Road recognition method, device, storage medium and computer program product

Abstract

The application discloses a road recognition method and device, a storage medium and a computer program product. Related embodiments can be applied in vehicle navigation, automatic driving, intelligent transportation, cloud technology and artificial intelligence scenarios. The method comprises: obtaining trajectory data of a travel object traveling in a target geographical space region, the trajectory data comprising traffic flow and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory; constructing a static flow graph based on the traffic flow and speed information of each trajectory point; calling a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph, the grayscale graph being used to represent the probability of each trajectory being recognized as a road; and calling a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, the road recognition result being used to indicate a target trajectory recognized as a road in the at least one trajectory. According to the embodiments, the road can be accurately recognized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a road recognition method and device, a storage medium and a computer program product. BACKGROUND

[0002] Road refers to infrastructure for various trackless vehicles and pedestrians to pass through, and is divided into urban roads, highways, factory and mine roads, forest area roads and rural roads according to use characteristics. Road recognition is an indispensable key step in road network generation. The existing road recognition method adopts a kernel density estimation (KDE) algorithm, that is, trajectory data is subjected to kernel density estimation processing to generate a trajectory histogram, and then morphological image processing is performed on the trajectory histogram to obtain a road recognition result. However, the KDE algorithm has a relatively high requirement for data quality of the trajectory data, and therefore, GPS data from multiple channels is usually uniformly sampled. However, the data uniformity of the GPS data from multiple channels is poor, which leads to an inability to accurately recognize roads. SUMMARY

[0003] The present application provides a road recognition method, device, storage medium and computer program product, which can accurately recognize roads.

[0004] In one aspect, the present application provides a road recognition method, which comprises:

[0005] Obtaining trajectory data of a travel object traveling in a target geographical space region; wherein the trajectory data comprises traffic flow and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory of the travel object;

[0006] Constructing a static flow graph based on the traffic flow and speed information of each trajectory point; wherein the static flow graph is used to represent flow distribution and speed distribution in the target geographical space region;

[0007] Calling a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a gray-scale graph; wherein the gray-scale graph is used to represent a probability that each trajectory is recognized as a road;

[0008] Calling a second convolutional neural network to perform road recognition processing on the gray-scale graph to obtain a road recognition result; wherein the road recognition result is used to indicate a target trajectory recognized as a road in the at least one trajectory.

[0009] In one aspect, the present application provides a road recognition device, which comprises:

[0010] The data acquisition unit is configured to acquire trajectory data of a travel object traveling in a target geographic space region, wherein the trajectory data comprises traffic flow and speed information of a plurality of trajectory points included in each trajectory of at least one trajectory.

[0011] The graph construction unit is configured to construct a static flow graph based on the traffic flow and speed information of each trajectory point, wherein the static flow graph is used to represent traffic distribution and speed distribution in the target geographic space region.

[0012] The processing unit is configured to invoke a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph, wherein the grayscale graph is used to represent a probability that each trajectory is recognized as a road.

[0013] The processing unit is further configured to invoke a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, wherein the road recognition result is used to indicate a target trajectory that is recognized as a road in the at least one trajectory.

[0014] In another aspect, an embodiment of the present application provides a computer device, which comprises a communication interface, and further comprises:

[0015] a processor adapted to implement one or more computer programs; and

[0016] a computer storage medium storing one or more computer programs, the one or more computer programs being adapted to be loaded and executed by the processor to perform the above road recognition method.

[0017] In another aspect, an embodiment of the present application provides a computer storage medium storing one or more computer programs, the one or more computer programs being adapted to be loaded and executed by a processor to perform the above road recognition method.

[0018] In another aspect, an embodiment of the present application provides a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device perform the above road recognition method.

[0019] In the embodiment of the present application, trajectory data of the driving object driving in the target geographic space region is acquired, the trajectory data includes traffic flow and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory, a static flow graph is constructed based on the traffic flow and speed information of each trajectory point, the static flow graph is used to represent the flow distribution and speed distribution in the target geographic space region, the massive trajectories of the target geographic space region in a historical time period can be described, and important feature information of the trajectories is retained. On this basis, a first convolutional neural network is called to perform trajectory recognition processing on the static flow graph to obtain a gray image, and a second convolutional neural network is called to perform road recognition processing on the gray image to obtain a road recognition result, the double-layer convolutional neural network can better complete the segmentation of the road and the background, and the problem that the road information is easily disturbed by non-road information such as vegetation shadow, high-rise buildings, rivers and the like is avoided, so that the road can be accurately recognized. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 is a flow diagram of a road recognition method provided by an embodiment of the present application;

[0022] Figure 2 is a construction diagram of a static flow graph provided by an embodiment of the present application;

[0023] Figure 3 is a construction diagram of a static flow unit graph provided by an embodiment of the present application;

[0024] Figure 4 is a structure diagram of a road recognition model provided by an embodiment of the present application;

[0025] Figure 5a is an acquisition diagram of a road recognition result provided by an embodiment of the present application;

[0026] Figure 5b is an acquisition diagram of a road recognition result provided by an embodiment of the present application;

[0027] Figure 5c is an acquisition diagram of a road recognition result provided by an embodiment of the present application;

[0028] Figure 6 is a flow diagram of another road recognition method provided by an embodiment of the present application;

[0029] Figure 7 is a schematic diagram of acquiring training track data provided by an embodiment of the present application;

[0030] Figure 8 is a structural schematic diagram of a road recognition device provided by an embodiment of the present application;

[0031] Figure 9 is a structural schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0033] Artificial intelligence (AI) is to use digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to design principles and implementation methods of various intelligent machines, so that the machine has the functions of perception, reasoning and decision-making.

[0034] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology (CV), speech processing technology, natural language processing technology, and machine learning (ML) / deep learning (DL) and other major directions. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its application covers all fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rule-based learning.

[0035] Based on the machine learning techniques mentioned above, the embodiment of the present application provides a road recognition scheme, which can first acquire trajectory data of a travel object traveling in a target geographic space region, the trajectory data including traffic flow and speed information of a plurality of trajectory points contained in each trajectory in at least one trajectory. Then, based on the traffic flow and speed information of each trajectory point, a static flow graph is constructed, the static flow graph being used to represent traffic distribution and speed distribution in the target geographic space region. Further, a first convolutional neural network is called to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph, the grayscale graph being used to represent a probability that each trajectory is recognized as a road, and a second convolutional neural network is called to perform road recognition processing on the grayscale graph to obtain a road recognition result, the road recognition result being used to indicate a target trajectory in the at least one trajectory that is recognized as a road.

[0036] The road recognition result in the embodiment of the present application can be used to generate a road network. The road network refers to a road system in a certain geographic space region, which is composed of various roads and interconnects and interweaves into a network-like distribution. A road network composed of all levels of highways is called a highway network. A road network composed of various roads in a city is called a city road network. Specifically, the road recognition result in the embodiment of the present application can be applied in application scenarios such as vehicle navigation and route optimization. For example, route optimization can specifically include road excavation or road shape correction.

[0037] In order to facilitate the understanding of the present application, the related terms are explained as follows.

