CN113902830B

CN113902830B - Method for generating track road network

Info

Publication number: CN113902830B
Application number: CN202111487400.6A
Authority: CN
Inventors: 李亚宁
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-12-08
Filing date: 2021-12-08
Publication date: 2022-03-04
Anticipated expiration: 2041-12-08
Also published as: CN113902830A

Abstract

The embodiment of the disclosure provides a track network generation method, a track network generation device, track network generation equipment and a computer-readable storage medium. The method of the embodiment of the disclosure uses a transmission structure of a message queue, models and divides track data, takes generated data blocks as a transmission unit and a processing unit, transmits the data blocks to deep learning processing through the message queue, realizes rapid production and consumption of the data through asynchronous pulling and uploading, and simultaneously uses the data generated by the deep learning in parallel to generate a part of a vector track network through the message queue again, thereby automatically generating a nationwide full-scale track network. The method disclosed by the invention obviously reduces the time required by processing each part for generating the track network, improves the generation efficiency of the track network, and ensures that the whole track network generation system has good timeliness, stability and transportability due to asynchronous parallel processing among the parts. The embodiment of the disclosure can be applied to the field of maps.

Description

Method for generating track road network

Technical Field

The present disclosure relates to the field of artificial intelligence and data mining, and more particularly, to a method, an apparatus, a device, and a storage medium for generating a trajectory road network.

Background

With the rapid development of electronic communication technology, internet technology and internet of things technology, road network data is becoming more and more important in city planning, road supervision, map value-added services and other aspects. No matter the external expansion of cities and the optimization of internal structures, the change of urban road network can be caused. The traditional urban road network data is mainly obtained by two modes of surveying and mapping by professional personnel and digitalizing a remote sensing image, but the two methods have high investment cost, high professional technical requirements of the professional personnel and long production period, and are difficult to meet the requirements of rapid development and change of cities on the road network data.

With the large installation and widespread use of Global Positioning System (GPS) devices, people can conveniently acquire a large amount of trajectory data. These trajectory data sets could potentially reflect urban road networks, human travel behavior, and urban traffic dynamics, among others. For example, a person's walking trajectory versus a vehicle's driving trajectory may reflect potential walking road networks and vehicle road networks. The automatic extraction of the track road network from a large number of GPS tracks can greatly improve the construction and updating speed of the map and greatly reduce the cost. In addition, the way of extracting the track road network from a large number of tracks can lead the obtained road network to have some information on traffic. However, the existing track road network generation method does not consider road network priori knowledge due to the existence of a large number of artificial experience parameters, the generated road network form effect is poor, the adaptivity, the noise immunity and the like for the traffic-intensive area with large data volume have great problems, and meanwhile, because the existing method is mostly an experimental property scheme, the capacity of automatically generating the track road network with large data volume (for example, nationwide) is unavailable at present.

Therefore, there is a need for an efficient and accurate method for generating a road network of trajectories that enables rapid and accurate nationwide automated road network generation.

Disclosure of Invention

In order to solve the above problem, the present disclosure forms a whole amount of a nationwide track network by generating a part of the track network in a block transmission unit using a transmission structure of a message queue to generate a plurality of parts of the track network through asynchronous processing.

The embodiment of the disclosure provides a track network generation method, a track network generation device, track network generation equipment and a computer-readable storage medium.

The embodiment of the disclosure provides a track road network generation method, which includes acquiring track data, wherein the track data comprises a first number of track data sets; for each of the first number of trajectory datasets, generating a corresponding three-channel color image block based on the trajectory dataset; generating a corresponding binary image block through deep learning based on the three-channel color image block; and vectorizing the binarized image blocks to generate a partial trajectory network, wherein the trajectory network is composed of a first number of partial trajectory networks corresponding to the first number of trajectory data sets.

An embodiment of the present disclosure provides a track network generation apparatus, including: a data acquisition module configured to acquire trajectory data, the trajectory data comprising a first number of trajectory data sets; an image generation module configured to, for each of the first number of trajectory datasets, generate a corresponding three-channel color image block based on the trajectory dataset; the background segmentation module is configured to generate a corresponding binarization image block through deep learning based on the three-channel color image block; and a road network generation module configured to perform vectorization on the binarized image blocks to generate a partial trajectory road network, the partial trajectory road network of a first number corresponding to the first number of trajectory data sets constituting the trajectory road network.

An embodiment of the present disclosure provides a track network generating device, including: one or more processors; and one or more memories, wherein said one or more memories have stored therein a computer executable program that, when executed by said processor, performs a trajectory network generation method as described above.

Embodiments of the present disclosure provide a computer-readable storage medium having stored thereon computer-executable instructions for implementing a trajectory network generation method as described above when executed by a processor.

Embodiments of the present disclosure provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the trajectory net generation method according to the embodiment of the present disclosure.

Compared with the conventional track network generation method, the method provided by the embodiment of the disclosure can generate a plurality of parts in the track network through asynchronous processing to form the nationwide track network, thereby shortening the time required by the nationwide track network generation and improving the track network generation efficiency.

The method provided by the embodiment of the disclosure uses a transmission structure of a message queue, models and divides track data, uses generated data blocks as a transmission unit and a processing unit, transmits the data blocks to deep learning processing through the message queue, realizes rapid production and consumption of the data through asynchronous pulling and uploading, and simultaneously uses the data generated by the deep learning for generating a part of a vector track network through the message queue again in a concurrent manner, thereby automatically generating a nationwide full-scale track network. The method provided by the embodiment of the disclosure obviously reduces the time required by processing each part for generating the track network, improves the generation efficiency of the track network, and ensures that the whole track network generation system has good timeliness, stability and transportability due to asynchronous parallel processing among the parts, thereby laying a foundation for quickly realizing the following road excavation and hanging of walking and vehicle automatic.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings that are required to be used in the description of the embodiments will be briefly described below. It is apparent that the drawings in the following description are only exemplary embodiments of the disclosure, and that other drawings may be derived from those drawings by a person of ordinary skill in the art without inventive effort.

