Road navigation method and system based on artificial intelligence

Abstract

The invention provides a road navigation method and system based on artificial intelligence, and relates to the technical field of road navigation, the method comprises the following steps: determining information of each passenger and the longest travel time and the shortest travel time of the travel by using an internal video processing model based on a shot video in an automobile; matching a plurality of navigation routes based on the longest driving distance, the shortest driving distance, the longest travel time, the shortest travel time and the target scenic spot information of the travel; screening out a plurality of target midway scenic spots of each navigation route from the plurality of initial midway scenic spots in each navigation route by using a target midway scenic spot determination model; displaying a plurality of navigation routes and a plurality of target midway scenic spots of each navigation route in the plurality of navigation routes; and navigating based on the selected target navigation route. The method can recommend appropriate scenic spots and the optimal navigation route, and the travel experience of the user is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of road navigation, and in particular to an artificial intelligence-based road navigation method and system. Background Art

[0002] In modern society, travel has become one of the most important activities in people's lives. With the growth of tourism demand, people are paying more and more attention to travel experience and personalized needs. In modern society, people's demand for travel is no longer limited to simply reaching the destination, but pays more attention to travel experience and personalized needs. Traditional navigation systems often only recommend popular scenic spots, and then provide the shortest or fastest route options after recommending popular scenic spots. Traditional navigation systems lack consideration for the travel time limit of passengers. Everyone's itinerary may have time constraints. Sometimes users do not necessarily need the fastest route. Sometimes users need to visit as many attractions as possible within a limited time to improve the travel experience. Traditional navigation systems lack consideration for user interests and preferences and cannot meet personalized needs.

[0003] Therefore, how to recommend suitable attractions and optimal navigation routes to improve users' travel experience is an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by the present invention is how to recommend suitable attractions and optimal navigation routes to enhance the user's travel experience.

[0005] According to a first aspect, the present invention provides a road navigation method based on artificial intelligence, including: obtaining a video shot in a car and a traffic status map; determining each passenger's information and the longest and shortest play time of this trip using an internal video processing model based on the video shot in the car; obtaining each passenger's historical play record; determining target scenic spot information, each passenger's interest stay time period during the trip, and each passenger's preference for scenic spots based on each passenger's information, each passenger's historical play record, the longest and shortest play time of this trip; determining the longest driving distance and the shortest driving distance of this trip based on the traffic status map, the longest and shortest play time of this trip, and the target scenic spot information; determining the longest driving distance and the shortest driving distance of this trip based on the longest driving distance of this trip , the shortest driving distance, the longest play time of the trip, the shortest play time, and the target scenic spot information are matched to obtain multiple navigation routes, each of the multiple navigation routes includes multiple initial midway scenic spots; based on the multiple navigation route information, the time period of interest to stay during the trip of each passenger, and the degree of attraction preference of each passenger, a target midway scenic spot determination model is used to screen out multiple target midway scenic spots for each navigation route from the multiple initial midway scenic spots in each navigation route, and the multiple target midway scenic spots for each navigation route are marked in each navigation route; multiple navigation routes and multiple target midway scenic spots for each navigation route in the multiple navigation routes are displayed, and a target navigation route selected by a user from the multiple navigation routes is obtained; navigation is performed based on the selected target navigation route.

[0006] Furthermore, the internal video processing model is a Transformer model, the input of the internal video processing model is the video shot inside the car, and the output of the internal video processing model is the information of each passenger and the longest and shortest travel time of this trip.

[0007] Furthermore, the historical travel records of each passenger include the travel attractions, travel time, and stopover time during the travel route.

[0008] Furthermore, the displaying of the plurality of navigation routes and the target scenic spots along the way of each of the plurality of navigation routes comprises: displaying the plurality of navigation routes and the target scenic spots along the way of each of the plurality of navigation routes on a vehicle-mounted display screen.

[0009] Furthermore, the method also includes: constructing multiple graph structures based on the multiple navigation routes and the target midway attractions of each of the multiple navigation routes, wherein each graph structure represents a navigation route, each graph structure includes multiple nodes and multiple edges between the multiple nodes, each node represents a target midway attraction, the node characteristics of the target midway attraction node are the attraction type, average stay time at the attraction, location of the attraction, and current traffic flow of the attraction, the multiple edges between the multiple nodes represent the connection relationship between different target midway attractions, and the characteristics of each of the multiple edges include the distance between two nodes and the estimated travel time; processing the multiple graph structures based on a graph neural network model to determine a reference navigation route, the input of the graph neural network model is the multiple graph structures, and the output of the graph neural network model is the reference navigation route.