[0038] The travel object includes but is not limited to a user riding on a vehicle or a walking user. The user riding on the vehicle includes but is not limited to a passenger or a driver, such as a car user, a public transportation user, a cycling user, a bus user, or a truck user, etc.

[0039] The trajectory data is a set of a series of position data, speed data, and time stamps. The trajectory data can include traffic flow and speed information of a plurality of trajectory points contained in each trajectory in at least one trajectory. The traffic flow of each trajectory point is determined based on position information of each trajectory point contained in each trajectory of the travel object. The position information of each trajectory point can be collected by a positioning device of the travel object. The speed information of each trajectory point can be collected by a speed sensor of the travel object. The speed sensor and the positioning device can be integrated in one electronic device or can be independent of different electronic devices. For example, if road recognition in a specified geographic space region is required, trajectory data of a plurality of travel objects traveling in the geographic space region in a preset time period can be acquired. The preset time period can be a preset time period, such as the last month or a specific time period, etc.

[0040] Trajectory refers to a figure formed by a moving point meeting certain conditions, and the trajectory can include a starting point and an ending point. The trajectory can include a plurality of trajectory points, and the plurality of trajectory points can at least include the starting point and the ending point. The positioning device can collect position information of the moving object at each trajectory point, and the speed sensor can collect speed information of the moving object at each trajectory point.

[0041] Static flow map, which can also be referred to as density map, refers to an RGB image including trajectory flow, speed and direction. The static flow map can be used to represent the flow distribution and speed distribution in a specified geographic space.

[0042] The road recognition result is used to indicate the recognized road. In one example, the road recognition result can be presented in the form of an image, for example, the road recognition result is a binary image, each white line in the binary image represents a road, and the black area in the binary image represents the background, that is, elements other than roads, such as mountains, rivers, buildings, etc. In another example, the road recognition result can be presented in the form of text, for example, the road recognition result can include the trajectory identifier of the target trajectory recognized as a road in at least one trajectory. Wherein, the trajectory identifier of any trajectory is used to identify the trajectory, for example, can be the number of the trajectory, etc.

[0043] In a specific implementation, the road recognition scheme proposed by the embodiments of the present application can be executed by a computer device, which can be a terminal device or a server; here, the terminal device can include but is not limited to: a computer, a smartphone, a tablet computer, a notebook computer, a smart home appliance, a vehicle-mounted terminal, a smart wearable device, etc.; here, the server can be a standalone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms, etc. Basic cloud computing services. Further optionally, the road recognition scheme proposed by the embodiments of the present application can also be executed by other computing electronic devices alone or in cooperation, and the embodiments of the present application do not make any limitation.

[0044] Referring to Figure 1 A flowchart of a road recognition method provided by the embodiments of the present application is shown. Figure 1 The road recognition method shown can be executed by a computer device, or can be executed by other computing electronic devices alone or in cooperation, and the embodiments of the present application take the computer device as an example. Figure 1 The road recognition method shown can include the following steps:

[0045] S101, obtain trajectory data of a travel object traveling in a target geographical space region, the trajectory data comprising traffic volume and speed information of a plurality of trajectory points contained in at least one trajectory.

[0046] The target geographical space region can refer to any geographical space region, such as a region requiring road network generation, a region requiring route optimization, etc., and is not limited in the embodiments of the present application.

[0047] In one example, the computer device can obtain trajectory data of a travel object traveling in a target geographical space region. For example, it is assumed that the travel objects include travel object A, travel object B, and travel object C. When traveling in the target geographical space region, travel object A, travel object B, and travel object C can collect position information and speed information of a plurality of trajectory points contained in at least one trajectory through respective positioning devices and speed sensors, and send the collected position information and speed information of each trajectory point to the computer device. The computer device can determine the traffic volume of each trajectory point based on the position information of the plurality of trajectory points contained in at least one trajectory of travel object A, travel object B, and travel object C. Optionally, the computer device can also obtain the position information and speed information of each trajectory point from the positioning devices and speed sensors of travel object A, travel object B, and travel object C in real time or at a preset interval.

[0048] In another example, the computer device can obtain trajectory data of at least one travel object traveling in a target geographical space region within a preset time period. For example, it is assumed that the preset time period is one month, and the computer device can obtain trajectory data of at least one travel object traveling in the target geographical space region within the last one month. For example, the travel objects traveling in the target geographical space region within the last one month include travel object A, travel object B, and travel object C. When traveling in the target geographical space region, travel object A, travel object B, and travel object C can collect position information and speed information of a plurality of trajectory points contained in at least one trajectory through respective positioning devices and speed sensors. Then, the computer device can obtain the position information and speed information of the plurality of trajectory points contained in at least one trajectory traveling in the target geographical space region within the last one month from the positioning devices and speed sensors of travel object A, travel object B, and travel object C, and determine the traffic volume of each trajectory point based on the position information of the plurality of trajectory points contained in at least one trajectory of travel object A, travel object B, and travel object C.

[0049] The travel modes of the travel objects A, B and C can be the same or different. For example, the travel objects A, B and C are all car users, or the travel object A is a car user, the travel object B is a pedestrian user, and the travel object C is a public transport user, and the like, which are not limited in the embodiments of the present application.

[0050] In an implementation manner, the computer device can acquire position information of each trajectory point included in each trajectory of the travel objects, and count the traffic volume of the trajectory points at the same position based on the position information of each trajectory point, to obtain the traffic volume of each trajectory point. For example, the computer device acquires position information of a plurality of trajectory points included in each trajectory of at least one travel object in at least one trajectory. Assuming that there are 100 trajectory points with the same position information, it indicates that the 100 trajectory points indicate the same geographical position, and thus it can be known that the traffic volume of the trajectory points at the geographical position is 100.

[0051] In S102, a static traffic graph is constructed based on the traffic volume and speed information of each trajectory point.

[0052] After the computer device acquires the trajectory data of the travel objects in the target geographical space region, the computer device can construct a static traffic graph based on the traffic volume and speed information of a plurality of trajectory points included in each trajectory of at least one travel object in at least one trajectory. The static traffic graph is used to represent the traffic distribution and speed distribution in the target geographical space region.

[0053] In an implementation manner, the computer device can generate a traffic graph layer based on the traffic volume of each trajectory point, the traffic graph layer being used to represent the traffic distribution in the target geographical space region. The computer device can also project the speed indicated by the speed information of each trajectory point in a first reference direction respectively, to obtain the speed component of each trajectory point in the first reference direction, and generate a first vector speed graph layer based on the speed component of each trajectory point in the first reference direction. The computer device can also project the speed indicated by the speed information of each trajectory point in a second reference direction respectively, to obtain the speed component of each trajectory point in the second reference direction, and generate a second vector speed graph layer based on the speed component of each trajectory point in the second reference direction. The first vector speed graph layer and the second vector speed graph layer are used to represent the speed distribution in the target geographical space region. Then, the computer device can construct a static traffic graph based on the traffic graph layer, the first vector speed graph layer and the second vector speed graph layer.