Fig. 1A is a schematic diagram illustrating a scenario of processing of a travel request from a user terminal according to an embodiment of the present disclosure;

FIG. 1B is a schematic diagram illustrating a return to a different travel scenario according to an embodiment of the present disclosure;

FIG. 1C is a schematic diagram illustrating an existing deep learning based trajectory net generation architecture, according to an embodiment of the present disclosure;

FIG. 2 is a flow chart illustrating a trajectory net generation method according to an embodiment of the present disclosure;

FIG. 3A is a schematic diagram illustrating a processing architecture of a trajectory net generation method according to an embodiment of the present disclosure;

FIG. 3B is a flow diagram illustrating a trace network generation via message queues according to an embodiment of the present disclosure;

FIG. 4A is a schematic diagram illustrating generating a mesh image based on pairs of trajectory points, according to an embodiment of the present disclosure;

FIG. 4B is a schematic diagram illustrating the generation of a three-channel color image block based on a grid image in accordance with an embodiment of the present disclosure;

FIG. 4C is a schematic diagram illustrating an overlap between three channel color image blocks according to an embodiment of the present disclosure;

FIG. 5A is a schematic diagram illustrating generation of a binarized image block based on a three-channel color image block according to embodiments of the present disclosure;

FIG. 5B is a flow diagram illustrating generation of a binarized image block based on a three-channel color image block according to embodiments of the present disclosure;

fig. 6 is a schematic diagram illustrating vectorization processing according to an embodiment of the present disclosure;

FIG. 7A is a diagram illustrating an example trajectory net generated in accordance with an embodiment of the present disclosure;

FIG. 7B is a schematic diagram illustrating a comparison of the generated trajectory network before and after application of the trajectory network generation method according to an embodiment of the present disclosure;

FIG. 8 is a schematic flow chart diagram illustrating subsequent data mining in accordance with an embodiment of the present disclosure;

FIG. 9 is a schematic diagram illustrating a trajectory net generation device according to an embodiment of the present disclosure;

FIG. 10 shows a schematic diagram of a trajectory net generation device according to an embodiment of the present disclosure;

FIG. 11 shows a schematic diagram of an architecture of an exemplary computing device, in accordance with embodiments of the present disclosure; and

FIG. 12 shows a schematic diagram of a storage medium according to an embodiment of the disclosure.

Detailed Description

In order to make the objects, technical solutions and advantages of the present disclosure more apparent, example embodiments according to the present disclosure will be described in detail below with reference to the accompanying drawings. It is to be understood that the described embodiments are merely a subset of the embodiments of the present disclosure and not all embodiments of the present disclosure, with the understanding that the present disclosure is not limited to the example embodiments described herein.

In the present specification and the drawings, steps and elements having substantially the same or similar characteristics are denoted by the same or similar reference numerals, and repeated description of the steps and elements will be omitted. Meanwhile, in the description of the present disclosure, the terms "first", "second", and the like are used only for distinguishing the description, and are not to be construed as indicating or implying relative importance or order.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to be limiting of the invention.

For the purpose of describing the present disclosure, concepts related to the present disclosure are introduced below.

The track road network generation method of the present disclosure may be based on Artificial Intelligence (AI). Artificial intelligence is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. For example, with respect to the artificial intelligence-based trajectory road network generating method, it is possible to determine a road network of walking trajectories corresponding to a road to be ridden or a road network of vehicle trajectories based on road data of a human or vehicle trajectories data of a vehicle in a manner similar to a manner in which a human recognizes various roads on the ground by naked eyes and draws a road network based on a geographic coordinate system. The artificial intelligence enables the track road network generating method disclosed by the invention to have the function of rapidly, accurately and automatically excavating a walking road network and a vehicle road network in the environment by researching the design principle and the implementation method of various intelligent machines.

For example, the trajectory road network generation method of the present disclosure may be based on deep learning. Deep learning is an algorithm based on characterization learning of data in machine learning. An observation (e.g., an image) may be represented using a number of ways, such as a vector of intensity values for each pixel, or more abstractly as a series of edges, a region of a particular shape, and so forth. Tasks (e.g., face recognition or facial expression recognition) are more easily learned from the examples using some specific representation methods. The benefit of deep learning is to replace the manual feature acquisition with unsupervised or semi-supervised feature learning and hierarchical feature extraction efficient algorithms. Wherein, optionally, the trajectory data processing method of the present disclosure may be based on a Convolutional Neural Network (CNN). The convolutional neural network is a feedforward neural network, and the artificial neurons of the convolutional neural network can respond to a part of surrounding units in a coverage range, so that the convolutional neural network has excellent performance on large-scale image processing. The convolutional neural network consists of one or more convolutional layers and an apical fully-connected layer (corresponding to a classical neural network), while also including associated weights and pooling layers. In embodiments of the present disclosure, a convolutional neural network may be used for binarization of a three-channel color image to form a binary digital image.

Alternatively, the trajectory data processing method of the present disclosure may be based on an Intelligent Transportation System (ITS). An Intelligent Transportation System is a comprehensive Transportation System which effectively and comprehensively applies advanced scientific technologies (information technology, computer technology, data communication technology, sensor technology, electronic control technology, automatic control theory, operational research, artificial intelligence and the like) to Transportation, service control and vehicle manufacturing and strengthens the relation among vehicles, roads and users, thereby forming a comprehensive Transportation System which ensures safety, improves efficiency, improves environment and saves energy.

Optionally, the trajectory data processing method of the present disclosure may also be based on a data mining technology for user travel data (e.g., walking data or vehicle data) in the intelligent transportation system. The data mining technology utilizes large-scale data generated by daily trip behaviors of users to mine implicit modes in the large-scale data so as to provide a powerful and flexible analysis and processing function for the large-scale data. As an application-oriented data analysis processing technology, the data mining technology can quickly, effectively and deeply analyze massive traffic information and mine the track network information hidden in massive traffic data.

Alternatively, the trajectory data processing method of the present disclosure may be based on a Message queue (Message queue). A message queue is a form of inter-process communication or communication between different threads of the same process, and a queue of software is used to process a series of inputs (typically from a user). Message queues provide an asynchronous communication protocol, and the records in each queue contain data specifying the time of occurrence, the type of input device, and specific input parameters, i.e., the sender and recipient of the message need not interact with the message queue at the same time. The message will be kept in the queue until the recipient retrieves it. For example, the message queue may be, for example, a message queue CKafka, which can effectively decouple the relationship between the producer and the consumer, so that the respective parts of the processing in the method of the present disclosure can be executed asynchronously and concurrently.

In summary, the solutions provided by the embodiments of the present disclosure relate to technologies such as artificial intelligence, intelligent transportation, and data mining, and the embodiments of the present disclosure will be further described with reference to the accompanying drawings.

Fig. 1A is a schematic diagram illustrating a scenario 100 of a process of a travel request from a user terminal according to an embodiment of the present disclosure. Fig. 1B is a schematic diagram illustrating a return to different travel scenarios according to an embodiment of the present disclosure. FIG. 1C is a schematic diagram illustrating an existing deep learning based trajectory network generation architecture according to an embodiment of the present disclosure.

Currently, there are applications (e.g., various applications typified by an Tencent map, a Baidu map, etc.) that can provide respective travel plans for various travel modes of a user, and the user can initiate various travel requests (e.g., a navigation request shown in FIG. 1B) among the applications installed on his user terminal. The user terminal may then transmit its data, e.g. its location data and input data including location information such as start and end points, to the server of the application over the network.

