Public landscape personalized interaction system and method based on multi-agent reinforcement learning
By combining a multi-agent reinforcement learning system with lidar and thermal imaging sensors, high-precision density monitoring and personalized route planning for public landscapes have been achieved. This solves the problems of insufficient personnel density monitoring and lagging route planning in existing technologies, and improves the visitor experience and landscape management efficiency.
Patent Information
- Application Number
- CN202511770831.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-03
AI Technical Summary
Existing public landscapes suffer from insufficient accuracy in monitoring population density, a lack of personalization and dynamism in route planning, and a lag in response to route adjustments, leading to a decline in visitor experience and low efficiency in landscape management.
A system based on multi-agent reinforcement learning is used to accurately monitor personnel density through lidar and thermal imaging sensors, generate personalized tour routes, and adjust the routes in real time to avoid congestion. The routes are optimized by combining historical density changes and real-time location.
It achieves high-precision density monitoring and personalized route planning, reduces the average waiting time for tourists due to congestion by 30%-50%, improves visitor satisfaction, increases the number of visitors to the scenic area by 15%-20%, optimizes visitor flow distribution, and reduces management costs.
Smart Images
Figure CN121599252A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of route planning technology, specifically to a personalized interactive system and method for public landscapes based on multi-agent reinforcement learning. Background Technology
[0002] With the acceleration of urbanization and the increasing demand for spiritual and cultural enrichment among residents, public landscapes (such as urban parks, theme parks, and cultural scenic areas) have become important venues for people's daily leisure and cultural experiences. However, public landscapes currently face many pressing issues in terms of service and management: Insufficient accuracy in personnel density monitoring: Traditional personnel density statistics in scenic areas rely heavily on manual counting, infrared beam sensors, or ordinary cameras. These methods are susceptible to environmental interference (such as changes in lighting or obstructions), have large statistical errors (often exceeding 20%), and cannot distinguish between dynamic object types (such as pedestrians and non-motorized vehicles), resulting in the inability to obtain accurate real-time information on regional congestion.
[0003] The planning of visitor routes lacks personalization and dynamism: existing route recommendations are mostly based on a fixed order of attractions or the shortest distance principle, without taking into account real-time population density distribution and individual visitor needs. Visitors often mistakenly enter congested areas, leading to a decline in their visitor experience. At the same time, landscape management also finds it difficult to achieve a balanced distribution of visitor flow through route guidance, resulting in some areas operating beyond capacity during peak hours while other areas remain underutilized.
[0004] Delayed response to route adjustments: Even when congestion is detected in a region, traditional management methods often adjust routes through manual broadcasts and sign updates. This results in low information transmission efficiency, limited coverage, and an inability to dynamically adjust routes based on real-time changes in tourist location and density, leading to poor route optimization results.
[0005] The contradiction between visitor experience and landscape carrying capacity is prominent: With the development of the cultural tourism industry, the daily number of visitors to public landscapes has been increasing year by year. How to maximize the carrying capacity and operational efficiency of landscapes while ensuring that visitors have a good visiting experience (such as reasonable visit duration and low congestion) has become a core challenge faced by the industry.
[0006] Against this backdrop, there is an urgent need for a personalized interactive technology for public landscapes that integrates high-precision perception, intelligent decision-making, and dynamic optimization to address the shortcomings of existing solutions in areas such as density monitoring, route planning, and real-time adjustments, thereby achieving a dual improvement in visitor experience and landscape management efficiency. Summary of the Invention
[0007] To address the shortcomings of existing technologies, this invention provides a personalized interaction system and method for public landscapes based on multi-agent reinforcement learning, which solves the problem of personalized interaction in public landscapes that does not integrate high-precision perception, intelligent decision-making, and dynamic optimization.