[0010] According to a second aspect, the present invention provides an artificial intelligence-based road navigation system, comprising: a first acquisition module, for acquiring a captured video and a traffic status map in a car; An internal video processing module, used to determine the information of each passenger and the longest and shortest play time of this trip based on the video shot in the car using an internal video processing model; The second acquisition module is used to obtain the historical travel records of each passenger; A target scenic spot determination module is used to determine the target scenic spot information based on the information of each passenger, the historical play records of each passenger, the longest play time of the current trip, the shortest play time, the time period of interest to stay during the trip of each passenger, and the degree of preference of each passenger for the scenic spot; A distance determination module, for determining the longest travel distance and the shortest travel distance of the current trip based on the traffic status map, the longest play time and the shortest play time of the current trip, and the target scenic spot information; A matching module, for matching a plurality of navigation routes based on the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time, and the target scenic spot information of the current trip, wherein each of the plurality of navigation routes includes a plurality of initial midway scenic spots; A target midway scenic spot determination module is used to screen out multiple target midway scenic spots of each navigation route from multiple initial midway scenic spots in each navigation route using a target midway scenic spot determination model based on multiple navigation route information, the interested stop time period of each passenger during the trip, and the scenic spot preference of each passenger, and mark the multiple target midway scenic spots of each navigation route in each navigation route; A display module, used to display multiple navigation routes and multiple target midway attractions of each of the multiple navigation routes, and obtain a target navigation route selected by a user from the multiple navigation routes; The navigation module is used for navigating based on the selected target navigation route.

[0011] Furthermore, the internal video processing model is a Transformer model, the input of the internal video processing model is the video shot inside the car, and the output of the internal video processing model is the information of each passenger and the longest and shortest travel time of this trip.

[0012] Furthermore, the historical travel records of each passenger include the travel attractions, travel time, and stopover time during the travel route.

[0013] Furthermore, the display module is also used to display the multiple navigation routes and the target intermediate scenic spots of each of the multiple navigation routes on the vehicle display screen.

[0014] Furthermore, the system is also used to: construct multiple graph structures based on the multiple navigation routes and the target midway attractions of each navigation route in the multiple navigation routes, wherein each graph structure represents a navigation route, each graph structure includes multiple nodes and multiple edges between the multiple nodes, each node represents a target midway attraction, the node features of the target midway attraction node are the attraction type of the target midway attraction, the average stay time at the attraction, the location of the attraction, and the current flow of people at the attraction, the multiple edges between the multiple nodes represent the connection relationship between different target midway attractions, and the features of each of the multiple edges include the distance between two nodes and the estimated travel time; The multiple graph structures are processed based on a graph neural network model to determine a reference navigation route, the input of the graph neural network model is the multiple graph structures, and the output of the graph neural network model is the reference navigation route.

[0015] The present invention provides an artificial intelligence-based road navigation method and system, which includes obtaining a video shot in a car and a traffic status map; determining each passenger's information and the longest and shortest play time of this trip using an internal video processing model based on the video shot in the car; obtaining each passenger's historical play record; determining target scenic spot information, each passenger's interest stay time period during the trip, and each passenger's preference for scenic spots based on each passenger's information, each passenger's historical play record, the longest and shortest play time of this trip; determining the longest driving distance and the shortest driving distance of this trip based on the traffic status map, the longest and shortest play time of this trip, and the target scenic spot information; determining the longest driving distance and the shortest driving distance of this trip based on the longest driving distance, The shortest driving distance, the longest play time of the trip, the shortest play time, and the target scenic spot information are matched to multiple navigation routes, each of which includes multiple initial midway scenic spots; based on multiple navigation route information, the time period of interest of each passenger during the trip, and the degree of preference of each passenger for scenic spots, a target midway scenic spot determination model is used to screen multiple target midway scenic spots of each navigation route from multiple initial midway scenic spots in each navigation route, and the multiple target midway scenic spots of each navigation route are marked in each navigation route; multiple navigation routes and multiple target midway scenic spots of each navigation route in the multiple navigation routes are displayed, and the target navigation route selected by the user from the multiple navigation routes is obtained; navigation is performed based on the selected target navigation route. The method can recommend suitable scenic spots and the best navigation route to enhance the user's travel experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a process flow of a road navigation method based on artificial intelligence provided by an embodiment of the present invention; Figure 2 A schematic diagram of a process for outputting a reference navigation route through a graph neural network model provided in an embodiment of the present invention; Figure 3 A schematic diagram of an artificial intelligence-based road navigation system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In an embodiment of the present invention, there is provided Figure 1 An artificial intelligence-based road navigation method is shown, and the artificial intelligence-based road navigation method includes steps S1 to S9: Step S1, obtaining the video shot inside the car and the traffic status map.