[0054] The static traffic flow map in this embodiment is a color rendering map generated by combining the average driving speed, traffic flow, and traffic direction of the target geographic spatial area. Since directional features cannot be directly modeled in the image, the computer device projects the speed into a first reference direction and a second reference direction, and calculates the speed components under the two reference directions respectively.

[0055] like Figure 2 Taking the illustrated diagram of static traffic flow map construction as an example, computer equipment can generate a traffic flow layer based on the traffic flow at each trajectory point. The traffic flow layer is also known as the R-channel layer, which physically represents the traffic flow value within the target geographic area. The R-channel layer can be constructed as follows: based on the traffic flow at each trajectory point, determine the R value of the corresponding pixel in the R-channel layer. The R value is positively correlated with the traffic flow at the corresponding trajectory point; that is, the larger the R value (0-255), the greater the traffic flow at the corresponding trajectory point, which visually appears as a larger proportion of red. Specifically, the computer equipment can iterate through the location information of each trajectory point on each trajectory and accumulate it on the digital map corresponding to the target geographic area to construct the R-channel layer.

[0056] Assuming the first reference direction is due north, the first vector velocity layer is the B-channel layer. The physical meaning of the B-channel layer represents the projected velocity of traffic within the target geographic area in the due north direction. The B-channel layer can be constructed by iterating through each trajectory point, calculating the projected velocity of that point in the due north direction, and then averaging and accumulating the projected velocities of all trajectory points in the due north direction. A larger B value (0-255) indicates a larger component of traffic velocity along the due north direction at the current location, which visually appears as a larger proportion of blue.

[0057] Assuming the second reference direction is due east, the second vector velocity layer, also known as the G-channel layer, physically represents the projected velocity of traffic within the target geographic area in the due east direction. The G-channel layer can be constructed by iterating through each trajectory point on each trajectory, calculating the projected velocity of that point in the due east direction, and then averaging and accumulating the projected velocities of all trajectory points in the due east direction. A larger G value (0-255) indicates a larger component of traffic velocity along the due east direction at the current location, which visually appears as a larger proportion of green.

[0058] This application preprocesses the input trajectory data, connects the points and models them in three-dimensional space, and constructs traffic flow layers, speed layers, and direction layers to depict massive trajectories of a certain historical geographic area, preserving important feature information of the trajectories, thereby ensuring accurate road identification.

[0059] In an embodiment, the static flow graph can include a plurality of static flow unit graphs. The computer device can perform spatiotemporal data segmentation on the digital map corresponding to the target geospatial region to obtain a plurality of grid graphs. For any grid graph, the computer device can construct a static flow unit graph of the grid graph according to the traffic flow and speed information of each trajectory point located in the grid graph.

[0060] For example, the computer device can perform three-layer grid segmentation on the digital map in the manner of a road network index tree. The first layer of grid segmentation is performed according to the granularity of a province, the second layer of grid segmentation is performed according to the granularity of a map parent library road network map sheet, and the third layer of grid segmentation is performed according to the granularity of a block. The size of the map sheet can be 8.3km*12.5km, and the size of the block can be 1024m*1024m.

[0061] As shown in the construction schematic diagram of the static flow unit graph, Figure 3 As shown in the construction schematic diagram of the static flow unit graph, after the computer device obtains a plurality of grid graphs corresponding to the target geospatial region, for any grid graph (i.e., a block), the computer device can obtain the traffic flow and speed information of each trajectory point located in the grid graph. Then, the computer device can generate a flow graph layer based on the traffic flow of each trajectory point located in the grid graph. The computer device can also project the speed indicated by the speed information of each trajectory point located in the grid graph in a first reference direction to obtain the speed component of each trajectory point in the first reference direction, and generate a first vector speed graph layer based on the speed component of each trajectory point in the first reference direction. In addition, the computer device can project the speed indicated by the speed information of each trajectory point located in the grid graph in a second reference direction to obtain the speed component of each trajectory point in the second reference direction, and generate a second vector speed graph layer based on the speed component of each trajectory point in the second reference direction. Then, the computer device can construct a static flow unit graph based on the flow graph layer, the first vector speed graph layer, and the second vector speed graph layer.

[0062] S103, calling the first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a gray-scale graph.

[0063] The gray-scale graph can be used to represent the probability that each trajectory is recognized as a road.

[0064] A convolutional neural network (CNN) is a feed-forward neural network whose artificial neurons can respond to only a small region of the coverage, and it has excellent performance for large image processing. The convolutional neural network is composed of one or more convolutional layers and a fully connected layer at the top (corresponding to the classic neural network), and also includes associated weights and pooling layers. CNN is known for its powerful remote sensing image classification ability, and roads have some natural attributes, such as connectivity, complexity, etc. Considering these attributes, the D-LinkNet model can better preserve detailed spatial and temporal information. Based on this, the first convolutional neural network can be a D-LinkNet model. Unlike remote sensing images, static traffic graphs constructed from trajectory data do not have the problem of being easily disturbed by non-road information such as vegetation shadows, high-rise buildings, rivers, etc. Therefore, the D-LinkNet model also has good effect in identifying roads in mountainous and vegetated areas. Therefore, the computer device can call the D-LinkNet model to perform trajectory recognition processing on the static traffic graph to obtain a grayscale image.

[0065] In S104, a second convolutional neural network is called to perform road recognition processing on the grayscale image to obtain a road recognition result.

[0066] The road recognition result can be used to indicate a target trajectory in at least one trajectory that is recognized as a road. Because the traffic and speed of different levels of roads are very different, and the roads in the real world have different width and shape characteristics on different road levels, such as national roads, provincial roads, village roads, and mountain roads. Therefore, if only the grayscale image is recognized, it cannot be accurately recognized as a road, especially in the case of complex parallel road interchange relationships. Therefore, after the grayscale image is obtained by processing through a D-LinkNet model, the grayscale image can be predicted by a second model, that is, a second convolutional neural network is called to perform road recognition processing on the grayscale image to obtain a road recognition result. After the second convolutional neural network, each pixel point of the road and the background obtains a binary segmentation result of approximately 0 and 1, which can better complete the segmentation of the road and the background.

[0067] In an implementation manner, the computer device can call the second convolutional neural network to compare the grayscale value of each pixel point in the grayscale image with a reference grayscale value to obtain a comparison result. Then, the computer device can generate a binary image according to the comparison result of each pixel point in the grayscale image, the binary image includes at least one road line, and the binary image is taken as the road recognition result, and one road line in the binary image is used to represent a road.

[0068] For example, a computer device can compare the grayscale values ​​of each pixel in a grayscale image with a reference grayscale value. Pixels with grayscale values ​​greater than the reference grayscale value will have their RGB values ​​updated to 255, 255, 255, while pixels with grayscale values ​​less than or equal to the reference grayscale value will have their RGB values ​​updated to 0, 0, 0, resulting in a binarized image. In this binarized image, the white areas represent road lines, and the black areas represent the background. In other words, areas with RGB values ​​of 255, 255, 255 represent road lines, and areas with RGB values ​​of 0, 0, 0 represent the background.

