OSM-based overpass identification and navigable construction method and system

Abstract

The invention provides an overpass identification and navigability construction method and system based on an OSM. According to the method, urban area road node information and track information are input, multiple groups of road node sets are obtained through a DBSCAN clustering algorithm, and each overpass rectangular area is constructed in each group of road node sets; screening an area to which each track point belongs in the same time range, constructing each vehicle track in each area, and obtaining association between each vehicle track point and a road node through an improved Dijkstra matching algorithm; constructing a BiRNN network model, and obtaining an optimized BiRNN network model through a minimum negative logarithm likelihood loss model and a stochastic gradient descent method; and track information corresponding to the vehicles in the rectangular area of the flyover is obtained in real time, and the label type of the vehicle track matching path is predicted through the optimized BiRNN network model. Through the automatic road topology recognition and classification method, the navigable construction efficiency of the complex road is improved, manual intervention is reduced, and the method has important practical application value.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and road topology recognition, and in particular relates to an OSM-based overpass recognition and navigability construction method and system. Background Art

[0002] With the rapid development of China's economy, the number of motor vehicles has increased significantly, and driving has become an important way for the public to travel. In this process, road map navigation plays an indispensable role as an auxiliary tool for driving, especially in complex sections such as overpasses. At present, the construction of navigation road networks mainly relies on commercial companies to conduct field surveys using surveying vehicles. This method is not only costly but also has a long update cycle. In contrast, OpenStreetMap (OSM), as a spontaneous geographic information project, provides a wide range of global road network data, but its accuracy is low, and there are often problems such as wrong directions, missing roads, and connection errors, which limits its direct application in vehicle navigation.

[0003] In order to solve this problem, in recent years, research has gradually emerged on the use of low-cost, high-frequency updated spatiotemporal data to check and correct road network data. These data mainly come from vehicles equipped with GPS or Beidou positioning systems, which can record location information in real time and become an important resource for road information mining. However, as the most complex part of the road network, the overpass has multiple layers of intersections and fast-changing traffic rules, which makes the construction of navigation information more challenging. Therefore, there is an urgent need for a new method that can effectively identify and update the navigability of the overpass, reduce the economic and labor costs of building a navigable road network, and shorten the update cycle. Summary of the invention

[0004] In order to realize automatic identification and navigability construction of complex overpasses, the present invention proposes an overpass identification and navigability construction method and system based on OSM.

[0005] The technical solution of the method of the present invention is an OSM-based overpass identification and navigability construction method, comprising the following steps:

[0006] Step 1: Input the road node information of each urban area and the trajectory information of multiple vehicles in the same time range of the urban area, and cluster them by DBSCAN clustering algorithm based on the longitude and latitude of multiple urban area road nodes to obtain multiple sets of urban area road node sets, and construct each overpass rectangular area in each set of urban area road nodes;

[0007] Step 2: Filter the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and construct the trajectory of each vehicle in each rectangular area of ​​the overpass. Match them using the improved Dijkstra map matching algorithm to obtain the road nodes in each rectangular area of ​​the overpass corresponding to each vehicle trajectory point in the same time range in the urban area.

[0008] Step 3: Build a BiRNN network model, construct a long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample, and construct a label corresponding to each sample, build a model that minimizes the negative log-likelihood loss, and optimize the training through the stochastic gradient descent method to obtain the optimized BiRNN network model;

[0009] Step 4: Obtain the trajectory information corresponding to each vehicle in the rectangular area of ​​the overpass in real time, and predict the label type of the matching path of each vehicle trajectory in the rectangular area of ​​the overpass in real time through the optimized BiRNN network model;

[0010] Preferably, the information of each urban area road node in step 1 includes: the longitude of each urban area road node, the latitude of each urban area road node, the road type corresponding to each urban area road node, the road shape type corresponding to each urban area road node, the index corresponding to each urban area road node, the connectivity of any two urban area road nodes, the connection order of any two urban area road nodes, and the direction angle between any two urban area road nodes.

[0011] The trajectory information of each vehicle within the same time range in the urban area described in step 1 includes: multiple trajectory points of each vehicle within the same time range in the urban area, and the number and index of trajectories within the same time range in the urban area.