[0018] In some embodiments, the video shot inside the car can be obtained by acquiring the video shot inside the car through a vehicle camera or a passenger's mobile phone or other device.

[0019] A traffic map is a chart or map used to display real-time traffic conditions on roads and transportation networks. Traffic maps include road congestion, traffic volume, vehicle speeds, and other traffic-related information.

[0020] Step S2, based on the video shot in the car, use the internal video processing model to determine the information of each passenger and the longest and shortest play time of this trip.

[0021] The video captured in the car includes image sequences and audio sequences.

[0022] Passenger information includes passenger age, gender, name, health status during the trip, and activity level during the trip.

[0023] The longest tour time and the shortest tour time of this trip refer to the best longest tour time and the best shortest tour time determined by comprehensively analyzing the information of each passenger. The longest tour time of this trip includes the time spent on the way to the target attraction and the time spent at the target attraction. The shortest tour time of this trip includes the time spent on the way to the target attraction and the time spent at the target attraction. The time spent on the way to the target attraction includes the navigation time and the time spent at the waypoints. The waypoints can be the attractions along the way.

[0024] The internal video processing model is a Transformer model, the input of the internal video processing model is the video shot inside the car, and the output of the monitoring processing model is the information of each passenger and the longest and shortest play time of this trip.

[0025] The video shot in the car is a time series data that records the activities of passengers. By organizing the video frames into sequences and inputting them into the Transformer model, the model can learn the temporal information and correlation in the sequence, thereby understanding the behavior patterns of passengers.

[0026] In some embodiments, the internal video processing model includes an image and audio segmentation layer, an image processing layer, an audio processing layer, and a play time determination layer. The image and audio segmentation layer, the image processing layer, the audio processing layer, and the play time determination layer are all Transformer models. The input of the image and audio segmentation layer is the video shot in the car, and the output of the image and audio segmentation layer is the image sequence in the video shot in the car and the audio sequence in the video shot in the car. The input of the image processing layer is the image sequence in the video shot in the car. The output of the image processing layer is the age, gender, name, first health level of each passenger, and first activity level of each passenger. The input of the audio processing layer is the audio sequence in the video shot in the car. The output of the audio processing layer is the second health level of each passenger and the second activity level of each passenger. The input of the play time determination layer is the age, gender, name, first health level of each passenger, first activity level of each passenger, second health level of each passenger, and second activity level of each passenger. The output of the play time determination layer is the health level of each passenger during play, the activity level during play, the longest play time of this trip, and the shortest play time.

[0027] The first health level of each passenger and the second health level of each passenger can be used to indicate the size of the passenger's health level. The larger the first health level and the second health level, the healthier the passenger. The first activity level of each passenger and the second activity level of each passenger can be used to indicate the size of the passenger's activity level. The larger the first activity level and the second activity level, the more active the passenger is when riding, and the higher the passenger's interest in playing.

[0028] Dividing the video processing task into an image processing layer and an audio processing layer can make full use of two different data modalities, image and audio. By analyzing images and audio through the Transformer model, more comprehensive and multi-angle passenger information can be obtained. For example, the audio processing layer can provide sound-related clues to help further assess the health and activity level of passengers. The fusion of multimodal information helps to improve the expressiveness and accuracy of the model. The play time determination layer can combine the information of each passenger to determine the longest and shortest play time for this trip.

[0029] The image processing layer can infer a person's health and activity by analyzing the posture, facial expression, and movement of each person in the image sequence of the video shot in the car. For example, if a person's posture appears tired or their expression is depressed, the image processing layer may think that they are in a low health level. For another example, if a person shows active activities and movements in the video, the image processing layer may think that they are more active. The audio processing layer can also infer a person's health and activity by analyzing the speech data of each person in the video shot in the car. When the passenger speaks in the video, the audio processing layer can process his voice signal and extract some features from it. For example, the voice signal contains features such as pitch, frequency, and volume of the sound, which can be used to infer a person's health and activity. For example, if a person's voice is full of energy and motivation, the voice is loud, and the speed of speech is fast, then it can be inferred that he is a healthy and active person. On the contrary, if a person's voice is full of dullness and depression, the voice is small, and the speed of speech is slow, then it can be inferred that he is not in good health and has low energy.