Optionally, the user terminal may specifically include a smartphone, a tablet, a laptop portable computer, a vehicle-mounted terminal, a wearable device, and so on. The user terminal may also be a client that installs a browser or various applications, including system applications and third party applications. The network may be an Internet of Things (Internet of Things) based on the Internet and/or a telecommunication network, which may be a wired network or a wireless network, for example, which may be a Local Area Network (LAN), a Metropolitan Area Network (MAN), a Wide Area Network (WAN), a cellular data communication network, or other electronic networks capable of implementing information exchange functions. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as cloud service, a cloud database, cloud computing, a cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and a big data and artificial intelligence platform.

As shown in fig. 1A, the server may generate corresponding navigation data on-line based on received user data, for example, location data of a user and various input data, such as location information of a start point and an end point set by the user, a travel pattern, and the like, and the navigation data may include a variety of travel plans determined by the server based on the user data. Then, the server may return the determined navigation data of the plurality of travel plans to the user terminal through the network and display the same as a navigation response to the navigation request of the user.

The interface displaying the various travel plans to the user may be an interface as shown in fig. 1B. The user may request navigation by entering a start point and an end point location, respectively, within the upper address box in the interface (e.g., in fig. 1B, the start point is the user's current location (shown as "my location") and the end point is the a subway station). The server may determine different travel plans for the user according to the selection of the travel mode by the user, for example, the travel plan when the user selects the walking mode may include a plurality of paths which cannot be passed by the motor vehicle. As shown in fig. 1B, two travel plans (plan one and plan two) returned from the server are shown in the interface, where each travel plan includes a corresponding travel time, travel distance, and number of intersections traveled. The interface shows the navigation route in the walking mode by taking the scheme one as an example, wherein the current position of the user, namely the starting point, is represented by a positioning symbol, the destination position, namely the end point, is represented by a mark 'end', and the black bold curve between the starting point and the end point represents the navigation route corresponding to the current travel scheme.

In the process of providing travel navigation for the user by using the application, data in the server of the application provides basic support for navigation, and in turn, the actual travel process of the user according to the travel navigation provides development directions for data expansion of the application. For example, it is desirable to provide missing road intelligence quickly, stably, and comprehensively for the missing road problem typical of such applications. The generation of a trajectory network throughout the missing road excavation process may be considered as the basis for implementing missing road excavation. As one of the infrastructures constituting ITS, the track network plays a very critical role in many applications. Many applications that provide services to users based on a track network require a high level of precision and accuracy for the track network, which may otherwise cause inconvenience and loss to the users due to the wrong information obtained from the track network.

In the existing trajectory network generation technology, a trajectory network generation method based on Kernel Density Estimation (KDE) and a trajectory network generation method based on deep learning may be included. The KDE-based track network generation method comprises the steps of firstly generating a track histogram by using the original track data through a kernel density track, then carrying out filtering processing and image processing to obtain final track network data, deducing the distribution of overall data based on limited samples, wherein the result of kernel density estimation is the probability density function estimation of the samples. However, the above-mentioned method for generating a trajectory road network uses a lot of artificial experience parameters without considering road network priori knowledge, resulting road network morphology is poor, and has a great problem in adaptivity, noise immunity and the like for a traffic-intensive area with a large data volume, and at the same time, there is no capability of automatically producing a trajectory road network with a large data volume (for example, nationwide).

In addition, a processing method based on deep learning exists in the conventional track network generation technology, as shown in fig. 1C. The whole processing architecture based on deep learning is transmitted and stored based on clusters, the processing flow of the processing architecture can comprise three parts of track modeling, deep learning and road network vectorization, as the processed data objects are a plurality of small file objects, a large amount of frequent read-write operations with the clusters and local message interaction exist in the processing process, and the whole processing flow is processed in series, the processing of the processing architecture is complex and the maintenance is difficult. Specifically, for the three parts, a calculation type task is executed in the track modeling, provinces within the country are divided by adopting a province operation (the method shown in fig. 1C is divided into two provinces) and corresponding track density maps are obtained through calculation respectively, then merging processing is carried out based on the track density map results of the parts to generate a national track density map, and the generated national track density map still needs to be transmitted to a next processing module through a cluster. In the deep learning part, two CNNs are adopted to carry out image processing twice to obtain an approximately binary road and background image, wherein data interaction between the two CNNs and a cluster also needs dumping through an intermediate storage platform, the involved thread (pulling, calculating, pushing and the like) processing and repeated disk dropping of transmission and storage are very complicated and difficult to check and maintain, and the cluster work efficiency is reduced because the cluster easily fails to access a small file object due to multi-node multi-time concurrent pulling and pushing operations of the small file object. Due to the serial processing of the parts in the whole processing flow, the method for generating road network data consumes a lot of time and is low in robustness, and when the method is applied to road network track generation related to a plurality of travel modes, due to the limited available resources, there may be resource contention and excessive waiting time between the various processes of track network generation, because before one process is completed, the other process cannot be started.

As described above, most of the existing methods for generating a trajectory network are experimental schemes, and there is no capability of automatically generating a trajectory network with a large data volume (e.g., nationwide).

The present disclosure is based on the above and provides a trajectory data processing method for generating a part of a trajectory network in units of blocks by using a transmission structure of a message queue to generate a plurality of parts of the trajectory network by asynchronous processing, thereby forming a whole amount of trajectory network nationwide.

FIG. 2 is a flow chart illustrating a trajectory net generation method 200 according to an embodiment of the present disclosure.

FIG. 3A is a schematic diagram illustrating a processing architecture of a trajectory net generation method according to an embodiment of the present disclosure; FIG. 3B is a flow chart illustrating the implementation of trace-road network generation through message queues according to an embodiment of the present disclosure.

Before describing the method steps of the method 200 of generating a road network according to the present disclosure in detail, the processing architecture of the method 200 of generating a road network according to the present disclosure may be set forth in its entirety.

As shown in fig. 3A, the processing architecture of the trajectory network generation method 200 is optimized and reconstructed based on the existing deep learning-based trajectory network generation architecture shown in fig. 1C, and a data structure with a message queue added among three processing modules, namely, trajectory modeling, deep learning and network vectorization, is used as a message middleware for data transmission, so that data pulling and uploading among the three processing modules and operations among the processing modules can be performed asynchronously. The data generated by the deep learning can be synchronously and concurrently transmitted to a road network vectorization module through the message queue to generate a vector track road network. Although the generation of the trajectory road network is not required to be performed by data dropping, in order to facilitate the information backtracking of the service information, it is necessary to store the intermediate data generated by the three processing modules, for example, to a unified Storage system (for example, a Cloud Object Storage (COS) system, which is applicable to various scenarios such as big data calculation and analysis).

As described above, compared to the cluster-based data storage and transmission structure in the existing deep learning based trajectory network generation architecture as shown in fig. 1C, the processing architecture of the trajectory network generation method 200 of the present disclosure separates data storage from data transmission so that the storage and transmission operations do not interfere with each other.