[0012] In step 1, each trajectory point of each vehicle within the same time range in the urban area includes: the longitude of each trajectory point of each vehicle within the same time range in the urban area, the latitude of each trajectory point of each vehicle within the same time range in the urban area, the direction angle between any two adjacent trajectory points of each vehicle within the same time range in the urban area, the collection time information of each vehicle trajectory point, the collection time sequence of each vehicle trajectory point, and the index corresponding to each vehicle trajectory point.

[0013] Step 1 constructs each overpass rectangular area as follows:

[0014] Step 1.1, based on the longitude and latitude of multiple urban area road nodes, clustering is performed using the DBCSAN clustering algorithm to obtain multiple groups of urban area road node sets.

[0015] Step 1.2: In each set of urban area road node sets, the road nodes whose road shape type is an overpass are screened out by a sliding window method as each set of urban area overpass road node sets.

[0016] Step 1.3, select the node with the smallest longitude and latitude in each set of urban area overpass road nodes as the upper left corner road node of each overpass rectangular area; select the node with the largest longitude and latitude in each set of urban area overpass road nodes as the lower right corner road node of each overpass rectangular area.

[0017] Step 1.4, construct each overpass rectangular area according to the road node at the upper left corner of each overpass rectangular area and the road node at the lower right corner of each overpass rectangular area.

[0018] Preferably, the step 2 selects the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range of the urban area belongs and constructs the trajectory of each vehicle in each rectangular area of ​​the overpass, as follows:

[0019] Combined with the longitude and latitude of each track point of each vehicle within the same time range in the urban area, multiple track points of each vehicle within the same time range in the urban area are divided into each rectangular area of ​​the overpass, and the rectangular area of ​​the overpass to which each track point of each vehicle within the same time range in the urban area belongs is obtained;

[0020] Further constructing the trajectory of each vehicle in each rectangular area of ​​each overpass according to the multiple trajectory points of each vehicle in the rectangular area of ​​each overpass;

[0021] In step 2, the improved Dijkstra map matching algorithm is used to match the road nodes in each rectangular area of ​​each overpass corresponding to each vehicle trajectory point within the same time range in the urban area, as follows:

[0022] Each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass is matched with the road node in the corresponding rectangular area of ​​each overpass by using the improved Dijkstra map matching algorithm, so as to obtain the road node in each rectangular area of ​​each overpass corresponding to each vehicle trajectory point in the same time range of the urban area;

[0023] The matching is performed by using an improved Dijkstra map matching algorithm, as follows:

[0024] The trajectory of each vehicle in each rectangular area of ​​the overpass is taken as input. When searching for matching nodes in the overpass area through the traditional Dijkstra algorithm in the order of collection time, the matching logic is changed to only perform a single highest confidence match, cancel the recursive matching mechanism, and obtain the matching node information results corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass.

[0025] According to the collection time sequence of each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass, the road nodes in each rectangular area of ​​each overpass corresponding to each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass are screened out in sequence, and the two connected road nodes are screened out in combination with the connectivity of any two road nodes corresponding to two adjacent collection times in each rectangular area of ​​each overpass, so as to construct a matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​each overpass;

[0026] The road node corresponding to the trajectory point with the smallest collection time in the trajectory of each vehicle in each rectangular area of ​​the overpass is used as the starting point of each vehicle in each rectangular area of ​​the overpass, and the road node corresponding to the trajectory point with the largest collection time in the trajectory of each vehicle in each rectangular area of ​​the overpass is used as the end point of each vehicle in each rectangular area of ​​the overpass;

[0027] The starting point and the end point of each vehicle in each rectangular area of ​​each overpass are taken as input, and the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​each overpass is obtained by optimizing the Viterbi algorithm in combination with multiple road nodes in each rectangular area of ​​each overpass;

[0028] If the result of comparing the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass with the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is inconsistent, the trajectory of the vehicle corresponding to each rectangular area of ​​the overpass is obtained, and the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is labeled and constructed into three types: road topology missing, road topology connectivity error, and road topology direction error;

[0029] If the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is consistent with the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass, the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is labeled, and four different labels are generated for the normal road topology type;

[0030] The direction angle between the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle and the direction angle between the road nodes corresponding to the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle are used as inputs. The CRF linear weighted feature model is used to calculate the direction consistency probability and emission probability P between the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle and the corresponding road nodes. re (c i ), transition probability Thus, each road node c in the same time range of the rectangular area of ​​the bridge is superimposed.j The joint probability P of the corresponding index sequence j As the model feature, the joint probability P j The operation method is as follows:

[0031]

[0032] Where T represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i Sequence of acquisition time sequence;

[0033] t represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i Collection time;

[0034] i represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i The corresponding index;

[0035] j represents each road node c within the same time range of the rectangular area of ​​the overpass j The corresponding index.