[0030] A person's health and activity level can be used to determine the maximum and minimum play time for this trip because health and activity levels are directly related to a person's physical strength and energy when performing activities. If a person is in good health and has a high level of activity, he may have more physical strength and energy to participate in activities, so he can withstand a longer play time. On the contrary, if a person is in poor health or less active, he may need more rest time to recover his physical strength and energy, so the most suitable play time will be relatively short. By analyzing the health and activity level of passengers, we can make a preliminary assessment of their physical strength and energy status, and based on this information, infer the maximum and minimum play time suitable for them. This helps to provide passengers with a better play experience and avoid fatigue or discomfort due to lack of physical strength and energy.

[0031] Step S3, obtaining the historical travel records of each passenger.

[0032] In some embodiments, the historical travel records of each passenger can be obtained according to the name of each passenger. The historical travel records of each passenger include the travel attractions, travel time, and the stopover time in the travel route. The stopover time in the travel route refers to the length of time that the passenger stays at a certain scenic spot or place in the travel route.

[0033] For example, passenger Xiao Ming's travel history shows that he visited the Beijing Palace Museum in the summer of 2022 and spent 4 hours there, stopping at the Temple of Heaven for 1 hour. Then the travel attraction is the Beijing Palace Museum, the travel time is 5 hours, and the stopover time during the travel route is 1 hour.

[0034] Step S4, based on the information of each passenger, the historical travel records of each passenger, the longest travel time and the shortest travel time of this trip, determine the target scenic spot information, the time period of interest to stay during the trip of each passenger, and the degree of preference of each passenger for the scenic spot.

[0035] In some embodiments, the target scenic spot information, the time period of interest for each passenger during the trip, and the degree of preference for each passenger for the scenic spot can be determined based on the scenic spot determination model. The input of the scenic spot determination model is the information of each passenger, the historical play records of each passenger, the longest play time and the shortest play time of the current trip, and the output of the scenic spot determination model is the target scenic spot information, the time period of interest for each passenger during the trip, and the degree of preference for each passenger for the scenic spot. The scenic spot determination model can be an artificial neural network model (ANN). The artificial neural network model is a way to implement artificial intelligence. The basic structure of the artificial neural network model includes an input layer, a hidden layer, and an output layer. The input layer receives external input data, and the hidden layer is responsible for processing and processing the input data, and finally passes the result to the output layer, which generates the final prediction or classification result. The artificial neural network is trained by the back propagation algorithm, which enables the network to learn and adapt to specific tasks by continuously adjusting the connection weights. The artificial neural network contains multiple hidden layers, so that more complex features and relationships can be learned. Through training, artificial neural networks can learn complex nonlinear relationships from a large amount of input data and map these relationships to the desired output results. The neural network model can perform personalized modeling and prediction based on the unique characteristics and historical records of each passenger, and determine the time period of interest and the degree of attraction preference of each passenger. Based on the time period of interest and the degree of attraction preference of each passenger, the most suitable target attraction information with the longest and shortest playing time for this trip can be determined, thereby providing each passenger with a personalized travel experience.

[0036] The target scenic spot information refers to the information of the scenic spot recommended for visiting determined according to the scenic spot determination model, and the target scenic spot information includes the name, location, and type of the scenic spot.

[0037] The time period of interest for each passenger during the trip indicates the time period of play in the intermediate scenic spot that each passenger can accept when navigating to the target scenic spot. For example, if the passenger's target scenic spot is the Forbidden City and the time period of interest for the passenger during the trip is 1 to 2 hours, it means that the passenger is willing to stay at the intermediate scenic spot on the way to the Forbidden City for 1 to 2 hours.

[0038] The degree of attraction preference of each passenger refers to the degree of preference of each passenger for different types of attractions. For example, the degree of attraction preference of each passenger includes the degree of preference for cultural and historical attractions, the degree of preference for natural scenery attractions, the degree of preference for shopping and entertainment attractions, etc.

[0039] Step S5, determining the longest driving distance and the shortest driving distance of this trip based on the traffic status map, the longest playing time and the shortest playing time of this trip, and the target scenic spot information.

[0040] In some embodiments, the traffic status map, the longest and shortest playing time of the trip, and the target scenic spot information can be processed based on a convolutional neural network model to determine the longest driving distance and the shortest driving distance of the trip.