In addition, as shown in fig. 3A, the processing architecture of the trajectory network generation method 200 of the present disclosure further applies a monitoring alarm mechanism to each of the three processing modules, wherein for a deep learning processing module with more complex data processing and more involved processing hardware (e.g., GPU for image processing), a heartbeat monitoring mechanism is employed to ensure the normal operation of the processing module, and for two processing modules of trajectory modeling and vectorization network, an alarm notification mechanism is employed to ensure the normal operation of the processing module.

As shown in FIG. 2, in step 201, trajectory data may be acquired, which may include a first number of trajectory data sets.

Alternatively, the trajectory data may be a set of a series of position data and time stamps, and the data volume of the trajectory data is generally large and has noise and interference. Therefore, before data mining is performed on the track data, the track data can be cleaned and filtered, and abnormal points in the track data can be screened out by using a specific algorithm, so that high-precision track data can be obtained.

Alternatively, the sources of these trajectory data may generally include cars, trucks, buses, public transportation, walks, rides, etc., and thus the trajectory data may be classified into data such as low-speed trajectory data (e.g., walking and riding trajectory data, etc.) and high-speed trajectory data (e.g., driving trajectory data, etc.) based on the user's travel patterns.

According to an embodiment of the present disclosure, the trajectory data may be walking and riding trajectory data, or vehicle trajectory data.

According to an embodiment of the present disclosure, the trajectory network respectively generated based on the walking and riding trajectory data and the vehicle trajectory data may be different. This is because the road network is a road system in which various roads, which are infrastructures for various vehicles and pedestrians to pass through (for example, urban roads, highways, factory roads, forest roads, rural roads, and the like according to their usage characteristics), are interconnected and distributed in a mesh shape in a certain area, and the roads that the vehicles and the pedestrians can pass through are different from each other, and thus the trajectory road network generated based on the roads on which the specific object passes through is also different.

Alternatively, trajectory networks for different travel modes (e.g., walking, driving, etc.) may be generated based on different trajectory data categories, so that when providing travel plans and navigation routes for users, the users can be selectively served based on a specific trajectory network according to actual needs of the users.

Alternatively, for a large amount of data track data covering a large area (e.g., nationwide), the generated track network is difficult to generate completely at one time, and generally, a part of the track network may be generated in units of a map sheet (mesh), where the map sheet may be a part of the coverage area (e.g.,

adjacent map sheets

1 and 2 shown in fig. 7A later, which are two adjacent partial track networks in the track network), so that a plurality of map sheets generated based on all the track data may constitute a complete track network. For the trajectory data of the present disclosure, the image may be further divided into more smaller image blocks, wherein the trajectory data included in each image block corresponds to one trajectory data set, and thus, the trajectory data of the present disclosure may include a first number of trajectory data sets.

In step 202, for each of the first number of trajectory data sets, a corresponding three-channel color image block may be generated based on the trajectory data sets.

Fig. 4A is a schematic diagram illustrating generating a mesh image based on pairs of trajectory points according to an embodiment of the present disclosure. FIG. 4B is a schematic diagram illustrating generating a three-channel color image block based on a grid image according to an embodiment of the disclosure. Fig. 4C is a schematic diagram illustrating an overlap between three-channel color image blocks according to an embodiment of the present disclosure.

According to an embodiment of the disclosure, the trajectory data in each of the first number of trajectory data sets may comprise a plurality of pairs of trajectory points.

Alternatively, the trajectory data in each trajectory data set may exist in the form of pairs of trajectory points. As shown in the right-hand trace indicated with the correct symbol in fig. 4A, by connecting the trace points in each pair of trace points two by two, a plurality of traces in the image block can be formed based on the trace data.

As described above, the conventional complex redundant process of generating a trajectory by using trajectory data matched by a network can be avoided by the trajectory generation method, in which a trajectory is generated only after trajectory data points are merged and sorted to form a point string, and this operation is no longer required in the trajectory data already existing in the form of trajectory point pairs. Therefore, the traditional track network matching problem can be avoided through the method of the embodiment of the disclosure.

As shown in the middle part of fig. 4A (spatial slicing and grid management), one image block (the largest square in fig. 4A) can be evenly sliced into a certain number of grids, which is the simplest one of data structures for storing geospatial information. Each grid may be defined as a pixel, the location of which may be marked by rows and columns. The track segments may be spatially mapped to a block of pixels of a certain size in the grid image, for example a block of pixels containing 9 pixels in fig. 4A.

Optionally, the precision of the trajectory road network and the fineness of the cell grid division have a direct relationship, and the finer the grid division, the smaller the area range covered by the cell grid, the higher the precision of the grid image. Therefore, the requirement of using the grid data structure to store data information with high precision is large.

According to an embodiment of the present disclosure, generating a corresponding three-channel color image block based on the trajectory dataset may include: generating a three-channel color image block corresponding to the trajectory data set based on a plurality of pairs of trajectory points in the trajectory data set. Wherein each pixel in the three-channel color image block has three-dimensional features corresponding to flow, speed, and direction of a trajectory at the pixel.

As shown in fig. 4B, an image block in each image frame may include a certain number of pixels, and the multi-channel feature of each pixel may be obtained by performing multi-channel feature modeling on trajectory data at the pixel. For example, the multi-channel features of each pixel may be generated based on the average speed, traffic flow, traffic direction, etc. of the trajectory of each pixel in the image block over a predetermined time range to generate a color rendering, i.e., trajectory modeling, for the image block.

Alternatively, each pixel in the image block may have three-dimensional features including a flow feature, a velocity feature in the X direction (denoted as velocity X feature in the figure), and a velocity feature in the Y direction (denoted as velocity Y feature in the figure). Thus, the color rendering of the generated image block is a three-channel color image block.

Alternatively, the three channel color image block may be an RGB (red green blue) three channel color image block. Where the R channel (red component) represents a flow map layer whose physical meaning represents the actual flow value within the geospatial region. The specific construction mode is that each track point of each track is traversed, accumulation is carried out in a space area, and at each pixel, the larger the value (0-255) is, the larger the flow passing through the pixel is, and the larger proportion of the red component is presented in the visual effect. The B channel (blue component) represents an X-direction (for example, due north) vector velocity map layer, the physical meaning of the map layer represents the projection speed of the actual traffic speed in the geographic space region in the X direction, the specific construction mode is that each track point of each track is traversed, the projection speed of the track point in the X direction is calculated, average cumulative calculation is carried out, at each pixel, the larger the value (0-255) is, the larger the traffic speed component in the X direction at the current pixel is, and the larger the proportion of the blue component is presented in the visual effect. The G channel (green component) represents a Y-direction (for example, the east direction) vector velocity map layer, the physical meaning of the map layer represents the projection speed of the actual traffic speed in the geographic space region in the Y direction, the specific construction mode is that each track point of each track is traversed, the projection speed of the track point in the Y direction is calculated, average cumulative calculation is carried out, at each pixel, the larger the value (0-255) is, the larger the traffic speed component in the Y direction representing the current position is, and the larger the proportion of the green component in the visual effect is presented.