[0008] To achieve the above objectives, the present invention provides the following technical solution: a personalized interaction method for public landscapes based on multi-agent reinforcement learning, comprising the following steps: Step 1: Based on the monitoring sensors involved in each scenic location, the population density of the area is detected. First, the presence of moving objects in the area is identified using lidar sensors. Then, thermal imaging sensors are used to confirm whether the moving objects are people. This process is used to assess the population density associated with the corresponding scenic area. The process includes: Based on the lidar sensors installed at the corresponding landscape locations, the contour features of objects associated with the landscape area are determined, and dynamic objects existing in the landscape area are marked based on the different radar detection features associated with consecutive time points. Then, based on the infrared detection indicators associated with the dynamic object obtained by the thermal imaging sensor, the object's temperature W is confirmed from the infrared detection indicators. i And the characteristic wavelength B of infrared radiation i Where i represents different dynamic objects, W i And B i Dynamic objects that belong to a preset range are marked as people, where the endpoint values of the preset range are all preset values; Let Z be the total number of people marked at the current moment. k Where k represents different landscape regions, and the area parameter set for the corresponding landscape region is marked as M. k M k This is the default value; Using: Personnel density = Z k ÷M k Determine the population density of the corresponding landscape area; Step Two: Determine the starting point of the visitors. Based on the starting point and the location of different scenic spots, generate several tour routes. Combine the population density and historical characteristics of each scenic area to determine the predicted density associated with that scenic area within the corresponding tour route. Then, comprehensively evaluate the recommendation priority of each different tour route to select the optimal recommended route. The specific selection method is as follows: Determine the starting point of the visitors, and then generate several tour routes by combining the different locations of different scenic spots. Each tour route is unique. The population density of the area corresponding to the scenic location is denoted as MD. kHere, k represents different landscape regions. Then, based on the historical characteristics of the current landscape location, the corresponding regional population density change curves for different periods are identified. The horizontal axis of these curves represents time, and the vertical axis represents regional population density. The total time length of the timeline is 24 hours. In the different regional population density change curves, the relationship with MD is confirmed. k Curve density TZ at the same time q Where q represents the population density variation curves in different regions, and will satisfy: |MD k -TZ q |≤0.1×MD k The regional population density change curve is denoted as the candidate curve; A set of random tour routes is selected and recorded as the routes to be processed. From the routes to be processed, the distance traveled by visitors from the starting point to different scenic spots is determined. Based on a set speed (determined in conjunction with practical experience), the travel time of visitors from the starting point to the corresponding scenic spot is determined. Simultaneously, combined with the current time, the associated time of visitors from the starting point to the corresponding scenic spot is determined. Different personnel densities corresponding to the associated time are determined in different candidate curves. The average of the confirmed personnel densities is processed and recorded as the predicted density. According to the travel direction of the routes to be processed, the predicted densities associated with different scenic spots in the routes to be processed are confirmed in turn. The confirmed predicted densities are averaged again and recorded as the first feature of the corresponding route to be processed. The route length of the corresponding route to be processed is then determined and recorded as the second feature. The comprehensive feature is determined by the formula: Comprehensive Feature = First Feature × C1 + Second Feature × C2, where C1 and C2 are preset fixed coefficient factors. Different comprehensive features associated with different tour routes are determined sequentially, and the minimum value is selected from the determined sets of comprehensive features. The tour route associated with the minimum value is recorded as the optimal recommended route.
[0009] Step 3: Based on the determined optimal recommended route, and considering the real-time location of the visitors and the real-time visitor density associated with different scenic spots, determine the degree of deviation between the predicted density and the actual visitor density. If the deviation is too large, the optimal recommended route is revised, and the revised optimal recommended route is displayed in real time. Otherwise, no processing is required. The specific method for determining the degree of deviation between the predicted density and the actual visitor density is as follows: Within the optimal recommended route, determine the next scenic spot that the visitors are to reach, and confirm the actual personnel density change curve associated with the next scenic spot from the time of the visitors' visit to the current time. Simultaneously record the associated time period, and then simultaneously confirm the predicted density change curve of the associated time period from several sets of candidate curves. From the predicted density change curve and the actual personnel density change curve, determine the density difference at the same moment, and the density difference ≥ 0. From the confirmed density difference, select the maximum value, and record the density difference associated with the maximum value as the undetermined value. Record the actual personnel density at the same moment as the undetermined value as M1. Identify whether the undetermined value meets the following condition: undetermined value ≥ 0.3 × M1. If it meets the condition, it means that the deviation is too large, and a correction signal is generated, which requires correction of the optimal recommended route. If it does not meet the condition, no processing is required.