[0041] The convolutional neural network model is an implementation of artificial intelligence. The convolutional neural network (CNN) can be a multi-layer neural network (for example, including at least two layers). The at least two layers may include at least one of a convolutional layer (CONV), a rectified linear unit (ReLU) layer, a pooling layer (POOL), or a fully connected layer (FC). The convolutional neural network can extract useful features from the image and gradually understand and learn the contextual information of the image. The input of the convolutional neural network model is the traffic status map, the longest play time, the shortest play time of this trip, and the target scenic spot information, and the output of the convolutional neural network model is the longest driving distance and the shortest driving distance of this trip. The convolutional neural network model can comprehensively consider the traffic status map, the longest play time, the shortest play time of this trip, and the target scenic spot information and output the longest driving distance and the shortest driving distance of this trip.

[0042] The convolutional neural network model can combine the longest playing time, the shortest playing time, the target attraction information and the traffic condition map of the trip to infer the longest driving distance and the shortest driving distance required to reach the destination under different traffic conditions.

[0043] The maximum driving distance of this trip refers to the maximum driving distance and the minimum driving distance allowed in the entire trip. For example, the maximum driving distance of this trip shall not exceed 200 kilometers, and the shortest driving distance shall not be less than 50 kilometers. In some embodiments, the shortest driving distance may be the driving distance corresponding to the shortest path.

[0044] Step S6, matching multiple navigation routes based on the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time, and the target scenic spot information of the trip, each of the multiple navigation routes includes multiple initial intermediate scenic spots.

[0045] Initial midway attractions are other attractions that the navigation route passes through during the entire trip, except for the starting point and the end point. For example, if the target attraction is a seaside resort, and the route to the seaside resort includes a commercial street attraction, the commercial street is an initial midway attraction.

[0046] In some embodiments, a plurality of navigation routes can be matched based on the navigation route generation model based on the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time of the trip, and the target scenic spot information, and each of the plurality of navigation routes includes a plurality of initial midway scenic spots. The navigation route generation model is an artificial neural network model. The input of the navigation route generation model is the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time of the trip, and the target scenic spot information of the trip, and the output of the navigation route generation model is a plurality of navigation routes, and each of the plurality of navigation routes includes a plurality of initial midway scenic spots.

[0047] Step S7, based on the information of multiple navigation routes, the time period of interest of each passenger during the trip, and the degree of preference of each passenger for attractions, a target midway attraction determination model is used to screen out multiple target midway attractions for each navigation route from the multiple initial midway attractions in each navigation route, and the multiple target midway attractions for each navigation route are marked in each navigation route.

[0048] Navigation route information: length of the navigation route, path planning, driving time, etc.

[0049] The target midway scenic spot is selected from among multiple navigation routes after screening as a midway scenic spot that the passengers on the route should pay attention to and meets the needs of the passengers.

[0050] As an example, there is a travel team of two people. Their travel destination is the main scenic spot area of ​​a city, and multiple navigation routes have been generated. Then, the target midway scenic spot determination model selects the target midway scenic spot that may best meet the needs of the passengers based on the scenic spot preference provided by the passengers and the time period of interest of each passenger during the trip. The target midway scenic spot that best meets the needs of the passengers is marked on each navigation route for the reference of the passengers.

[0051] The target midway scenic spot determination model can be an artificial neural network model. The input of the target midway scenic spot determination model is multiple navigation route information, the interest stay time period of each passenger during the journey, and the attraction preference degree of each passenger. The output of the target midway scenic spot determination model is multiple target midway scenic spots for each navigation route. As an example, the input multiple navigation route information, the interest stay time period of the passengers during the journey, and the attraction preference degree can be extracted by a multi-layer neural network structure. Each feature can represent a certain aspect of information, such as the distance of the navigation route, the preference degree of the passengers for different attractions, etc. The extracted features will be fused together to comprehensively consider the relationship between multiple input information. For example, feature fusion can be achieved by connecting layers, weighted summation or other fusion methods. In this way, information such as navigation route information, interest stay time period during the journey, and attraction preference degree can be integrated into a comprehensive representation. The fused features will be input to the output layer of the neural network model to calculate the score of each midway scenic spot. The neural network model can generate an evaluation result for each midway scenic spot by learning the patterns and rules of a large number of data samples. These evaluation results can determine whether each midway attraction is suitable as a target midway attraction based on the needs, preferences and itinerary information of the passengers. Based on the output of the neural network model, multiple initial midway attractions for each navigation route can be scored and ranked. The midway attractions with the highest scores are selected as target midway attractions, indicating that they best meet the needs and preferences of the passengers.