[0036] Preferably, the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass is constructed in step 3, specifically as follows:

[0037] The matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass described in step 2 is used as the initial sample;

[0038] The joint probability P described in step 2 j As the main feature, the connectivity, emission probability, and transition probability of any two adjacent nodes in the rectangular area of ​​the overpass are used as secondary features. After normalization, Z-score standardization is performed, and finally feature splicing is performed to generate a trajectory point c in the rectangular area of ​​the overpass. i The long vector x i , and at the same time integrate the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass to obtain the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample input to the BiRNN model;

[0039] Each sample is numerically represented by a long vector, and the label corresponding to each sample is the target that the model needs to predict, which is output by the BiRNN model;

[0040] The matching path type corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass is used as the label corresponding to the initial sample;

[0041] The model that minimizes the negative log-likelihood loss is defined as:

[0042]

[0043] Among them, M represents the number of trajectories within the same time range in the rectangular area of ​​the overpass;

[0044] m represents the index of the number of trajectories within the same time range in the rectangular area of ​​the overpass;

[0045] Represents the trajectory point c in the rectangular area of ​​the overpass i The true label under the m-th trajectory number index indicates the actual type of each sample;

[0046] Represents the road node c in the rectangular area of ​​the overpass j The predicted probability under the mth trajectory number index, representing each input vector x i Type prediction.

[0047] The present invention also proposes an OSM-based overpass identification and navigability construction system, comprising:

[0048] The module for constructing the rectangular area of ​​an overpass is used to input the road node information of each urban area and the trajectory information of multiple vehicles in the same time range of the urban area, and to cluster the longitude and latitude of multiple road nodes in the urban area through the DBSCAN clustering algorithm to obtain multiple sets of urban area road nodes, and to construct each rectangular area of ​​an overpass in each set of urban area road nodes;

[0049] The road node matching module is used to screen the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and construct the trajectory of each vehicle in each rectangular area of ​​the overpass, and to match the improved Dijkstra map matching algorithm to obtain the road nodes in each rectangular area of ​​the overpass corresponding to each vehicle trajectory point in the same time range in the urban area;

[0050] The BiRNN network model optimization module is used to construct the BiRNN network model, construct the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample, construct the label corresponding to each sample, and construct a model that minimizes the negative log-likelihood loss. The optimized BiRNN network model is obtained through the stochastic gradient descent method optimization training;

[0051] The vehicle trajectory matching path real-time prediction module is used to obtain the trajectory information corresponding to each vehicle in the rectangular area of ​​the overpass in real time, and predict the label type of each vehicle trajectory matching path in the rectangular area of ​​the overpass in real time through the optimized BiRNN network model.

[0052] The advantage of the present invention is that it provides a highly automated method for identifying road topology in an overpass area, which can effectively identify and correct topological errors in the overpass area. By combining node clustering and sliding window technology, the present invention significantly improves the accuracy and efficiency of road topology identification and reduces the need for manual intervention. In addition, the present invention adopts a deep learning model, which can process large-scale traffic data in real time, optimize traffic flow, and reduce the risk of traffic accidents. The system has self-learning capabilities, can quickly adapt to unknown situations, improve the robustness and accuracy of the model, and provide strong technical support for the optimization of urban traffic management and intelligent transportation systems. In summary, the present invention has significant practical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 : A flow chart of a method according to an embodiment of the present invention.

[0054] Figure 2 : Schematic diagram of overpass identification according to an embodiment of the present invention.

[0055] Figure 3 : Schematic diagram of the specific implementation of the map matching result of an embodiment of the present invention.