[0010] Preferably, in step three, the process of correcting the optimal recommended route includes: For the corrected signal, determine the next scenic spots that the visitors need to reach, and record the current location of the visitors as the starting point. Then, use the same processing method as in step two to redetermine the optimal recommended route. The revised optimal recommended route will then be displayed for visitors to choose from.
[0011] Preferred, a public landscape personalized interaction system based on multi-agent reinforcement learning includes: At the density confirmation end, the population density of the area is detected based on the monitoring sensors involved in each landscape location. The dynamic objects in the area are first identified by the lidar detection sensor, and then the dynamic objects are identified as people by the thermal imaging sensor. In this way, the population density of the corresponding landscape area is assessed. The optimal route selection process involves determining the starting point of the visitors and generating several tour routes based on the starting point and the location of different scenic spots. It also involves determining the predicted density of the corresponding scenic spot in the corresponding tour route by combining the population density and historical characteristics of the area associated with each scenic spot. Finally, it involves comprehensively evaluating the recommendation priority of each different tour route to select the optimal recommended route. The real-time route correction system determines the deviation between the predicted density and the actual density of visitors based on the determined optimal recommended route, combined with the real-time location of visitors and the real-time personnel density associated with different scenic spots. If the deviation is too large, the optimal recommended route is corrected again and the corrected optimal recommended route is displayed in real time; otherwise, no processing is required.
[0012] This invention provides a personalized interaction system and method for public landscapes based on multi-agent reinforcement learning. Compared with existing technologies, it has the following advantages: Based on the tourist's starting point, landscape location, real-time density, and historical density change curves, the system quantitatively evaluates route priority through comprehensive features (first feature - predicted density mean, second feature - route length) and generates the optimal recommended route. This route ensures the shortest tourist travel distance while avoiding predicted congestion areas, meeting the personalized visiting needs of different tourists, reducing the average waiting time for tourists due to congestion by 30%-50%, and significantly improving the satisfaction of the visiting experience. By comparing the deviation between predicted and actual density in real time (a correction is triggered when the maximum density difference is ≥0.3 × actual density), the optimal route is replanned from the visitor's current location. This dynamic adjustment mechanism has a fast response time (correction cycle ≤10 minutes), which can promptly avoid sudden congestion and ensure that visitors are always in a low-congestion state, while helping landscape management to achieve a dynamic and balanced distribution of visitor flow; Through precise density monitoring and intelligent route guidance, the distribution of visitors to public landscapes is more balanced, avoiding overload in some areas. According to calculations, the daily effective reception capacity of the landscape can be increased by 15%-20% after adopting this solution, while reducing facility maintenance costs and management manpower input, thereby maximizing the operational benefits of the landscape. Attached Figure Description
[0013] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation
[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0015] As a first embodiment of this application, please refer to Figure 1 This application provides a method for personalized interaction in public landscapes based on multi-agent reinforcement learning, including the following steps: Step 1: Based on the monitoring sensors involved in each scenic location, the population density of the area is detected. First, the dynamic objects in the area are identified by the lidar detection sensor, and then the dynamic objects are identified as people by the thermal imaging sensor. This is used to assess the population density of the corresponding scenic area. Specifically, different scenic areas have different population characteristics. Based on the corresponding monitoring sensors, the different population characteristics associated with different scenic areas can be effectively identified, thereby determining the population density associated with the corresponding scenic area. Step 2: Determine the starting point of the visitors (usually marked with a QR code, which can be scanned to confirm the starting point, allowing the system to plan a tour route). Based on the starting point and the location of different scenic spots, generate several tour routes. Combine the population density and historical characteristics of each scenic area to determine the predicted density of the corresponding scenic area in the corresponding tour route. Then, comprehensively evaluate the recommendation priority of each different tour route to select the optimal recommended route. Specifically, when visiting the scenic spots, visitors can effectively plan the tour route between scenic spots based on the determined starting point and the location of each scenic area. Step 3: Based on the determined optimal recommended route, combined with the real-time location of the visitors and the real-time personnel density associated with different scenic spots, determine the degree of deviation between the predicted density and the actual personnel density. If the deviation is too large, the optimal recommended route will be revised and the revised optimal recommended route will be displayed in real time. Otherwise, no processing is required. Specifically, step two in the actual prediction process cannot be guaranteed to be accurate. It is only based on a comprehensive reference of the population density flow characteristics in historical data. Therefore, in order to further optimize the visitor route, it is necessary to comprehensively optimize the recommended route in combination with the actual operation process.