According to an embodiment of the present disclosure, there is an overlapping portion between the three-channel color image block and at least one other three-channel color image block, the at least one other three-channel color image block including at least one three-channel color image block adjacent to at least one of the upper, lower, left, and right sides of the three-channel color image block.

Due to the above-described blocking process, edge distortion may occur at boundary positions in the generated trajectory network corresponding to adjacent blocks, that is, if two adjacent image blocks are independent (that is, there is no common trajectory data), the trajectory network generated based on the two image blocks may not be aligned at the adjacent boundaries of the two image blocks. Therefore, in order to eliminate the above-mentioned edge distortion phenomenon, the division of the image blocks may be optimized such that two adjacent image blocks have the same trajectory data on their adjacent boundaries, for example, four image blocks marked with (r), (c), and (c) in fig. 4B, where each two adjacent image blocks have a shared edge.

Specifically, as shown in fig. 4C, (a) shows a division example of independent multiple image blocks, and (b) is an improvement of the image block division method of (a) as described herein. In (b), the coverage of the original image block may be enlarged (as indicated by the dotted line and the arrow portion), so that there may be an overlapping portion (shown as a darker shaded portion in the figure) between two adjacent image blocks.

Through the overlapping processing, the distortion of the adjacent three-channel color image blocks at the boundary can be eliminated, so that the generated track network curve is smoother.

In step 203, a corresponding binarized image block may be generated by deep learning based on the three-channel color image block.

FIG. 5A is a schematic diagram illustrating generation of a binarized image block based on a three-channel color image block according to embodiments of the present disclosure. Fig. 5B is a flow diagram illustrating generation of a binarized image block based on a three-channel color image block according to an embodiment of the present disclosure.

In the existing deep learning-based trajectory road network generation method as described previously, the processing section of deep learning includes a two-step serial process for performing two divisions of roads and background as shown in (a) in fig. 5A. The first image segmentation processing part firstly draws a track density graph of each part generated by track modeling from the cluster, and completes the combination of images of each part after the processing of a first CNN model (CNN 1) so as to obtain a gray scale graph with the pixel value of each pixel within the range of 0-255, and after the processing of the step is completed, the local data is required to be deleted and the data of the cluster is required to be uploaded. In the second image division processing part, almost the same operation as that of the first image division processing part is actually performed, in which the result of the first image division which is uploaded to the cluster by the first image division processing part is first pulled from the cluster, the second CNN model (CNN 2) which is subjected to the second image division processing after the image division and the respective images after the model processing are merged, and the deletion of local data and the upload of data to the cluster are performed after the processing is completed.

As described above, in the deep learning processing step in the existing deep learning based trajectory road network generation method, the processing time actually consumed in the data pulling and uploading with the cluster and the data deleting part may be more than the actual CNN model processing time, especially in the case of small file processing, frequent data input and output is more time-consuming, and the two-step serial processing further doubles the running time.

Therefore, in the processing architecture shown in fig. 3A, the deep learning processing module in the track network generation method 200 of the present disclosure implements asynchronous pulling and uploading of data through the message queue, thereby avoiding the above two-step serial processing. Specifically, (b) in fig. 5A shows a deep learning processing flow in the trajectory road network generation method 200 of the present disclosure.

Thus, in accordance with an embodiment of the present disclosure, step 203 may include operations that are performed asynchronously as shown in fig. 5B.

Optionally, the deep learning processing part of the trajectory network generation method 200 of the present disclosure implements fast data pulling and data uploading asynchronously through data interaction with the message queue (shown as the first message queue in fig. 5A (b)).

As shown in fig. 5B, a second number of three-channel color image blocks may be read out from the first message queue in step 501.

For example, in fig. 5A (b), a second number of processes may first be pulled from the first message queue to await image segmentation processing.

In step 502, M binarized image patches that correspond one-to-one to M of the second number of three-channel color image patches may be generated by a plurality of neural networks trained in advance.

According to an embodiment of the present disclosure, the selection of M of the second number of three-channel color image blocks may be achieved through a message queue (e.g., shown as q1 in (b) of fig. 5A).

Alternatively, depending on the actual processing power of the limited processing unit for running the CNN model, the process (three-channel color image block) that was first pulled from the first message queue may be first streamed to the processing unit via the message queue q1 for segmentation of the road from the background. For example, the processing unit may also be capable of processing the M three-channel color image blocks simultaneously at a particular time, and thus the M three-channel color image blocks pulled first out of the second number of three-channel color image blocks may be input to the processing unit via the message queue q 1. Message queues the application in this disclosure is similar to the caching application of a CPU.

According to an embodiment of the present disclosure, step 502 may include: generating, by a pre-trained first neural network, M gray scale image blocks in one-to-one correspondence with M of the second number of three-channel color image blocks, wherein each pixel in each gray scale image block has a value in a range of 0 to 255; and generating M binarized image blocks in one-to-one correspondence with the M gray image blocks through a pre-trained second neural network, wherein each pixel in each binarized image block has a value between 0 and 1.

Optionally, for various scenes (travel patterns) or various lines of business, the neural network models may be trained based on different categories of trajectory data training sets, so that trajectory road network data may be provided for the scenes more accurately, thereby improving user experience.

Optionally, a first neural network (shown as cnn 1) may be used to generate a patch of gray scale images based on a three-channel color image patch, and a second neural network (shown as cnn 2) may further generate an approximately binarized image patch based on the patch of gray scale images generated by the first neural network. The segmentation of the road and the background in the original track density graph can be realized through the first neural network and the second neural network. Wherein intermediate messages between the two image segmentation processes (cnn 1 and cnn 2) are not repeatedly stored and read any more, so that the computing unit can perform the computing task with higher efficiency.

Alternatively, the neural network model employed for deep learning may be a semantic segmentation neural network such as D-LinkNet, which processes the road extraction task as a two-classification semantic segmentation task to generate pixel-level labels for roads. It should be understood that the above-mentioned D-LinkNet is used as an example only in the present disclosure, and is not limited thereto, and other neural network models that can achieve similar effects can be also applied to the trajectory network generation method of the present disclosure.

In step 503, a third number of binarized image blocks may be written into the first message queue, the third number of binarized image blocks including N of the M binarized image blocks.

According to an embodiment of the present disclosure, M and N are natural numbers.

Alternatively, in this processing portion, a third number of processes that have completed the image segmentation process may be uploaded to the first message queue concurrently with the pulling from the data in the first message queue.

According to an embodiment of the present disclosure, the selection of N of the M binarized image blocks may be implemented through another message queue (e.g., shown as q2 in (b) of fig. 5A) similar to that described with reference to step 502.