[0052] Mark the target midway attractions on each navigation route on the route map or navigation software so that passengers can clearly see the location and order of each midway attraction. In this way, passengers can choose which target midway attractions to visit during the trip according to their own interests and time arrangements.

[0053] Step S8, displaying a plurality of navigation routes and a plurality of target scenic spots along the way of each of the plurality of navigation routes, and obtaining a target navigation route selected by the user from the plurality of navigation routes.

[0054] The target navigation route is a navigation route selected by the user from multiple navigation routes. The purpose of this step is to provide users with multiple navigation routes and corresponding target midway attractions options so that users can make decisions based on their preferences, time and interests. By displaying multiple navigation routes and multiple target midway attractions for each of the multiple navigation routes, users can better compare and select a travel route that suits them, thereby enhancing travel satisfaction.

[0055] In some embodiments, the plurality of navigation routes and the target intermediate scenic spots of each of the plurality of navigation routes may be displayed on an in-vehicle display screen.

[0056] In some embodiments, a reference navigation route can also be output through a graph neural network model for customer reference, and the reference navigation route can be displayed on a user terminal.

[0057] Figure 2 A schematic diagram of a process for outputting a reference navigation route through a graph neural network model provided in an embodiment of the present invention, wherein outputting a reference navigation route through a graph neural network model includes steps S21-S22: Step S21, constructing multiple graph structures based on the multiple navigation routes and the target midway attractions of each navigation route in the multiple navigation routes, wherein each graph structure represents a navigation route, each graph structure includes multiple nodes and multiple edges between the multiple nodes, each node represents a target midway attraction, the node characteristics of the target midway attraction node are the attraction type of the target midway attraction, the average stay time at the attraction, the location of the attraction, and the current flow of people at the attraction, the multiple edges between the multiple nodes represent the connection relationship between different target midway attractions, and the characteristics of each of the multiple edges include the distance between two nodes and the estimated travel time.

[0058] In graph structure data, nodes represent entities or concepts, and each node contains a set of features that describe the attributes or states of the entity. In the present application, each graph structure represents a navigation route, and each node represents a target midway attraction. On the navigation route from city A to city B, the nodes in the graph structure can represent the various target midway attractions, and the edges represent the distances and estimated travel times between these target midway attractions. The estimated travel time can be determined by existing navigation system software.

[0059] In graph structure data, edges represent the relationship between nodes. In this application, edges represent the connection relationship between different target midway attractions, and the characteristics of each edge in multiple edges include the distance between two nodes and the estimated travel time.

[0060] Executing step S21 may construct multiple graph structures based on multiple navigation routes and target scenic spots along the way of each navigation route, so as to better understand the relationship between different roads and target scenic spots along the way and provide more accurate navigation services.

[0061] In some embodiments, as an example, executing step S21 requires performing the following steps: Collect data: In order to build multiple graph structures, it is necessary to collect information such as the attraction type, average stay time at the attraction, location of the attraction, current traffic flow of the attraction, etc. of the roads and target attractions from maps and other available data sources.

[0062] Build graph structures: Based on the collected data and algorithms, build multiple graph structures. Each graph structure represents a navigation route, each node represents a target intermediate attraction, and the edge represents the distance and estimated travel time between two nodes.

[0063] Add node features: Add feature information to each node, such as the name, type, historical and cultural background of the target scenic spot.

[0064] Add edge features: Add feature information to each edge, such as the distance and estimated travel time between two target intermediate attractions.

[0065] Step S22: Process the multiple graph structures based on a graph neural network model to determine a reference navigation route, wherein the input of the graph neural network model is the multiple graph structures, and the output of the graph neural network model is the reference navigation route.

[0066] The graph neural network model includes a graph neural network (GNN) and a fully connected layer. A graph neural network is a neural network that directly acts on graph structure data. Graph structure data is a data structure consisting of nodes and edges. Graph neural networks are a way to implement artificial intelligence.

[0067] In some embodiments, multiple graph structures can be merged into a large graph structure, and then trained and predicted using a graph neural network model. As an example, each navigation route can be regarded as a subgraph, each subgraph containing multiple nodes and edges. Then, all subgraphs can be merged into a large graph structure, where each subgraph corresponds to a feature of a subgraph type. In this way, multiple map data can be considered simultaneously in the graph neural network model, thereby more accurately predicting the navigation route.