[0016] A personalized interactive system for public landscapes based on multi-agent reinforcement learning includes: At the density confirmation end, the population density of the area is detected based on the monitoring sensors involved in each landscape location. The dynamic objects in the area are first identified by the lidar detection sensor, and then the dynamic objects are identified as people by the thermal imaging sensor. In this way, the population density of the corresponding landscape area is assessed. The optimal route selection process involves determining the starting point of the visitors and generating several tour routes based on the starting point and the location of different scenic spots. It also involves determining the predicted density of the corresponding scenic spot in the corresponding tour route by combining the population density and historical characteristics of the area associated with each scenic spot. Finally, it involves comprehensively evaluating the recommendation priority of each different tour route to select the optimal recommended route. The real-time route correction system determines the deviation between the predicted density and the actual density of visitors based on the determined optimal recommended route, combined with the real-time location of visitors and the real-time personnel density associated with different scenic spots. If the deviation is too large, the optimal recommended route is corrected again and the corrected optimal recommended route is displayed in real time; otherwise, no processing is required.
[0017] As a second embodiment of this application, based on Embodiment 1, there is a further embodiment, specifically including the process of determining the population density of the corresponding area within the landscape area in step one: Based on the lidar sensors installed at the corresponding landscape locations, the contour features of objects associated with the landscape area are determined. Based on the different radar detection features associated with consecutive moments, dynamic objects existing in the landscape area are marked. Specifically, the lidar emits laser beams at high frequency (usually hundreds of thousands to millions of points per second) towards the target area. By calculating the round-trip time (TOF) or phase difference, the three-dimensional coordinates (x, y, z) of each reflection point are generated, forming a "point cloud" data of the coverage area. The precise coordinates of people in space (with an error usually ≤10cm) can be obtained, supporting the differentiation of targets at different heights (such as standing people, children, and people on seats). Through matching and differential calculation of continuous frame point clouds, the movement direction, speed (such as walking speed 0.5-1.5m / s), and path trajectory of people are obtained. After point cloud clustering, the number of people in the area can be counted. Combined with spatial grid division (such as 1m×1m), high-precision density distribution data is generated. Since the method of detecting dynamic objects by lidar is relatively common in existing technologies, it will not be elaborated on here. Then, based on the infrared detection indicators associated with the dynamic object obtained by the thermal imaging sensor, the object's temperature W is confirmed from the infrared detection indicators. i And the characteristic wavelength B of infrared radiation i Where i represents different dynamic objects, W i And B i Dynamic objects that belong to the preset range are marked as people. The endpoint values of the preset range are all preset values, which are determined in advance by the operator based on experience. The preset range for temperature calibration is generally set to [36°, 37°], and the preset range for wavelength calibration is generally set to [8μm, 14μm]. Let Z be the total number of people marked at the current moment. k (A detection needs to be performed once within each time period, for example, once every 10 minutes. This cycle is to be determined in advance by relevant personnel.) Here, k represents different landscape areas, and the area parameter set for the corresponding landscape area is denoted as M. k M k These are preset values, and their specific values are determined in advance by the operator based on actual measurements. Using: Personnel density = Z k ÷M k Determine the population density of the corresponding landscape area; Specifically, based on the determined density characteristics, in the subsequent feature analysis and processing, in order to ensure that each visitor has a better visit experience, it is necessary to adjust the visitor's route to guarantee that the visitor will have a better visit experience during the actual visit.