Alternatively, the data transmission inside the deep learning processing module may still be implemented by message queues (e.g., using two message queues q1 and q 2), so that a certain number of processes can be pulled from the second number of processes for real-time processing according to the actual processing capability of the processing unit, and the processes that completed image segmentation first are uploaded online to an intermediate message queue (shown as q 2) to flow into the first message queue through the intermediate message queue for subsequent processing.

According to the embodiment of the disclosure, during the period of generating the corresponding binary image blocks through the deep learning based on the three-channel color image blocks, heartbeat detection is performed on the plurality of relevant GPUs to detect the operating states of the plurality of GPUs.

Alternatively, a heartbeat monitoring thread may be added to the storage system by writing a file to detect the operating state of a processing unit (e.g., GPU) running the neural network model, thereby enabling a fast response to a failure (e.g., a fast restart of a particular processing unit) to ensure proper operation of the process.

According to an embodiment of the present disclosure, the generated three-channel color image block and the binarized image block may be transmitted through a first message queue.

Accordingly, by the application of the message queue, unnecessary input-output consumption is avoided, the burden on the processing unit is reduced and the work efficiency thereof is improved, so that more efficient deep learning can be realized with less computing resources, for example, by the deep learning process shown in (b) in fig. 5A, the number of GPUs originally used for image segmentation of three-channel color image blocks can be reduced, while deep learning of nationwide trajectory modeling data can be realized in a shorter time.

In step 204, vectorizing the binarized image blocks may be performed to generate a partial trajectory network, the partial trajectory network of the first number corresponding to the first number of trajectory data sets constituting the trajectory network.

Since the road in the binarized image block obtained based on the neural network model through the deep learning process in step 203 is determined based on a certain probability, the binarized image block needs to be subjected to the vectorization process to convert the probabilistic road into a real-world vector road network.

Vectorizing the binarized image block to generate a partial trajectory network according to an embodiment of the present disclosure may include extracting a plurality of road center lines from the binarized image block, the extracted plurality of road center lines constituting a network topology corresponding to the binarized image block; and performing point-edge decomposition on the road network topological graph, and generating the partial track road network based on morphological analysis and topological analysis.

Optionally, the vectorization of the road network is mainly to extract roads, and includes first extracting road center lines of the probabilistic road network generated by the deep learning module, where the extracted road center lines form a road network topological graph, where the road network topological graph may include a point-edge set composed of intersections such as road network center lines and line segments of the road network center lines divided by the intersections.

Optionally, the road vectorization of the probabilistic road network may be implemented by performing point-edge decomposition on the road network topological graph based on the point-edge set, and performing morphological analysis and topological analysis on new points and edges obtained by the decomposition.

Alternatively, in order to ensure the quality of road lines in the road network, methods such as adaptive threshold, skeletonization algorithm, connected domain analysis, and bezier smoothing may be adopted to process the binarized image block. In addition, for the normalized output of the road network, 8 neighborhood analysis, road aggregation and splitting and rarefying algorithms can be applied to the binary image blocks. It should be understood that the above algorithm is only used as an example of the algorithm adopted by the road vectorization processing part in the track network generation method of the present disclosure, and is not limited.

As described above, the image data completing the deep learning may be transmitted to the road network vectorization part through the first message queue for vectorization processing.

Fig. 6 is a schematic diagram illustrating vectorization processing according to an embodiment of the present disclosure.

Optionally, the road network vectorization part may implement simultaneous processing of image data of multiple processes by using a streaming cluster in which multiple processing nodes (shown as image vectorization 1, …, image vectorization Z in fig. 6) are distributed and deployed in a large data cluster streaming processing manner, so as to greatly reduce time required for road network vectorization, and even implement road network vectorization while completing deep learning of a last trajectory density map.

Optionally, all the generated partial trajectory network may be stored to the file storage system via the object storage system as the finally generated trajectory network.

As described above, the specific operations in the steps of the trajectory network generation method according to the embodiment of the present disclosure, that is, the operations in the three processing modules of trajectory modeling, deep learning and network vectorization in the processing architecture of the trajectory network generation method 200 shown in fig. 3A are described.

Since the processing architecture of the trajectory network generation method 200 of the present disclosure implements asynchronous data pulling and uploading between three processing modules by adding a message queue as a message middleware between the three processing modules of trajectory modeling, deep learning and network vectorization for data transmission, in addition to the description presented above with respect to step 202 and 204, according to an embodiment of the present disclosure, step 202 and 204 in the trajectory network generation method 200 may further include an asynchronously executed operation as shown in fig. 3B.

In step 301, a first generated three-channel color tile of the first number of three-channel color tiles generated for the first number of track data sets may be written first into a first message queue.

Optionally, for trajectory modeling of a first number of trajectory data sets, the three-channel color image block generated first may flow into the first message queue first, and flow into the deep learning processing module first when the deep learning processing may be performed.

In step 302, a three-channel color image block may be read out from the first message queue to generate a binarized image block based on the three-channel color image block, and a first generated binarized image block of a first number of binarized image blocks generated for the first number of three-channel color image blocks is written into the first message queue first.

Alternatively, the three-channel color tile read from the first message queue may be the earliest of all three-channel color tiles currently flowing into the first message queue in the first message queue.

As described above, the binarized image block, which also completes the deep learning first, is first streamed into the first message queue. It can be understood that the order of the binary image blocks flowing into the first message queue is not necessarily the same as the order of the three-channel color image blocks corresponding to the binary image blocks flowing into the first message queue, because the operations in the three processing modules of trajectory modeling, deep learning and road network vectorization are asynchronous and concurrent, and data is transmitted between the three processing modules similarly to water flow.

In step 303, a binarized image block may be read from the first message queue, and vectorized to generate the partial trajectory network.

Similarly, the binarized image block that first flows into the first message queue flows into the vectorization processing module when vectorization processing is available to generate its corresponding partial trajectory network.

As described above, in the entire process flow of the trajectory network generation method of the present disclosure, a synchronized vector trajectory network can be generated by generating the first image block of the trajectory model. After each of the first number of trajectory data sets has been streamed from the trajectory modeling processing module to the vectorization processing module and vectorized, a complete trajectory network corresponding to the first number of trajectory data sets may be generated.

FIG. 7A is a diagram illustrating an example trajectory net generated in accordance with an embodiment of the present disclosure; fig. 7B is a schematic diagram illustrating a comparison of the generated trajectory network before and after the application of the trajectory network generation method according to the embodiment of the present disclosure.

As shown in fig. 7A, two geographically adjacent drawings (drawing 1 and drawing 2) are used as examples to illustrate a part of the trajectory network generated by the trajectory network generation method according to the embodiment of the disclosure. Wherein the road is shown with white lines and the black part is a background part except the road.