[0068] The graph neural network model can receive multiple graph structures as input and output a reference navigation route. The graph neural network model includes an input layer, an intermediate layer, and an output layer. The input layer of the graph neural network model includes multiple node and edge features, each node represents a target midway attraction, and the edge represents the distance and estimated travel time between two nodes. The intermediate layer includes multiple neurons for learning the complex relationship between various factors. The output layer outputs the reference navigation route.

[0069] During the navigation process, multiple factors need to be considered comprehensively, such as road conditions, traffic congestion, user preferences, target midway attractions, etc. There are complex nonlinear relationships between these factors, and traditional rules and algorithms are often difficult to accurately model and process. The graph neural network model has a strong nonlinear modeling capability and can learn and represent the complex relationships between various factors. Specifically, by treating each target midway attraction as a node and the distance and estimated travel time between two nodes as an edge, a graph structure containing the target midway attractions can be constructed. Using the graph neural network model, the complex relationships between various factors can be learned from multiple input graph structures, such as the correlation between attractions, the impact of road conditions on the navigation route, etc. The model can automatically learn and extract features to more accurately predict the best reference navigation route for user reference.

[0070] Step S9: Navigate based on the selected target navigation route.

[0071] In some embodiments, the selected target navigation route may be navigated by a GPS-based navigation device, a smartphone application, or a vehicle navigation system.

[0072] Based on the same inventive concept, Figure 3 A schematic diagram of an artificial intelligence-based road navigation system provided by an embodiment of the present invention, the artificial intelligence-based road navigation system comprising: A first acquisition module 31 is used to acquire the video shot in the car and the traffic status map; An internal video processing module 32 is used to determine the information of each passenger and the longest and shortest play time of this trip based on the video shot in the car using an internal video processing model; The second acquisition module 33 is used to obtain the historical travel records of each passenger; The target scenic spot determination module 34 is used to determine the target scenic spot information based on the information of each passenger, the historical play records of each passenger, the longest play time and the shortest play time of the current trip, the time period of interest to stay during the trip of each passenger, and the degree of preference of each passenger for the scenic spot; The distance determination module 35 is used to determine the longest driving distance and the shortest driving distance of the current trip based on the traffic status map, the longest playing time and the shortest playing time of the current trip, and the target scenic spot information; A matching module 36 is used to match multiple navigation routes based on the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time, and the target scenic spot information of the current trip, each of the multiple navigation routes includes multiple initial midway scenic spots; The target midway scenic spot determination module 37 is used to use the target midway scenic spot determination model to screen out multiple target midway scenic spots of each navigation route from multiple initial midway scenic spots in each navigation route based on multiple navigation route information, the interested stop time period of each passenger during the trip, and the scenic spot preference of each passenger, and mark the multiple target midway scenic spots of each navigation route in each navigation route; A display module 38, used to display a plurality of navigation routes and a plurality of target midway attractions of each of the plurality of navigation routes, and to obtain a target navigation route selected by a user from the plurality of navigation routes; The navigation module 39 is used for navigating based on the selected target navigation route.

Claims

1. A road navigation method based on artificial intelligence, characterized in that: include: Get the video footage and traffic status map shot in the car; Based on the video shot in the car, use an internal video processing model to determine the information of each passenger and the longest and shortest play time of this trip; Get the historical travel records of each passenger; Determine target scenic spot information, each passenger's interest stay time during the trip, and each passenger's preference for scenic spots based on each passenger's information, each passenger's historical travel records, the longest travel time and the shortest travel time of the current trip; Determine the longest driving distance and the shortest driving distance of this trip based on the traffic status map, the longest playing time and the shortest playing time of this trip, and the target scenic spot information; Matching a plurality of navigation routes based on the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time, and the target scenic spot information of the current trip, wherein each of the plurality of navigation routes includes a plurality of initial midway scenic spots; Based on the information of multiple navigation routes, the time period of interest of each passenger during the trip, and the preference of each passenger for scenic spots, a target scenic spot determination model is used to select multiple target scenic spots for each navigation route from multiple initial scenic spots in each navigation route, and the multiple target scenic spots for each navigation route are marked in each navigation route; Displaying multiple navigation routes and multiple target midway attractions of each of the multiple navigation routes, and obtaining a target navigation route selected by a user from the multiple navigation routes; Navigation is performed based on the selected target navigation route.

