula is: Sim temporal (C,F j )=1-|UTR(C)-UTR(F j )| Among them, UTR(T)=|T un | / |T| is the proportion of the time corresponding to the unobservable part of a coarse-grained trajectory in the total time of the complete trajectory. For trajectory T, |T| represents the total time of the trajectory, and |T un | represents the time corresponding to the unobservable part of the trajectory.

3. The vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals according to claim 1 is characterized in that: In step S2, the spatiotemporal embedding method is as follows: first, Skip-Gram is used to generate spatial embeddings for each intersection in the road network, and each timestamp is represented as a quadruple; next, the quadruple is input into a two-layer fully connected neural network and converted into a dense vector of the same dimension as the spatial embedding; finally, for the i-th trajectory point in the trajectory, the spatial and temporal embedding vectors are combined into a spatiotemporal vector.

4. The vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals according to claim 1 is characterized in that: Step S4 first uses the inter-track attention mechanism to compare the similarity of the co-occurrence relations between the coarse-grained trajectory high-level representation sequence and the j-th auxiliary trajectory high-level representation sequence, and then combines the two to generate a fused representation sequence. The average sampling strategy is used to aggregate all fused representation sequences into an aggregated representation sequence to aggregate the co-occurrence relations that appear separately.

5. The vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals according to claim 4 is characterized in that: Step S4 uses the inter-track attention mechanism formula as follows: in, represents the attention mechanism between trajectories, The computational logic of is consistent with the attention between intersections, but the input is the high-level representation vector of the coarse-grained trajectory C and the high-level representation sequence of the j-th auxiliary trajectory, emb i (FUSE j ) represents the fused representation vector.

6. The vehicle trajectory recovery method with full coverage of the road network and full penetration of individuals according to claim 1 is characterized in that: Step S5 uses another formula of the inter-track attention mechanism as follows: in, represents another inter-track attention mechanism, emb i-1 (C) is the (i-1)th representation in EMB(C), EMB(AGG) is the aggregate representation sequence, emb i (REC) represents the representation vector corresponding to the missing trajectory point in the missing trajectory REC.

7. A vehicle trajectory recovery device with full coverage of the road network and full penetration of individuals, characterized in that: The method comprises the following modules to implement the method according to any one of claims 1 to 6: An auxiliary trajectory selection module, used to take the coarse-grained trajectory to be restored as a query trajectory and select multiple fine-grained trajectories as auxiliary trajectories; The spatiotemporal embedding module is used to embed each point of a given coarse-grained trajectory and its corresponding auxiliary trajectory into a dense vector, capture the spatiotemporal correlation between different trajectory points, and generate a spatiotemporal embedding representation sequence for each trajectory; A fine-grained trajectory mobility information extraction module is used to extract co-occurrence relations from the spatiotemporal embedding representation sequence of each auxiliary trajectory using the inter-junction attention mechanism, and encode the spatiotemporal embedding representation sequence of the auxiliary trajectory into a high-level representation sequence; A coarse-grained trajectory mobility information extraction module is used to extract the co-occurrence relationship between observable intersections from the spatiotemporal embedding representation sequence of each coarse-grained trajectory using another inter-intersection attention mechanism, and encode the spatiotemporal embedding representation sequence of the coarse-grained trajectory into a high-level representation sequence; The mobility information fusion module is used to fuse the high-level representation sequence of the auxiliary trajectory and the high-level representation sequence of the coarse-grained trajectory using the inter-trajectory attention mechanism to obtain each fused representation sequence, and then use the average sampling strategy to aggregate all the fused representation sequences into an aggregated representation sequence; The trajectory generation module is used to generate a representation vector of missing trajectory points in the missing trajectory of the unobservable intersection area.

8. An electronic device comprising: The processor, the communication interface, the memory and the communication bus communicate with each other through the communication bus, wherein: Memory, used to store computer programs; A processor, for implementing the method according to any one of claims 1 to 6 when executing a program stored in a memory.

Images