In fig. 7A, two map sheets may be aligned with each other, so that a plurality of partial trajectory network generated by the trajectory network generation method according to the embodiment of the disclosure may form a complete and smooth-lined trajectory network graph.

Fig. 7B shows a part of the trajectory network generated before the trajectory network generation method according to the embodiment of the present disclosure is applied, and a part of the trajectory network generated after the trajectory network generation method according to the embodiment of the present disclosure is applied, in (a) and (B).

As shown in fig. 7B, (B) has more complete lines (e.g., a comparison of white sharp points) than (a), (a) originally discontinuous short line segments can all dig out complete roads based on the track network generation method of the present disclosure, and this is achieved on the basis of less time cost consumption.

In addition, by simply switching the track data source, track road networks for various scenes (travel modes) can be provided, and nationwide road network data support can be provided for various service lines.

Therefore, the track road network generation method of the embodiment of the disclosure provides a strong basic support for subsequent data mining.

FIG. 8 is a schematic flow chart diagram illustrating subsequent data mining in accordance with an embodiment of the present disclosure.

As shown in fig. 8, the data mining process may mainly include track preprocessing, track network generation, network differentiation, confidence screening, intelligence hooking, and intelligence operation online. The generation of the track road network is used as the basis of the subsequent data mining processing, and has great influence on the whole data mining processing.

For example, for the problem of missing roads in typical walking and riding in the navigation function of map application and the like, the more comprehensive and more accurate track network generated by the track network generation method according to the embodiment of the disclosure lays a good foundation for quickly realizing missing road excavation work.

Fig. 9 is a schematic diagram illustrating a trajectory net generating device 900 according to an embodiment of the present disclosure.

The trajectory road network generating device 900 may comprise a data acquiring module 901, an image generating module 902, a background segmenting module 903 and a road network generating module 904.

According to an embodiment of the present disclosure, the data acquisition module 901 may be configured to acquire trajectory data including a first number of trajectory data sets.

Alternatively, the trajectory data may be trajectory data of walking and riding, or trajectory data of vehicle running.

For example, the sources of these trajectory data may generally include cars, trucks, buses, public transportation, walks, rides, etc., and thus the trajectory data may be classified into data such as low-speed trajectory data (e.g., walking and riding trajectory data, etc.) and high-speed trajectory data (e.g., driving trajectory data, etc.) based on the user's travel patterns.

The image generation module 902 may be configured to, for each of the first number of trajectory data sets, generate a corresponding three-channel color image patch based on the trajectory data set.

Alternatively, each pixel in the generated three-channel color image block may have three-dimensional features including a flow feature, a velocity feature in the X-direction (denoted as velocity X-feature in the figure), and a velocity feature in the Y-direction (denoted as velocity Y-feature in the figure). For example, the three channel color image block may be an RGB (red green blue) three channel color image block.

Optionally, the image generation module 902 may be configured to perform operations as described with reference to step 202.

The background segmentation module 903 may be configured to generate a corresponding binarized image block by deep learning based on the three-channel color image block.

Optionally, the deep learning processing module in the track road network generating method 200 of the present disclosure implements asynchronous pulling and uploading of data through a message queue. Specifically, the operations performed by the background segmentation module 903 may refer to the operations performed asynchronously of the deep learning process flow as described with reference to (b) in fig. 5A.

The road network generating module 904 may be configured to vectorize the binarized image blocks to generate a partial road network of tracks, the partial road network of tracks of the first number corresponding to the first number of sets of track data constituting the road network of tracks.

Optionally, the road network generating module 904 may be configured to perform the operations as described with reference to step 204.

For example, the road vectorization of the probabilistic road network can be realized by performing point-edge decomposition on the road network topological graph based on the point-edge set, and performing morphological analysis and topological analysis on new points and edges obtained by decomposition.

According to the embodiment of the present disclosure, the generated three-channel color image block and the generated binarized image block are transmitted among the image generating module 902, the background segmenting module 903, and the road network generating module 904 through a first message queue.

According to an embodiment of the present disclosure, the image generation module 902, the background segmentation module 903 and the road network generation module 904 asynchronously perform the following operations: the image generation module 902 writes a first three-channel color image block of a first number of three-channel color image blocks generated for the first number of trajectory data sets into a first message queue; the background segmentation module 903 reads out a three-channel color image block from the first message queue to generate a binarized image block based on the three-channel color image block, and writes a first generated binarized image block of a first number of binarized image blocks generated for the first number of three-channel color image blocks into the first message queue; and the road network generating module 904 reads out a binarized image block from the first message queue, and performs vectorization on the binarized image block to generate the partial trajectory road network.

As described above, by adding a message queue as message middleware for data transmission among the three processing modules of the image generation module 902, the background segmentation module 903 and the road network generation module 904, asynchronous data pulling and uploading among the three modules can be realized.

According to still another aspect of the present disclosure, a trajectory network generating device is also provided. FIG. 10 shows a schematic diagram of a trajectory net generation device 2000 according to an embodiment of the present disclosure.

As shown in FIG. 10, the trajectory net generation device 2000 may include one or more processors 2010 and one or more memories 2020. Wherein said memory 2020 has stored therein computer readable code which when executed by said one or more processors 2010 may perform a trajectory network generation method as described above.

The processor in the embodiments of the present disclosure may be an integrated circuit chip having signal processing capabilities. The processor may be a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present application may be implemented or performed. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, which may be of the X86 or ARM architecture.

In general, the various example embodiments of this disclosure may be implemented in hardware or special purpose circuits, software, firmware, logic or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device. While aspects of embodiments of the disclosure have been illustrated or described as block diagrams, flow charts, or using some other pictorial representation, it is well understood that the blocks, apparatus, systems, techniques or methods described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.

For example, a method or apparatus in accordance with embodiments of the present disclosure may also be implemented by way of the architecture of computing device 3000 shown in fig. 11. As shown in fig. 11, computing device 3000 may include a bus 3010, one or more CPUs 3020, a Read Only Memory (ROM) 3030, a Random Access Memory (RAM) 3040, a communication port 3050 to connect to a network, input/output components 3060, a hard disk 3070, and the like. A storage device in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used in the processing and/or communication of the trajectory network generation method provided by the present disclosure, as well as program instructions executed by the CPU. Computing device 3000 can also include user interface 3080. Of course, the architecture shown in FIG. 10 is merely exemplary, and one or more components of the computing device shown in FIG. 11 may be omitted as needed in implementing different devices.

According to yet another aspect of the present disclosure, there is also provided a computer-readable storage medium. Fig. 12 shows a schematic diagram 4000 of a storage medium according to the present disclosure.