2. The artificial intelligence-based road navigation method according to claim 1, characterized in that: The internal video processing model is a Transformer model, the input of the internal video processing model is the video shot inside the car, and the output of the internal video processing model is the information of each passenger and the longest and shortest travel time of this trip.

3. The road navigation method based on artificial intelligence as claimed in claim 1, characterized in that: The historical travel records of each passenger include the travel attractions, travel time, and stopover time during the travel route.

4. The road navigation method based on artificial intelligence as claimed in claim 1, characterized in that: The displaying of the plurality of navigation routes and the target scenic spots along the way of each of the plurality of navigation routes comprises: displaying the plurality of navigation routes and the target scenic spots along the way of each of the plurality of navigation routes on a vehicle-mounted display screen.

5. The road navigation method based on artificial intelligence as claimed in claim 1, characterized in that: The method further includes: constructing multiple graph structures based on the multiple navigation routes and the target midway attractions of each of the multiple navigation routes, wherein each graph structure represents a navigation route, each graph structure includes multiple nodes and multiple edges between the multiple nodes, each node represents a target midway attraction, the node features of the target midway attraction node are the attraction type, average stay time, location, and current flow of people at the attraction, the multiple edges between the multiple nodes represent the connection relationship between different target midway attractions, and the features of each of the multiple edges include the distance between two nodes and the estimated travel time; The multiple graph structures are processed based on a graph neural network model to determine a reference navigation route, the input of the graph neural network model is the multiple graph structures, and the output of the graph neural network model is the reference navigation route.

6. A road navigation system based on artificial intelligence, characterized in that: include: A first acquisition module is used to acquire the video and traffic status map shot in the car; An internal video processing module, used to determine the information of each passenger and the longest and shortest play time of this trip based on the video shot in the car using an internal video processing model; The second acquisition module is used to obtain the historical travel records of each passenger; A target scenic spot determination module is used to determine the target scenic spot information based on the information of each passenger, the historical travel records of each passenger, the longest travel time and the shortest travel time of the current trip, the time period of interest of each passenger during the trip, and the degree of preference of each passenger for the scenic spot; A distance determination module, used to determine the longest driving distance and the shortest driving distance of this trip based on the traffic status map, the longest playing time and the shortest playing time of this trip, and the target scenic spot information; A matching module, for matching a plurality of navigation routes based on the longest driving distance, the shortest driving distance, the longest playing time, the shortest playing time, and the target scenic spot information of the current trip, wherein each of the plurality of navigation routes includes a plurality of initial midway scenic spots; A target midway scenic spot determination module is used to screen out multiple target midway scenic spots of each navigation route from multiple initial midway scenic spots in each navigation route using a target midway scenic spot determination model based on multiple navigation route information, the interested stop time period of each passenger during the trip, and the scenic spot preference of each passenger, and mark the multiple target midway scenic spots of each navigation route in each navigation route; A display module, used to display multiple navigation routes and multiple target midway attractions of each of the multiple navigation routes, and obtain a target navigation route selected by a user from the multiple navigation routes; The navigation module is used for navigating based on the selected target navigation route.

7. The artificial intelligence-based road navigation system according to claim 6, characterized in that: The internal video processing model is a Transformer model, the input of the internal video processing model is the video shot inside the car, and the output of the internal video processing model is the information of each passenger and the longest and shortest travel time of this trip.

8. The artificial intelligence-based road navigation system according to claim 6, characterized in that: The historical travel records of each passenger include the travel attractions, travel time, and stopover time during the travel route.

9. The artificial intelligence-based road navigation system according to claim 6, characterized in that: The display module is also used to display the multiple navigation routes and the target midway attractions of each of the multiple navigation routes on the vehicle display screen.

10. The artificial intelligence-based road navigation system according to claim 6, characterized in that: The system is also used to: construct multiple graph structures based on the multiple navigation routes and the target midway attractions of each navigation route in the multiple navigation routes, wherein each graph structure represents a navigation route, each graph structure includes multiple nodes and multiple edges between the multiple nodes, each node represents a target midway attraction, the node features of the target midway attraction node are the attraction type, average stay time at the attraction, location of the attraction, and current flow of people at the attraction, the multiple edges between the multiple nodes represent the connection relationship between different target midway attractions, and the features of each of the multiple edges include the distance between two nodes and the estimated travel time; The multiple graph structures are processed based on a graph neural network model to determine a reference navigation route, the input of the graph neural network model is the multiple graph structures, and the output of the graph neural network model is the reference navigation route.

Images