[0056] Figure 4 : Schematic diagram of a model of an embodiment of the present invention.

[0057] Figure 5 : Schematic diagram of the result structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] In specific implementation, the method proposed in the technical solution of the present invention can be implemented by technical personnel in this field using computer software technology to realize automatic operation process. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of the present invention and computer equipment running the corresponding computer program, should also be within the scope of protection of the present invention.

[0060] The present invention takes China's spaceborne aerosol and cloud high spectral resolution lidar ACDL as an example, which has fixed system hardware parameters.

[0061] Combine the following Figures 1 to 4The specific implementation of the present invention is an OSM-based overpass identification and navigability construction method and system.

[0062] like Figure 1 The figure is a method flow chart of an embodiment of the present invention, and the specific implementation method is a method for identifying and navigating overpasses based on OSM, which is as follows:

[0063] This experimental scenario covers data collection, feature extraction, model training and real-time prediction. It is based on NVIDIA RTX3060 graphics card, Intel i7 processor, 32GB memory, Windows 11 system, Python3.9 environment and a variety of data processing tools based on it.

[0064] Step 1: Input the road node information of each urban area and the trajectory information of multiple vehicles in the same time range of the urban area, and cluster them by DBSCAN clustering algorithm based on the longitude and latitude of multiple urban area road nodes to obtain multiple sets of urban area road node sets, and construct each overpass rectangular area in each set of urban area road nodes;

[0065] The information of each urban area road node described in step 1 includes: the longitude of each urban area road node, the latitude of each urban area road node, the road type corresponding to each urban area road node, the road shape type corresponding to each urban area road node, the index corresponding to each urban area road node, the connectivity of any two urban area road nodes, the connection order of any two urban area road nodes, and the direction angle between any two urban area road nodes.

[0066] The trajectory information of each vehicle within the same time range in the urban area described in step 1 includes: multiple trajectory points of each vehicle within the same time range in the urban area, and the number and index of trajectories within the same time range in the urban area.

[0067] In step 1, each trajectory point of each vehicle within the same time range in the urban area includes: the longitude of each trajectory point of each vehicle within the same time range in the urban area, the latitude of each trajectory point of each vehicle within the same time range in the urban area, the direction angle between any two adjacent trajectory points of each vehicle within the same time range in the urban area, the collection time information of each vehicle trajectory point, the collection time sequence of each vehicle trajectory point, and the index corresponding to each vehicle trajectory point.

[0068] Step 1 constructs each overpass rectangular area as follows:

[0069] Step 1.1, based on the longitude and latitude of multiple urban area road nodes, clustering is performed using the DBCSAN clustering algorithm to obtain multiple groups of urban area road node sets.

[0070] Step 1.2: In each set of urban area road node sets, the road nodes whose road shape type is an overpass are screened out by a sliding window method as each set of urban area overpass road node sets.

[0071] Step 1.3, select the node with the smallest longitude and latitude in each set of urban area overpass road nodes as the upper left corner road node of each overpass rectangular area; select the node with the largest longitude and latitude in each set of urban area overpass road nodes as the lower right corner road node of each overpass rectangular area.

[0072] Step 1.4, construct each overpass rectangular area according to the road node at the upper left corner of each overpass rectangular area and the road node at the lower right corner of each overpass rectangular area.

[0073] like Figure 2 An example of step 1 of the present invention is shown, including identifying each rectangular area of ​​an overpass.

[0074] Step 2: Filter the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and construct the trajectory of each vehicle in each rectangular area of ​​the overpass. Match them using the improved Dijkstra map matching algorithm to obtain the road nodes in each rectangular area of ​​the overpass corresponding to each vehicle trajectory point in the same time range in the urban area.