[0018] As a third embodiment of this application, based on embodiment one, there is a further embodiment, specifically including the following method for selecting the optimal recommended route in step two: Determine the starting point of the visitors, and then generate several tour routes by combining the different locations of different scenic spots. Each tour route is unique. The population density of the area corresponding to the scenic location is denoted as MD. k Here, k represents different landscape regions. Then, based on the historical characteristics of the current landscape location, the corresponding regional population density change curves for different periods are identified. The horizontal axis of these curves represents time, and the vertical axis represents regional population density. The total time length of the timeline is 24 hours (i.e., the corresponding time period). In the different regional population density change curves, the relationship with MD is confirmed. k Curve density TZ at the same time q Where q represents the population density variation curves in different regions, and will satisfy: |MD k -TZ q |≤0.1×MD k The regional population density change curve is denoted as the candidate curve; A set of random tour routes is selected as the routes to be processed. From the routes to be processed, the distance traveled by visitors from the starting point to different scenic spots is determined. Based on a set speed (determined in conjunction with practical experience), the travel time of visitors from the starting point to the corresponding scenic spot is determined. Simultaneously, combined with the current time, the associated time of visitors from the starting point to the corresponding scenic spot is determined. Different personnel densities corresponding to the associated time are determined within different candidate curves. The average of the confirmed personnel densities is processed and recorded as the predicted density. According to the travel direction of the routes to be processed, the predicted densities associated with different scenic spots within the routes to be processed are confirmed sequentially. The average of the confirmed predicted densities is processed again and recorded as the first feature of the corresponding routes to be processed. The route length of the corresponding routes to be processed is then determined and recorded as the second feature. The formula is: Comprehensive Feature = First Feature × C1 + Second Feature × C2. This determines the comprehensive feature associated with the route to be processed. C1 and C2 are preset fixed coefficient factors, and their specific values are determined by the operator based on experience. In this formula, the smaller the values associated with the first feature and the second feature, the smaller the associated comprehensive feature. That is to say, the shorter the travel distance and the lower the associated personnel density, the smaller the resulting comprehensive feature, thus effectively ensuring the actual processing effect of the comprehensive feature. Different comprehensive features associated with different tour routes are determined sequentially, and the minimum value is selected from several sets of determined comprehensive features. The tour route associated with the minimum value is recorded as the optimal recommended route and displayed. When visitors see this optimal recommended route, they can effectively follow this optimal recommended route to avoid congested areas during the tour and avoid them in time, so as to achieve the best attraction tour effect. As a fourth embodiment of this application, based on embodiment one, there is a further embodiment, specifically including the following method for determining the degree of deviation between the predicted density and the actual population density in step three: Within the optimal recommended route, determine the next scenic spot that the visitors will arrive at, and confirm the actual personnel density change curve associated with the next scenic spot from the visit time to the current time. Simultaneously record the associated time period (visit time - current time), and then simultaneously confirm the predicted density change curve of the associated time period from several sets of candidate curves. From the predicted density change curve and the actual personnel density change curve, determine the density difference at the same moment, and the density difference ≥ 0 (that is, it needs to be processed by absolute value). From the confirmed density difference, select the maximum value, and record the density difference associated with the maximum value as the undetermined value. Record the actual personnel density at the same moment as the undetermined value as M1. Identify whether the undetermined value meets the following condition: undetermined value ≥ 0.3 × M1. If it meets the condition, it means that the deviation is too large, and a correction signal is generated, and the optimal recommended route needs to be corrected. If it does not meet the condition, no processing is required. The process of correcting the optimal recommended route includes: For the corrected signal, determine the next scenic spots that the visitors need to reach, and record the current location of the visitors as the starting point. Then, use the same processing method as in step two to redetermine the optimal recommended route. The revised optimal recommended route will be displayed for visitors to choose from; Specifically, the real-time correction method used here can effectively prevent visitors from being in the optimal visiting state, thereby significantly reducing congestion time, improving visiting efficiency, and simultaneously increasing the daily reception capacity of the corresponding attractions.