As shown in fig. 12, the computer storage medium 4020 has stored thereon computer readable instructions 4010. The computer readable instructions 4010, when executed by a processor, may perform a trajectory network generation method according to embodiments of the present disclosure described with reference to the above figures. The computer readable storage medium in embodiments of the present disclosure may be either volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be Random Access Memory (RAM), which acts as external cache memory. By way of example and not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Synchronous Link Dynamic Random Access Memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memories of the methods described herein are intended to comprise, without being limited to, these and any other suitable types of memory. It should be noted that the memories of the methods described herein are intended to comprise, without being limited to, these and any other suitable types of memory.

Embodiments of the present disclosure also provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the trajectory net generation method according to the embodiment of the present disclosure.

It is to be noted that the flowchart 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 disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises at least one executable instruction for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The exemplary embodiments of the present disclosure described in detail above are merely illustrative, and not restrictive. It will be appreciated by those skilled in the art that various modifications and combinations of these embodiments or features thereof may be made without departing from the principles and spirit of the disclosure, and that such modifications are intended to be within the scope of the disclosure.

Claims

1. A method for generating a trajectory network comprises the following steps:

obtaining trajectory data, the trajectory data comprising a first number of trajectory data sets;

for each of the first number of trajectory data sets, generating a corresponding three-channel color patch based on the trajectory data set, each pixel in the three-channel color patch having a three-dimensional feature corresponding to a flow, a speed, and a direction of a trajectory at the pixel;

generating a corresponding binary image block through deep learning based on the three-channel color image block; and

vectorizing the binarized image blocks to generate a partial trajectory network, wherein the trajectory network is composed of a first number of partial trajectory networks corresponding to the first number of trajectory data sets.

2. The method of claim 1 wherein the generated three channel color tile and binarized tile are transmitted through a first message queue, the method comprising the following operations performed asynchronously:

writing a first three-channel color image block of a first number of three-channel color image blocks generated for the first number of track data sets into a first message queue;

reading a three-channel color image block from the first message queue to generate a binary image block based on the three-channel color image block, and writing a first generated binary image block of a first number of binary image blocks generated for the first number of three-channel color image blocks into the first message queue; and

and reading a binary image block from the first message queue, and carrying out vectorization on the binary image block to generate the partial track network.

3. The method of claim 2, wherein the trajectory data in each of the first number of trajectory data sets includes a plurality of pairs of trajectory points, generating a corresponding three-channel color image patch based on the trajectory data sets comprising:

generating a three-channel color image block corresponding to the trajectory data set based on a plurality of trajectory point pairs in the trajectory data set;

and the three-channel color image block and at least one other three-channel color image block are overlapped, and the at least one other three-channel color image block comprises at least one three-channel color image block adjacent to at least one of the upper side, the lower side, the left side and the right side of the three-channel color image block.

4. The method of claim 3, wherein generating a corresponding binarized image block by deep learning based on the three-channel color image block comprises the following operations performed asynchronously:

reading a second number of three-channel color image blocks from the first message queue;

generating M binaryzation image blocks corresponding to M of the second number of three-channel color image blocks one by one through a plurality of pre-trained neural networks; and

writing a third number of binarized image blocks into the first message queue, wherein the third number of binarized image blocks comprises N of the M binarized image blocks;

wherein, the selection of the M of the three-channel color image blocks of the second number and the N of the M binarization image blocks is realized by two message queues;

wherein M and N are natural numbers.

5. The method of claim 4, wherein generating M binarized patches of images that correspond one-to-one with M of the second number of three channel color patches through a pre-trained plurality of neural networks comprises:

generating, by a pre-trained first neural network, M gray scale image blocks in one-to-one correspondence with M of the second number of three-channel color image blocks, wherein each pixel in each gray scale image block has a value in a range of 0 to 255; and

generating M binarized image blocks in one-to-one correspondence with the M gray image blocks through a pre-trained second neural network, wherein each pixel in each binarized image block has a value between 0 and 1.

6. The method of claim 4 wherein during generation of a corresponding binarized tile by deep learning based on the three channel color tiles, heartbeat detection is performed on an associated plurality of GPUs to detect an operational state of the plurality of GPUs.

7. The method of claim 1, wherein vectorizing the binarized image block to generate a partial trajectory network comprises:

extracting a plurality of road center lines from the binarized image block, wherein the extracted road center lines form a road network topological graph corresponding to the binarized image block; and

and performing point-edge decomposition on the road network topological graph, and generating the partial track road network based on morphological analysis and topological analysis.

8. The method of claim 1, wherein the trajectory data is walking and riding trajectory data or is vehicle trajectory data;

the track network generated based on the walking and riding track data and the vehicle track data is different.

9. A trajectory network generating device comprising:

a data acquisition module configured to acquire trajectory data, the trajectory data comprising a first number of trajectory data sets;

an image generation module configured to, for each of the first number of trajectory data sets, generate a corresponding three-channel color patch based on the trajectory data set, each pixel in the three-channel color patch having a three-dimensional feature corresponding to a flow, a speed, and a direction of a trajectory at the pixel;

the background segmentation module is configured to generate a corresponding binarization image block through deep learning based on the three-channel color image block; and

a road network generating module configured to perform vectorization on the binarized image blocks to generate a partial trajectory road network, the partial trajectory road network of a first number corresponding to the first number of trajectory data sets constituting the trajectory road network.

10. The apparatus of claim 9, wherein the generated three channel color image patch and binarized image patch are transmitted between the image generation module, the background segmentation module, and the road network generation module via a first message queue, the image generation module, the background segmentation module, and the road network generation module asynchronously performing the following:

the image generation module writes a first three-channel color image block of a first number of three-channel color image blocks generated for the first number of track data sets into a first message queue;

the background segmentation module reads a three-channel color image block from the first message queue to generate a binary image block based on the three-channel color image block, and writes a binary image block generated earlier in a first number of binary image blocks generated for the first number of three-channel color image blocks into the first message queue; and

and the road network generation module reads out a binarization image block from the first message queue, and vectorizes the binarization image block to generate the partial track road network.

11. The apparatus of claim 10, wherein the trajectory data in each of the first number of trajectory data sets includes a plurality of pairs of trajectory points, the image generation module generating the corresponding three-channel color image patch based on the trajectory data sets comprising:

12. The apparatus of claim 11, wherein the background segmentation module to generate a corresponding binarized image block through deep learning based on the three-channel color image block comprises the following operations performed asynchronously:

wherein M and N are natural numbers.

13. The apparatus of claim 11, wherein said road network generating module vectorizes said binarized image block, generating a partial trajectory road network comprising:

14. A trajectory network generating device comprising:

one or more processors; and

one or more memories having stored therein a computer-executable program that, when executed by the processor, performs the method of any of claims 1-8.

15. A computer program product stored in a computer readable storage medium and comprising computer instructions which, when executed by a processor, cause a computer device to perform the method of any one of claims 1-8.

16. A computer-readable storage medium having stored thereon computer-executable instructions for implementing the method of any one of claims 1-8 when executed by a processor.