[0075] Step 2 screens the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and constructs the trajectory of each vehicle in each rectangular area of ​​the overpass, as follows:

[0076] Combined with the longitude and latitude of each track point of each vehicle within the same time range in the urban area, multiple track points of each vehicle within the same time range in the urban area are divided into each rectangular area of ​​the overpass, and the rectangular area of ​​the overpass to which each track point of each vehicle within the same time range in the urban area belongs is obtained;

[0077] Further constructing the trajectory of each vehicle in each rectangular area of ​​each overpass according to the multiple trajectory points of each vehicle in the rectangular area of ​​each overpass;

[0078] In step 2, the improved Dijkstra map matching algorithm is used to match the road nodes in each rectangular area of ​​each overpass corresponding to each vehicle trajectory point within the same time range in the urban area, as follows:

[0079] Each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass is matched with the road node in the corresponding rectangular area of ​​each overpass by using the improved Dijkstra map matching algorithm, so as to obtain the road node in each rectangular area of ​​each overpass corresponding to each vehicle trajectory point in the same time range of the urban area;

[0080] The matching is performed by using an improved Dijkstra map matching algorithm, as follows:

[0081] The trajectory of each vehicle in each rectangular area of ​​the overpass is taken as input. When searching for matching nodes in the overpass area through the traditional Dijkstra algorithm in the order of collection time, the matching logic is changed to only perform a single highest confidence match, cancel the recursive matching mechanism, and obtain the matching node information results corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass.

[0082] According to the collection time sequence of each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass, the road nodes in each rectangular area of ​​each overpass corresponding to each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass are screened out in sequence, and the two connected road nodes are screened out in combination with the connectivity of any two road nodes corresponding to two adjacent collection times in each rectangular area of ​​each overpass, so as to construct a matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​each overpass;

[0083] The road node corresponding to the trajectory point with the smallest collection time in the trajectory of each vehicle in each rectangular area of ​​the overpass is used as the starting point of each vehicle in each rectangular area of ​​the overpass, and the road node corresponding to the trajectory point with the largest collection time in the trajectory of each vehicle in each rectangular area of ​​the overpass is used as the end point of each vehicle in each rectangular area of ​​the overpass;

[0084] The starting point and the end point of each vehicle in each rectangular area of ​​each overpass are taken as input, and the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​each overpass is obtained by optimizing the Viterbi algorithm in combination with multiple road nodes in each rectangular area of ​​each overpass;

[0085] If the result of comparing the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass with the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is inconsistent, the trajectory of the vehicle corresponding to each rectangular area of ​​the overpass is obtained, and the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is labeled and constructed into three types: road topology missing, road topology connectivity error, and road topology direction error;

[0086] If the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is consistent with the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass, the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is labeled, and four different labels are generated for the normal road topology type;

[0087] The direction angle between the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle and the direction angle between the road nodes corresponding to the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle are used as inputs. The CRF linear weighted feature model is used to calculate the direction consistency probability and emission probability P between the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle and the corresponding road nodes. re (c i ), transition probability Thus, each road node c in the same time range of the rectangular area of ​​the bridge is superimposed. j The joint probability P of the corresponding index sequence j As the model feature, the joint probability P j The operation method is as follows:

[0088]

[0089] Where T represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i Sequence of acquisition time sequence;

[0090] t represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i Collection time;

[0091] i represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i The corresponding index;

[0092] j represents each road node c within the same time range of the rectangular area of ​​the overpass j The corresponding index.

[0093] like Figure 3 Shown are some examples of road topology error label types that may appear in step 2 of the present invention, including missing road topology, road topology connectivity errors, and road topology direction errors.

[0094] Step 3: Build a BiRNN network model, construct a long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample, and construct a label corresponding to each sample, build a model that minimizes the negative log-likelihood loss, and optimize the training through the stochastic gradient descent method to obtain the optimized BiRNN network model;

[0095] The long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass is constructed in step 3, as follows:

[0096] The matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass described in step 2 is used as the initial sample;

[0097] The joint probability P described in step 2 j As the main feature, the connectivity, emission probability, and transition probability of any two adjacent nodes in the rectangular area of ​​the overpass are used as secondary features. After normalization, Z-score standardization is performed, and finally feature splicing is performed to generate a trajectory point c in the rectangular area of ​​the overpass. i The long vector x i , and at the same time integrate the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass to obtain the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample input to the BiRNN model;

[0098] Each sample is numerically represented by a long vector, and the label corresponding to each sample is the target that the model needs to predict, which is output by the BiRNN model;

[0099] The matching path type corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass is used as the label corresponding to the initial sample;

[0100] The model that minimizes the negative log-likelihood loss is defined as:

[0101]

[0102] Among them, M represents the number of trajectories within the same time range in the rectangular area of ​​the overpass;

[0103] m represents the index of the number of trajectories within the same time range in the rectangular area of ​​the overpass;

[0104] Represents the trajectory point c in the rectangular area of ​​the overpass i The true label under the m-th trajectory number index indicates the actual type of each sample;

[0105] Represents the road node c in the rectangular area of ​​the overpass j The predicted probability under the mth trajectory number index, representing each input vector xi Type prediction.