[0069] In one embodiment, the first convolutional neural network and the second convolutional neural network can be two independent neural network models, or they can be integrated into a single neural network model, such as a road recognition model. Figure 4 Taking the road recognition model shown as an example, this model can include a first convolutional neural network and a second convolutional neural network, with the output of the first convolutional neural network serving as the input to the second convolutional neural network. For instance, this road recognition model can be a double-deep CNN model, which includes a first deep convolutional neural network and a second deep convolutional neural network. After the computer device constructs a static traffic flow map, it can input the static traffic flow map into the first deep convolutional neural network to obtain a grayscale image. Then, the grayscale image is input into the second deep convolutional neural network to obtain the road recognition result, i.e., a binarized image.

[0070] like Figure 5a , Figure 5b as well as Figure 5c Taking the road recognition result shown as an example, the computer equipment constructs a static traffic flow map based on trajectory data of a certain geographic spatial area. This static traffic flow map is input into the first layer of a deep convolutional neural network to obtain a grayscale image. Then, the grayscale image is input into the second layer of the deep convolutional neural network to obtain the road recognition result, i.e., a binarized image. The aforementioned geographic spatial area includes not only urban areas with complex grade-separated intersections and high trajectory density, but also suburban areas and mountainous roads with sparse trajectory data, all of which achieved good recognition results. Figure 5a , Figure 5b as well as Figure 5cIt can be known from the shown road recognition result that the embodiment of the application performs road recognition through two-layer convolutional neural networks (i.e., the first convolutional neural network and the second convolutional neural network), has great anti-noise capability, has great advantages in recognizing long-line roads, and lays a solid foundation for realizing missing road excavation and road shape correction. In addition, the application does not need a large number of artificial experience parameters, and can better recognize roads through an adaptive transformation function even in a traffic-intensive area. In addition, the road recognition based on the static flow graph can better construct a road network graph for parallel roads, and can effectively establish a road network graph for online-offline roads and branch mouths.

[0071] In one embodiment, if the static flow graph includes a plurality of static flow unit graphs, the computer device can call the first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a gray-scale graph in the following manner: calling the first convolutional neural network to perform trajectory recognition processing on each static flow unit graph to obtain a gray-scale graph of each static flow unit graph. The computer device can call the second convolutional neural network to perform road recognition processing on the gray-scale graph to obtain a road recognition result in the following manner: calling the second convolutional neural network to perform road recognition processing on the gray-scale graph of each static flow unit graph to obtain a road recognition result of each static flow unit graph, and performing splicing processing on the road recognition results of each static flow unit graph to obtain a road recognition result of the target geographic space region.

[0072] In the embodiment of the application, since the D-LinkNet model aims to receive a 1024x1024 image as input and preserve detailed spatial information. Therefore, the computer device can perform spatiotemporal data segmentation on the digital map in advance to obtain a plurality of grid graphs, which can adapt to the data size required by the model input and middleware, and further rationalize the data scale.

[0073] In another embodiment, after the computer device obtains a plurality of grid graphs, the computer device can establish data labels of each grid graph for message identity recognition. Then, in the splicing processing, the road recognition results of each static flow unit graph can be spliced according to the data labels of each grid graph to obtain a road recognition result of the target geographic space region.

[0074] In the embodiment of the application, the trajectory data of the driving object driving in the target geographic space region is obtained, the trajectory data includes traffic flow and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory, a static flow graph is constructed based on the traffic flow and speed information of each trajectory point, a first convolutional neural network is called to perform trajectory recognition processing on the static flow graph to obtain a gray-scale graph, and a second convolutional neural network is called to perform road recognition processing on the gray-scale graph to obtain a road recognition result, which can improve the accuracy of road recognition.

[0075] Based on the related embodiments of the above road recognition method, the embodiments of the present application provide another road recognition method, which can recognize the road in a deep learning manner. The trained road recognition model can be obtained by supervised training of the road recognition model, and then the trained road recognition model is called to recognize the road. Referring to Figure 6 The flowchart of another road recognition method provided by the embodiments of the present application is shown. Figure 6 The road recognition method shown can be executed by a computer device, or by other electronic devices with computing power alone or in cooperation. The embodiments of the present application take the computer device as an example. Specifically, the model training process and the model calling process can be executed by the same device, or by different devices. The embodiments of the present application take the same device as an example. Figure 6 The road recognition method shown can include the following steps:

[0076] S601, obtaining training trajectory data and a labeled label of the training trajectory data, the training trajectory data including traffic flow and speed information of a plurality of trajectory points contained in each training trajectory in at least one training trajectory, and the labeled label being used to indicate a target training trajectory recognized as a road in the at least one training trajectory.

[0077] In one embodiment, since a large number of training samples are needed for model training, if each trajectory data is manually labeled, the labeling speed is slow; if labeling is performed only based on the road network, a large amount of noise will be encountered. Therefore, in order to quickly obtain a large amount of training trajectory data with labeled labels for model training, the embodiments of the present application propose a weakly supervised labeling scheme based on an algorithm to assist in quickly accumulating labeled labels. That is, the computer device can construct training samples, i.e., training trajectory data and labeled labels of the training trajectory data, in a weakly supervised labeling manner.

[0078] Specifically, the computer device can acquire trajectory feature maps, which include traffic flow and speed information of multiple trajectory points contained in at least one pre-trained trajectory. Then, based on the traffic flow and speed information of the multiple trajectory points contained in each pre-trained trajectory, the computer device can construct a pre-trained static traffic flow map. Next, a first convolutional neural network is invoked to perform trajectory recognition processing on the pre-trained static traffic flow map to obtain a pre-trained grayscale image, and an initial second convolutional neural network is invoked to perform road recognition processing on the pre-trained grayscale image to obtain a pre-trained road recognition result. Further, the computer device can calibrate the pre-trained road recognition result to obtain a calibrated pre-trained road recognition result, perform image enhancement processing on the trajectory feature map to obtain an enhanced trajectory feature map, and use the trajectory feature map and the enhanced trajectory feature map as the training trajectory data, and use the calibrated pre-trained road recognition result as the annotation label for the training trajectory data.

[0079] Image enhancement processing refers to the use of image enhancement or image rotation techniques in image processing. After image enhancement processing, multiple samples can be obtained, thus expanding the sample size.

[0080] by Figure 7 Taking the training trajectory data acquisition process as an example, the computer device can use a single pixel as a sample and a specified road network as initial label data to quickly train a weak model, i.e., an initial second convolutional neural network. The initial second convolutional neural network processes the data to obtain the labeled tags for the pre-trained trajectory data. Then, the labeled tags for the pre-trained trajectory data are manually calibrated to obtain manually labeled tags. Furthermore, the computer device can use image enhancement to automatically expand the accurate dataset after manual labeling. For example, after manually labeling 500 trajectory feature maps using the above method, image enhancement processing is performed on each trajectory feature map to obtain 11 enhanced trajectory feature maps. The labeled tags of each trajectory feature map and its 11 enhanced trajectory feature maps are consistent, thus obtaining 500*(1+11) = 6000 trajectory feature maps (i.e., training samples).

[0081] S602, based on the traffic flow and speed information of multiple trajectory points contained in each training trajectory, constructs a training static traffic map.

[0082] S603, the first convolutional neural network is invoked to perform trajectory recognition processing on the training static flow map to obtain the training grayscale map.

[0083] S604, the initial second convolutional neural network is invoked to perform road recognition processing on the training grayscale image to obtain the training road recognition result.

[0084] S605, obtain a difference between a trajectory indicated by the training road recognition result and a target training trajectory indicated by the labeled label.

[0085] S606, train the initial second convolutional neural network in a direction of reducing the difference to obtain the second convolutional neural network.

[0086] In one example, the first convolutional neural network can be pre-trained, and the computer device only trains the second convolutional neural network, and the specific training process is as described in steps S601 to S606.

[0087] In one embodiment, the computer device can adjust a reference gray value in the initial second convolutional neural network in a direction of reducing the difference to train the initial second convolutional neural network to obtain the second convolutional neural network, and the second convolutional neural network includes the adjusted reference gray value. Based on this, when the computer device calls the trained second convolutional neural network to perform road recognition on the gray image, the computer device can call the second convolutional neural network to compare the gray value of each pixel point in the gray image with the adjusted reference gray value to obtain a comparison result, and then generate a binary image according to the comparison result of each pixel point in the gray image, and take the binary image as the road recognition result.

[0088] In another example, the first convolutional neural network and the second convolutional neural network are integrated in the same model, i.e., a road recognition model, that is, the road recognition model includes the first convolutional neural network and the second convolutional neural network. Then the computer device can train the road recognition model. Specifically, the computer device can obtain training trajectory data and labeled labels of the training trajectory data, construct a training static flow graph based on the traffic flow and speed information of the plurality of trajectory points included in each training trajectory, call the first convolutional neural network in the initial road recognition model to perform trajectory recognition processing on the training static flow graph to obtain a training gray image, and then call the second convolutional neural network in the initial road recognition model to perform road recognition processing on the training gray image to obtain a training road recognition result. Then, the computer device can obtain a difference between a trajectory indicated by the training road recognition result and a target training trajectory indicated by the labeled label, and train the initial road recognition model in a direction of reducing the difference to obtain the road recognition model. Wherein, training the initial road recognition model can refer to adjusting parameters of the first convolutional neural network and the second convolutional neural network in the initial road recognition model, or adjusting parameters of the second convolutional neural network in the initial road recognition model. Exemplarily, the parameters of the second convolutional neural network can include a reference gray value.

[0089] S607, acquire trajectory data of the driving object in the target geographic space area, the trajectory data includes traffic flow and speed information of multiple trajectory points contained in at least one trajectory of the driving object.

[0090] S608 constructs a static traffic flow map based on the traffic flow and speed information of each trajectory point.

[0091] S609, the first convolutional neural network is invoked to perform trajectory recognition processing on the static flow graph to obtain a grayscale image.

[0092] S610, the second convolutional neural network is invoked to perform road recognition processing on the grayscale image to obtain the road recognition result.

[0093] Steps S607 to S610 can be found above. Figure 1 The specific description of the embodiments described herein will not be repeated in the embodiments of this application.

[0094] In this embodiment, training trajectory data and annotation labels for the training trajectory data are obtained. The training trajectory data includes traffic flow and speed information of multiple trajectory points contained in at least one training trajectory. The annotation labels are used to indicate the target training trajectory identified as a road in at least one training trajectory. Based on the traffic flow and speed information of multiple trajectory points contained in each training trajectory, a training static traffic flow map is constructed. A first convolutional neural network is called to perform trajectory recognition processing on the training static traffic flow map to obtain a training grayscale map. An initial second convolutional neural network is called to perform road recognition processing on the training grayscale map to obtain a training road recognition result. The difference between the trajectory indicated by the training road recognition result and the target training trajectory indicated by the annotation labels is obtained. The initial second convolutional neural network is trained in the direction of reducing the difference to obtain a second convolutional neural network. Road recognition can be performed through the trained second convolutional neural network, which can improve the accuracy of road recognition.

[0095] Based on the above description of the road recognition method, this application also discloses a road recognition device. This road recognition device can be a computer program (including program code) running on the aforementioned computer device. The road recognition device can perform actions such as... Figure 1 and Figure 6 For the road recognition method shown, please refer to [link / reference]. Figure 8 The road recognition device may include at least: a data acquisition unit 801, a map construction unit 802, and a processing unit 803.

[0096] The data acquisition unit 801 is configured to acquire trajectory data of a travel object traveling in a target geographic space region; wherein the trajectory data comprises traffic volume and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory of the travel object;

[0097] The graph construction unit 802 is configured to construct a static flow graph based on the traffic volume and speed information of each trajectory point; wherein the static flow graph is used to represent the flow distribution and speed distribution in the target geographic space region.

[0098] The processing unit 803 is configured to call a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph; wherein the grayscale graph is used to represent the probability that each trajectory is recognized as a road.

[0099] The processing unit 803 is further configured to call a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result; wherein the road recognition result is used to indicate a target trajectory in the at least one trajectory that is recognized as a road.

[0100] In an embodiment, when the graph construction unit 802 constructs a static flow graph based on the traffic volume and speed information of each trajectory point, it can be specifically configured to perform the following operations:

[0101] Generate a flow graph layer based on the traffic volume of each trajectory point; wherein the flow graph layer is used to represent the flow distribution in the target geographic space region.

[0102] Project the speed indicated by the speed information of each trajectory point in a first reference direction respectively to obtain the speed component of each trajectory point in the first reference direction, and generate a first vector speed graph layer based on the speed component of each trajectory point in the first reference direction.

[0103] Project the speed indicated by the speed information of each trajectory point in a second reference direction respectively to obtain the speed component of each trajectory point in the second reference direction, and generate a second vector speed graph layer based on the speed component of each trajectory point in the second reference direction; wherein the first vector speed graph layer and the second vector speed graph layer are used to represent the speed distribution in the target geographic space region.

[0104] Construct the static flow graph based on the flow graph layer, the first vector speed graph layer and the second vector speed graph layer.

[0105] In another embodiment, the data acquisition unit 801 acquires the traffic volume of each trajectory point in the following manner:

[0106] acquire position information of the travel object at each trajectory point included in each trajectory;

[0107] count the traffic volume of the trajectory points at the same position based on the position information of each trajectory point, to obtain the traffic volume of each trajectory point.

[0108] In yet another implementation, the static traffic map includes a plurality of static traffic unit maps; when invoking the first convolutional neural network to perform trajectory recognition processing on the static traffic map to obtain the grayscale map, the processing unit 803 can be specifically configured to perform the following operations:

[0109] invoke the first convolutional neural network to perform trajectory recognition processing on each static traffic unit map to obtain a grayscale map of each static traffic unit map;

[0110] When invoking the second convolutional neural network to perform road recognition processing on the grayscale map to obtain a road recognition result, the processing unit 803 can be specifically configured to perform the following operations:

[0111] invoke the second convolutional neural network to perform road recognition processing on the grayscale map of each static traffic unit map to obtain a road recognition result of each static traffic unit map;

[0112] perform splicing processing on the road recognition results of each static traffic unit map to obtain a road recognition result of the target geographic space region.