[0106] like Figure 4 The figure shows a schematic diagram of the operation principle of the BiRNN model in step 3 of the implementation of the present invention.

[0107] Step 4: Obtain the trajectory information corresponding to each vehicle in the rectangular area of ​​the overpass in real time, and predict the label type of the real-time matching path of each vehicle trajectory in the rectangular area of ​​the overpass through the optimized BiRNN network model.

[0108] The technical solution of the system embodiment of the present invention is an OSM-based overpass identification and navigability construction system, including:

[0109] The module for constructing the rectangular area of ​​an overpass is used to input the road node information of each urban area and the trajectory information of multiple vehicles in the same time range of the urban area, and to cluster the longitude and latitude of multiple road nodes in the urban area through the DBSCAN clustering algorithm to obtain multiple sets of urban area road nodes, and to construct each rectangular area of ​​an overpass in each set of urban area road nodes;

[0110] The road node matching module is used to screen the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and construct the trajectory of each vehicle in each rectangular area of ​​the overpass, and to match the improved Dijkstra map matching algorithm to obtain the road nodes in each rectangular area of ​​the overpass corresponding to each vehicle trajectory point in the same time range in the urban area;

[0111] The BiRNN network model optimization module is used to construct the BiRNN network model, construct the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample, construct the label corresponding to each sample, and construct a model that minimizes the negative log-likelihood loss. The optimized BiRNN network model is obtained through the stochastic gradient descent method optimization training;

[0112] The vehicle trajectory matching path real-time prediction module is used to obtain the trajectory information corresponding to each vehicle in the rectangular area of ​​the overpass in real time, and predict the label type of each vehicle trajectory matching path in the rectangular area of ​​the overpass in real time through the optimized BiRNN network model.

[0113] like Figure 5The figure shows a schematic diagram of the output structure of the BiRNN model operation result in step 4 of an embodiment of the present invention. The result shows that the present invention can classify the matching paths corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass into road topology missing, road topology connectivity error, road topology direction error, and road topology normal, respectively, providing strong technical support for the optimization of urban traffic management and intelligent transportation systems.

[0114] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0115] It should be understood that the above description of the embodiments is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested by the present invention shall be based on the attached claims.

Claims

1. A method for identifying and navigating overpasses based on OSM, characterized in that: The following steps are involved: Step 1: Input the road node information of each urban area and the trajectory information of multiple vehicles in the same time range of the urban area, and cluster them by DBSCAN clustering algorithm based on the longitude and latitude of multiple urban area road nodes to obtain multiple sets of urban area road node sets, and construct each overpass rectangular area in each set of urban area road nodes; Step 2: Filter the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and construct the trajectory of each vehicle in each rectangular area of ​​the overpass. Match them using the improved Dijkstra map matching algorithm to obtain the road nodes in each rectangular area of ​​the overpass corresponding to each vehicle trajectory point in the same time range in the urban area. Step 3: Build a BiRNN network model, construct a long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample, and construct a label corresponding to each sample, build a model that minimizes the negative log-likelihood loss, and optimize the training through the stochastic gradient descent method to obtain the optimized BiRNN network model; Step 4: Obtain the trajectory information corresponding to each vehicle in the rectangular area of ​​the overpass in real time, and predict the label type of the matching path of each vehicle trajectory in the rectangular area of ​​the overpass in real time through the optimized BiRNN network model.

2. The OSM-based overpass identification and navigability construction method according to claim 1, characterized in that: The information of each urban area road node described in step 1 includes: the longitude of each urban area road node, the latitude of each urban area road node, the road type corresponding to each urban area road node, the road shape type corresponding to each urban area road node, the index corresponding to each urban area road node, the connectivity of any two urban area road nodes, the connection order of any two urban area road nodes, and the direction angle between any two urban area road nodes.

3. The OSM-based overpass identification and navigability construction method according to claim 2, characterized in that: The trajectory information of each vehicle within the same time range in the urban area in step 1 includes: multiple trajectory points of each vehicle within the same time range in the urban area, the number and index of trajectories within the same time range in the urban area; In step 1, each trajectory point of each vehicle within the same time range in the urban area includes: the longitude of each trajectory point of each vehicle within the same time range in the urban area, the latitude of each trajectory point of each vehicle within the same time range in the urban area, the direction angle between any two adjacent trajectory points of each vehicle within the same time range in the urban area, the collection time information of each vehicle trajectory point, the collection time sequence of each vehicle trajectory point, and the index corresponding to each vehicle trajectory point.

4. The OSM-based overpass identification and navigability construction method according to claim 3 is characterized by: Step 1 constructs each overpass rectangular area as follows: Step 1.1, based on the longitude and latitude of multiple urban area road nodes, clustering is performed using the DBCSAN clustering algorithm to obtain multiple sets of urban area road node sets; Step 1.2: In each set of urban area road node sets, the road nodes whose road shape type is an overpass are screened out by a sliding window method as each set of urban area overpass road node sets. Step 1.3, select the node with the smallest longitude and latitude in each set of urban area overpass road nodes as the upper left corner road node of each overpass rectangular area; select the node with the largest longitude and latitude in each set of urban area overpass road nodes as the lower right corner road node of each overpass rectangular area; Step 1.4, construct each overpass rectangular area according to the road node at the upper left corner of each overpass rectangular area and the road node at the lower right corner of each overpass rectangular area.

5. The OSM-based overpass identification and navigability construction method according to claim 4 is characterized by: Step 2 screens the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and constructs the trajectory of each vehicle in each rectangular area of ​​the overpass, as follows: Combined with the longitude and latitude of each track point of each vehicle within the same time range in the urban area, multiple track points of each vehicle within the same time range in the urban area are divided into each rectangular area of ​​the overpass, and the rectangular area of ​​the overpass to which each track point of each vehicle within the same time range in the urban area belongs is obtained; The trajectory of each vehicle in each rectangular area of ​​each overpass is further constructed according to the multiple trajectory points of each vehicle in the rectangular area of ​​each overpass.

6. The OSM-based overpass identification and navigability construction method according to claim 5, characterized in that: In step 2, the improved Dijkstra map matching algorithm is used to match the road nodes in each rectangular area of ​​each overpass corresponding to each vehicle trajectory point within the same time range in the urban area, as follows: Each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass is matched with the corresponding road node in each rectangular area of ​​each overpass through the improved Dijkstra map matching algorithm, so as to obtain the road node in each rectangular area of ​​each overpass corresponding to each vehicle trajectory point in the same time range of the urban area.

7. The OSM-based overpass identification and navigability construction method according to claim 6 is characterized by: The matching is performed by using an improved Dijkstra map matching algorithm, as follows: Taking the trajectory of each vehicle in each rectangular area of ​​the overpass as input, the matching logic is changed when searching for matching nodes in the overpass area through the traditional Dijkstra algorithm in the order of acquisition time, only a single highest confidence match is performed, the recursive matching mechanism is cancelled, and the matching node information results corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass are obtained; According to the collection time sequence of each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass, the road nodes in each rectangular area of ​​each overpass corresponding to each trajectory point of each vehicle's trajectory in each rectangular area of ​​each overpass are screened out in sequence, and the two connected road nodes are screened out in combination with the connectivity of any two road nodes corresponding to two adjacent collection times in each rectangular area of ​​each overpass, so as to construct a matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​each overpass; The road node corresponding to the trajectory point with the smallest collection time in the trajectory of each vehicle in each rectangular area of ​​the overpass is used as the starting point of each vehicle in each rectangular area of ​​the overpass, and the road node corresponding to the trajectory point with the largest collection time in the trajectory of each vehicle in each rectangular area of ​​the overpass is used as the end point of each vehicle in each rectangular area of ​​the overpass; The starting point and the end point of each vehicle in each rectangular area of ​​each overpass are taken as input, and the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​each overpass is obtained by combining multiple road nodes in the rectangular area of ​​each overpass through the Viterbi algorithm.