[0113] In yet another implementation, when constructing a static traffic map based on the traffic volume and speed information of each trajectory point, the graph construction unit 802 can be specifically configured to perform the following operations:

[0114] perform spatiotemporal data segmentation on a digital map corresponding to the target geographic space region to obtain a plurality of grid maps;

[0115] For any grid map, construct a static traffic unit map of the any grid map according to the traffic volume and speed information of each trajectory point located in the any grid map.

[0116] In yet another implementation, when invoking the second convolutional neural network to perform road recognition processing on the grayscale map to obtain a road recognition result, the processing unit 803 can be specifically configured to perform the following operations:

[0117] invoke the second convolutional neural network to compare the grayscale value of each pixel point in the grayscale map with a reference grayscale value to obtain a comparison result;

[0118] generate a binary image according to the comparison result of each pixel point in the grayscale map; wherein the binary image includes at least one road line;

[0119] the binary image as the road recognition result; wherein one road line in the binary image is used to represent one road.

[0120] In yet another implementation, the data obtaining unit 801 is further configured to obtain training trajectory data and a labeled label of the training trajectory data; wherein the training trajectory data comprises traffic flow and speed information of a plurality of trajectory points contained in each training trajectory of at least one training trajectory; and the labeled label is used to indicate a target training trajectory that is recognized as a road in the at least one training trajectory.

[0121] The graph constructing unit 802 is further configured to construct a training static flow graph based on the traffic flow and speed information of the plurality of trajectory points contained in each training trajectory.

[0122] The processing unit 803 is further configured to invoke the first convolutional neural network to perform trajectory recognition processing on the training static flow graph to obtain a training grayscale graph.

[0123] The processing unit 803 is further configured to invoke an initial second convolutional neural network to perform road recognition processing on the training grayscale graph to obtain a training road recognition result.

[0124] The processing unit 803 is further configured to obtain a difference between a trajectory indicated by the training road recognition result and a target training trajectory indicated by the labeled label.

[0125] The processing unit 803 is further configured to train the initial second convolutional neural network in a direction of reducing the difference to obtain the second convolutional neural network.

[0126] In yet another implementation, the processing unit 803 trains the initial second convolutional neural network in a direction of reducing the difference to obtain the second convolutional neural network, including:

[0127] adjusting a reference grayscale value in the initial second convolutional neural network in the direction of reducing the difference to train the initial second convolutional neural network to obtain the second convolutional neural network, the second convolutional neural network comprising the adjusted reference grayscale value.

[0128] The invoking the second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, including:

[0129] The invoking the second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, including:

[0130] A binary image is generated according to the comparison results of each pixel point in the gray image, and the binary image is taken as the road recognition result.

[0131] In yet another embodiment, the data acquisition unit 801, when acquiring the training trajectory data and the labeled label of the training trajectory data, can be specifically configured to perform the following operations:

[0132] Acquire a trajectory feature map, the trajectory feature map including traffic flow and speed information of a plurality of trajectory points included in each of a plurality of pre-training trajectories;

[0133] Construct a pre-training static flow graph based on the traffic flow and speed information of the plurality of trajectory points included in each of the plurality of pre-training trajectories;

[0134] Call the first convolutional neural network to perform trajectory recognition processing on the pre-training static flow graph to obtain a pre-training gray image;

[0135] Call the initial second convolutional neural network to perform road recognition processing on the pre-training gray image to obtain a pre-training road recognition result;

[0136] Calibrate the pre-training road recognition result to obtain a calibrated pre-training road recognition result;

[0137] Perform image enhancement processing on the trajectory feature map to obtain an enhanced trajectory feature map;

[0138] Take the trajectory feature map and the enhanced trajectory feature map as the training trajectory data;

[0139] Take the calibrated pre-training road recognition result as the labeled label of the training trajectory data.

[0140] According to one embodiment of the present application, Figure 1 and Figure 6 The steps involved in the method shown in Figure 8 The units in the road recognition device shown in can be executed by.

[0141] According to another embodiment of the present application, Figure 8The units in the road recognition device shown are divided based on logical functions. The units can be combined into one or several other units respectively or all, or some of the units can be further split into multiple units with smaller functions to form, which can achieve the same operation without affecting the implementation of the technical effects of the embodiments of the present application. In other embodiments of the present application, the road recognition device can also include other units. In actual applications, these functions can also be assisted by other units, and can be achieved by cooperation of multiple units.

[0142] According to another embodiment of the present application, the road recognition device as shown in Figure 1 or Figure 6 The computer program (including program code) related to each step of the method shown can be used to construct the road recognition device as shown in Figure 8 and to implement the road recognition method of the embodiments of the present application. The computer program can be recorded on, for example, a computer storage medium, loaded into the above-mentioned computer device through the computer storage medium, and run in the computer device.

[0143] In the embodiments of the present application, the data acquisition unit 801 acquires trajectory data of the travel object in the target geographic space region, and the trajectory data includes traffic flow and speed information of a plurality of trajectory points included in each trajectory in at least one trajectory. The graph construction unit 802 constructs a static flow graph based on the traffic flow and speed information of each trajectory point. The processing unit 803 calls a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph, and calls a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, which can improve the accuracy of road recognition.

[0144] Based on the above method embodiments and device embodiments, the present application further provides a computer device. Referring to Figure 9 , a structural schematic diagram of a computer device provided by the embodiments of the present application. Figure 9 The computer device shown can at least include a processor 901, a communication interface 902, and a computer storage medium 903. The processor 901, the communication interface 902, and the computer storage medium 903 can be connected through a bus or other means.

[0145] The computer storage medium 903 can be stored in the memory of the computer device, and is used to store a computer program including program instructions. The processor 901 is used to execute the program instructions stored in the computer storage medium 903. The processor 901 (or CPU (Central Processing Unit, central processor)) is the computing core and control core of the computer device, and is adapted to implement one or more instructions, and is specifically adapted to load and execute one or more instructions to implement the road recognition method process or the corresponding function.

[0146] The computer storage medium 903 can be stored in the memory of the computer device, and is used to store a computer program including program instructions. The processor 901 is used to execute the program instructions stored in the computer storage medium 903. The processor 901 (or CPU (Central Processing Unit, central processor)) is the computing core and control core of the computer device, and is adapted to implement one or more instructions, and is specifically adapted to load and execute one or more instructions to implement the road recognition method process or the corresponding function.

[0147] In one embodiment, the processor 901 can load and execute one or more instructions stored in the computer storage medium 903 to implement the corresponding steps of the method in the road recognition method embodiment described above. Figure 1 and Figure 6 In one embodiment, the processor 901 can load and execute one or more instructions stored in the computer storage medium 903 to implement the corresponding steps of the method in the road recognition method embodiment described above.