8. The OSM-based overpass identification and navigability construction method according to claim 7, characterized in that: If the result of comparing the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass with the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is inconsistent, the trajectory of the vehicle corresponding to each rectangular area of ​​the overpass is obtained, and the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is labeled and constructed into three types: road topology missing, road topology connectivity error, and road topology direction error; If the optimal path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is consistent with the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass, the matching path corresponding to the trajectory of each vehicle in each rectangular area of ​​the overpass is labeled, and four different labels are generated for the normal road topology type; The direction angle between the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle and the direction angle between the road nodes corresponding to the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle are used as inputs. The CRF linear weighted feature model is used to calculate the direction consistency probability and emission probability P between the trajectory points at two adjacent collection times in the trajectory of each overpass rectangular area corresponding to the vehicle and the corresponding road nodes. re (c i ), transition probability Thus, each road node c in the same time range of the rectangular area of ​​the bridge is superimposed. j The joint probability P of the corresponding index sequence j As the model feature, the joint probability P j The operation method is as follows: Where T represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i Sequence of acquisition time sequence; t represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i Collection time; i represents each vehicle trajectory point c within the same time range in the rectangular area of ​​the overpass i The corresponding index; j represents each road node c within the same time range of the rectangular area of ​​the overpass j The corresponding index.

9. The OSM-based overpass identification and navigability construction method according to claim 8, characterized in that: The long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass is constructed in step 3, as follows: The matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass described in step 2 is used as the initial sample; The joint probability P described in step 2 j As the main feature, the connectivity, emission probability, and transition probability of any two adjacent nodes in the rectangular area of ​​the overpass are used as secondary features. After normalization, Z-score standardization is performed, and finally feature splicing is performed to generate a trajectory point c in the rectangular area of ​​the overpass. i The long vector x i , and at the same time integrate the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass to obtain the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample input to the BiRNN model; Each sample is numerically represented by a long vector, and the label corresponding to each sample is the target that the model needs to predict, which is output by the BiRNN model; The matching path type corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass is used as the label corresponding to the initial sample; The model that minimizes the negative log-likelihood loss is defined as: Among them, M represents the number of trajectories within the same time range in the rectangular area of ​​the overpass; m represents the index of the number of trajectories within the same time range in the rectangular area of ​​the overpass; Y i (m) Represents the trajectory point c in the rectangular area of ​​the overpass i The true label under the m-th trajectory number index indicates the actual type of each sample; Represents the road node c in the rectangular area of ​​the overpass j The predicted probability under the mth trajectory number index, representing each input vector x i Type prediction.

10. An OSM-based overpass identification and navigability construction system, characterized in that: include: The module for constructing the rectangular area of ​​an overpass is used to input the road node information of each urban area and the trajectory information of multiple vehicles in the same time range of the urban area, and to cluster the longitude and latitude of multiple road nodes in the urban area through the DBSCAN clustering algorithm to obtain multiple sets of urban area road nodes, and to construct each rectangular area of ​​an overpass in each set of urban area road nodes; The road node matching module is used to screen the rectangular area of ​​the overpass to which each trajectory point of each vehicle in the same time range in the urban area belongs and construct the trajectory of each vehicle in each rectangular area of ​​the overpass, and to match the improved Dijkstra map matching algorithm to obtain the road nodes in each rectangular area of ​​the overpass corresponding to each vehicle trajectory point in the same time range in the urban area; The BiRNN network model optimization module is used to construct the BiRNN network model, construct the long vector of the matching path corresponding to the trajectory of each vehicle in the rectangular area of ​​the overpass as each sample, construct the label corresponding to each sample, and construct a model that minimizes the negative log-likelihood loss. The optimized BiRNN network model is obtained through the stochastic gradient descent method optimization training; The vehicle trajectory matching path real-time prediction module is used to obtain the trajectory information corresponding to each vehicle in the rectangular area of ​​the overpass in real time, and predict the label type of each vehicle trajectory matching path in the rectangular area of ​​the overpass in real time through the optimized BiRNN network model.

Images