[0148] Obtain trajectory data of the traveling object traveling in the target geographic space region; wherein the trajectory data includes traffic flow and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory of the traveling object;

[0149] Based on the traffic flow and speed information of each trajectory point, a static flow graph is constructed; wherein the static flow graph is used to represent the traffic distribution and speed distribution in the target geographic space region;

[0150] The first convolutional neural network is called to perform trajectory recognition processing on the static flow graph, to obtain a grayscale graph; wherein the grayscale graph is used to represent the probability that each trajectory is recognized as a road;

[0151] The second convolutional neural network is called to perform road recognition processing on the grayscale graph, to obtain a road recognition result; wherein the road recognition result is used to indicate a target trajectory in the at least one trajectory that is recognized as a road.

[0152] In one embodiment, the processor 901, when constructing a static flow graph based on traffic flow and speed information of each trajectory point, can be specifically configured to perform the following operations:

[0153] Based on the traffic flow of each trajectory point, a flow graph layer is generated; wherein the flow graph layer is used to represent the flow distribution in the target geographic space region;

[0154] The speed of each trajectory point indicated by the speed information of each trajectory point is projected in a first reference direction respectively, to obtain the speed component of each trajectory point in the first reference direction, and based on the speed component of each trajectory point in the first reference direction, a first vector speed graph layer is generated;

[0155] The speed of each trajectory point indicated by the speed information of each trajectory point is projected in a second reference direction respectively, to obtain the speed component of each trajectory point in the second reference direction, and based on the speed component of each trajectory point in the second reference direction, a second vector speed graph layer is generated; wherein the first vector speed graph layer and the second vector speed graph layer are used to represent the speed distribution in the target geographic space region;

[0156] Based on the flow graph layer, the first vector speed graph layer and the second vector speed graph layer, the static flow graph is constructed.

[0157] In one embodiment, the processor 901, when obtaining the traffic flow of each trajectory point, can be specifically configured to perform the following operations:

[0158] The position information of each trajectory point contained in the trajectories of the moving object is obtained;

[0159] Based on the position information of each trajectory point, the traffic flow of trajectory points located at the same position is counted to obtain the traffic flow of each trajectory point.

[0160] In one embodiment, the static flow graph includes a plurality of static flow unit graphs; when the processor 901 calls the first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph, it can be specifically configured to perform the following operations:

[0161] The first convolutional neural network is called to perform trajectory recognition processing on each static traffic unit graph, to obtain a grayscale graph of each static traffic unit graph.

[0162] When the processor 901 calls the second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, the processor 901 can be specifically configured to perform the following operations:

[0163] The second convolutional neural network is called to perform road recognition processing on the grayscale graph of each static traffic unit graph, to obtain a road recognition result of each static traffic unit graph.

[0164] The road recognition results of the static traffic unit graphs are spliced to obtain a road recognition result of the target geographic space region.

[0165] In one embodiment, when the processor 901 constructs a static traffic graph based on the traffic flow and speed information of each trajectory point, the processor 901 can be specifically configured to perform the following operations:

[0166] The digital map corresponding to the target geographic space region is spatiotemporally segmented to obtain a plurality of grid graphs;

[0167] For any grid graph, a static traffic unit graph of the any grid graph is constructed according to the traffic flow and speed information of each trajectory point located in the any grid graph.

[0168] In one embodiment, when the processor 901 calls the second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, the processor 901 can be specifically configured to perform the following operations:

[0169] The second convolutional neural network is called to compare the grayscale value of each pixel point in the grayscale graph with a reference grayscale value, to obtain a comparison result.

[0170] A binary image is generated according to the comparison result of each pixel point in the grayscale graph; wherein the binary image includes at least one road line.

[0171] The binary image is taken as the road recognition result; wherein one road line in the binary image represents one road.

[0172] In one embodiment, the processor 901 is further configured to perform the following operations:

[0173] Training trajectory data and a labeled label of the training trajectory data are obtained; wherein the training trajectory data includes traffic flow and speed information of a plurality of trajectory points included in each training trajectory of at least one training trajectory; and the labeled label is used to indicate a target training trajectory recognized as a road in the at least one training trajectory.

[0174] construct a training static flow map based on the traffic flow and speed information of the plurality of trajectory points contained in each of the training trajectories;

[0175] invoke the first convolutional neural network to perform trajectory recognition processing on the training static flow map to obtain a training grayscale map;

[0176] invoke an initial second convolutional neural network to perform road recognition processing on the training grayscale map to obtain a training road recognition result;

[0177] obtain a difference between a trajectory indicated by the training road recognition result and a target training trajectory indicated by the labeled label;

[0178] train the initial second convolutional neural network in a direction of reducing the difference to obtain the second convolutional neural network.

[0179] In one embodiment, the processor 901, when training the initial second convolutional neural network in a direction of reducing the difference to obtain the second convolutional neural network, can be specifically configured to perform the following operation:

[0180] adjust a reference grayscale value in the initial second convolutional neural network in the direction of reducing the difference to train the initial second convolutional neural network to obtain the second convolutional neural network, the second convolutional neural network comprising the adjusted reference grayscale value;

[0181] The invoking the second convolutional neural network to perform road recognition processing on the grayscale map to obtain a road recognition result comprises:

[0182] invoke the second convolutional neural network to compare the grayscale value of each pixel point in the grayscale map with the adjusted reference grayscale value to obtain a comparison result;

[0183] generate a binary image according to the comparison result of each pixel point in the grayscale map, and take the binary image as the road recognition result.

[0184] In one embodiment, the processor 901, when obtaining training trajectory data and a labeled label of the training trajectory data, can be specifically configured to perform the following operation:

[0185] obtain a trajectory feature map, the trajectory feature map comprising traffic flow and speed information of a plurality of trajectory points contained in each of at least one pre-training trajectory;

[0186] construct a pre-training static flow map based on the traffic flow and speed information of the plurality of trajectory points contained in each of the pre-training trajectories;

[0187] invoke the first convolutional neural network to perform trajectory recognition processing on the pre-trained static flow graph to obtain a pre-trained grayscale graph;

[0188] invoke an initial second convolutional neural network to perform road recognition processing on the pre-trained grayscale graph to obtain a pre-trained road recognition result;

[0189] perform calibration processing on the pre-trained road recognition result to obtain a calibrated pre-trained road recognition result;

[0190] perform image enhancement processing on the trajectory feature graph to obtain an enhanced trajectory feature graph;

[0191] use the trajectory feature graph and the enhanced trajectory feature graph as the training trajectory data;

[0192] use the calibrated pre-trained road recognition result as a label of the training trajectory data.

[0193] In the embodiments of the application, the processor 901 obtains trajectory data of a driving object driving in a target geographical space region, the trajectory data including traffic flow and speed information of a plurality of trajectory points included in each trajectory of at least one trajectory, constructs a static flow graph based on the traffic flow and speed information of each trajectory point, invokes a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph, and invokes a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result, thereby improving the accuracy of road recognition.

[0194] The embodiments of the application provide a computer program product or a computer program, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to perform the method embodiments as shown in the above Figure 1 and Figure 6 The computer readable storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), or the like.

[0195] The above merely illustrates the specific implementation of the application, but the protection scope of the application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A road recognition method characterized by comprising: The method comprises: acquiring trajectory data of a travel object traveling in a target geographical space region; wherein the trajectory data comprises traffic flow and speed information of a plurality of trajectory points contained in each trajectory of at least one trajectory of the travel object; constructing a static flow graph based on the traffic flow and speed information of each trajectory point; wherein the static flow graph is used to represent traffic distribution and speed distribution in the target geographical space region; calling a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph; wherein the grayscale graph is used to represent the probability that each trajectory is recognized as a road; calling a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result; wherein the road recognition result is used to indicate a target trajectory in the at least one trajectory that is recognized as a road.

2. The method of claim 1, wherein, The method further comprises: generating a flow graph layer based on the traffic flow of each trajectory point; wherein the flow graph layer is used to represent traffic distribution in the target geographical space region; projecting the speed indicated by the speed information of each trajectory point in a first reference direction respectively to obtain a speed component of each trajectory point in the first reference direction, and generating a first vector speed graph layer based on the speed component of each trajectory point in the first reference direction; projecting the speed indicated by the speed information of each trajectory point in a second reference direction respectively to obtain a speed component of each trajectory point in the second reference direction, and generating a second vector speed graph layer based on the speed component of each trajectory point in the second reference direction; wherein the first vector speed graph layer and the second vector speed graph layer are used to represent speed distribution in the target geographical space region; constructing the static flow graph based on the flow graph layer, the first vector speed graph layer and the second vector speed graph layer.

3. The method of claim 1 or 2, wherein, The method further comprises: acquiring position information of each trajectory point contained in each trajectory of the travel object; counting the traffic flow of trajectory points located at the same position based on the position information of each trajectory point to obtain the traffic flow of each trajectory point.

4. The method of claim 1, wherein, The static flow graph comprises a plurality of static flow unit graphs. The method further comprises: calling the first convolutional neural network to perform trajectory recognition processing on each static flow unit graph to obtain a grayscale graph of each static flow unit graph. The method further comprises: calling the second convolutional neural network to perform road recognition processing on the grayscale graph of each static flow unit graph to obtain a road recognition result of each static flow unit graph; splicing the road recognition results of each static flow unit graph to obtain a road recognition result of the target geographical space region.

5. The method of claim 4, wherein, The method further comprises: spatiotemporal data segmentation is performed on a digital map corresponding to the target geospatial region to obtain a plurality of grid maps; For any grid map, a static flow unit graph of the any grid map is constructed according to traffic flow and speed information of each trajectory point located in the any grid map.

6. The method of claim 1, wherein, The second convolutional neural network is called to perform road recognition processing on the gray-scale graph to obtain a road recognition result, including: The second convolutional neural network is called to compare the gray-scale values of each pixel point in the gray-scale graph with a reference gray-scale value to obtain a comparison result; A binary image is generated according to the comparison result of each pixel point in the gray-scale graph; wherein the binary image includes at least one road line; The binary image is taken as the road recognition result; wherein one road line in the binary image is used to represent one road.

7. The method of claim 1, wherein, The method further includes: training trajectory data and a labeled label of the training trajectory data are obtained; wherein the training trajectory data includes traffic flow and speed information of a plurality of trajectory points contained in each training trajectory in at least one training trajectory; the labeled label is used to indicate a target training trajectory recognized as a road in the at least one training trajectory; a training static flow graph is constructed based on the traffic flow and speed information of the plurality of trajectory points contained in each training trajectory; the first convolutional neural network is called to perform trajectory recognition processing on the training static flow graph to obtain a training gray-scale graph; an initial second convolutional neural network is called to perform road recognition processing on the training gray-scale graph to obtain a training road recognition result; a difference between a trajectory indicated by the training road recognition result and a target training trajectory indicated by the labeled label is obtained; the initial second convolutional neural network is trained in a direction of reducing the difference to obtain the second convolutional neural network.

8. The method of claim 7, wherein, The initial second convolutional neural network is trained in a direction of reducing the difference to obtain the second convolutional neural network, including: a reference gray-scale value in the initial second convolutional neural network is adjusted in the direction of reducing the difference to train the initial second convolutional neural network to obtain the second convolutional neural network, the second convolutional neural network including the adjusted reference gray-scale value; The second convolutional neural network is called to perform road recognition processing on the gray-scale graph to obtain a road recognition result, including: The second convolutional neural network is called to compare the gray-scale values of each pixel point in the gray-scale graph with the adjusted reference gray-scale value to obtain a comparison result; A binary image is generated according to the comparison result of each pixel point in the gray-scale graph, and the binary image is taken as the road recognition result.

9. The method of claim 7, wherein, The training trajectory data and the labeled label of the training trajectory data are obtained, including: a trajectory feature graph is obtained, the trajectory feature graph including traffic flow and speed information of a plurality of trajectory points contained in each pre-training trajectory in at least one pre-training trajectory; a pre-training static flow graph is constructed based on the traffic flow and speed information of the plurality of trajectory points contained in each pre-training trajectory. calling the first convolutional neural network to perform trajectory recognition processing on the pre-trained static flow graph to obtain a pre-trained grayscale graph; calling an initial second convolutional neural network to perform road recognition processing on the pre-trained grayscale graph to obtain a pre-trained road recognition result; performing calibration processing on the pre-trained road recognition result to obtain a calibrated pre-trained road recognition result; performing image enhancement processing on the trajectory feature graph to obtain an enhanced trajectory feature graph; taking the trajectory feature graph and the enhanced trajectory feature graph as the training trajectory data; taking the calibrated pre-trained road recognition result as a label of the training trajectory data.

10. A road recognition apparatus characterized by comprising: comprising: a data acquisition unit configured to acquire trajectory data of a traveling object traveling in a target geographic space region; wherein the trajectory data comprises traffic flow and speed information of a plurality of trajectory points included in each trajectory of at least one trajectory of the traveling object; a graph construction unit configured to construct a static flow graph based on the traffic flow and speed information of each trajectory point; wherein the static flow graph is used to represent traffic distribution and speed distribution in the target geographic space region; a processing unit configured to call a first convolutional neural network to perform trajectory recognition processing on the static flow graph to obtain a grayscale graph; wherein the grayscale graph is used to represent the probability of each trajectory being recognized as a road; the processing unit is further configured to call a second convolutional neural network to perform road recognition processing on the grayscale graph to obtain a road recognition result; wherein the road recognition result is used to indicate a target trajectory in the at least one trajectory that is recognized as a road.

11. A computer device, comprising: comprising: a processor adapted to implement one or more computer programs; a computer storage medium storing one or more computer programs adapted to be loaded and executed by the processor to perform the road recognition method according to any one of claims 1-9.

12. A computer storage medium, characterized in that, The computer storage medium stores one or more computer programs adapted to be loaded and executed by the processor to perform the road recognition method according to any one of claims 1-9.

13. A computer program product, characterised in that, The computer program product comprises a computer program adapted to be loaded and executed by the processor to perform the road recognition method according to any one of claims 1-9. The computer program product comprises a computer program adapted to be loaded and executed by the processor to perform the road recognition method according to any one of claims 1-9.

Images