[0019] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.
[0020] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A personalized interaction method for public landscapes based on multi-agent reinforcement learning, characterized in that, Includes the following steps: Step 1: Detect the population density of the area based on the monitoring sensors involved in each landscape location. First, use lidar sensors to identify dynamic objects in the area, and then use thermal imaging sensors to confirm whether the dynamic objects are people. In this way, assess the population density of the corresponding landscape area. Step 2: Determine the starting point of the visitors. Based on the starting point and the location of different scenic spots, generate several tour routes. Combine the population density and historical characteristics of each scenic area to determine the predicted density of the corresponding scenic area in the corresponding tour route. Then, comprehensively evaluate the recommendation priority of each different tour route to select the optimal recommended route.
2. The method for personalized interaction of public landscapes based on multi-agent reinforcement learning according to claim 1, characterized in that, Step one, the process of determining the population density of corresponding areas within the landscape area, includes: Based on the lidar sensors installed at the corresponding landscape locations, the contour features of objects associated with the landscape area are determined, and dynamic objects existing in the landscape area are marked based on the different radar detection features associated with consecutive time points. Then, based on the infrared detection indicators associated with the dynamic object obtained by the thermal imaging sensor, the object's temperature W is confirmed from the infrared detection indicators. i And the characteristic wavelength B of infrared radiation i Where i represents different dynamic objects, W i And B i Dynamic objects that belong to a preset range are marked as people, where the endpoint values of the preset range are all preset values; Let Z be the total number of people marked at the current moment. k Where k represents different landscape regions, and the area parameter set for the corresponding landscape region is marked as M. k M k This is the default value; Using: Personnel density = Z k ÷M k Determine the population density of the corresponding landscape area.
3. The method for personalized interaction of public landscapes based on multi-agent reinforcement learning according to claim 1, characterized in that, In step two, the specific method for determining the optimal recommended route based on the predicted density is as follows: Determine the starting point of the visitors, and then generate several tour routes by combining the different locations of different scenic spots. Each tour route is unique. The population density of the area corresponding to the scenic location is denoted as MD. k Here, k represents different landscape regions. Then, based on the historical characteristics of the current landscape location, the corresponding regional population density change curves for different periods are identified. The horizontal axis of these curves represents time, and the vertical axis represents regional population density. The total time length of the timeline is 24 hours. In the different regional population density change curves, the relationship with MD is confirmed. k Curve density TZ at the same time q Where q represents the population density variation curves in different regions, and will satisfy: |MD k -TZ q |≤0.1×MD k The regional population density change curve is denoted as the candidate curve; A set of random tour routes is selected as the routes to be processed. From the routes to be processed, the distance traveled by visitors from the starting point to different scenic spots is determined. Based on a set speed (determined in conjunction with practical experience), the travel time of visitors from the starting point to the corresponding scenic spot is determined. Simultaneously, combined with the current time, the associated time of visitors from the starting point to the corresponding scenic spot is determined. Different personnel densities corresponding to the associated time are determined within different candidate curves. The average of the confirmed personnel densities is processed and recorded as the predicted density. According to the travel direction of the routes to be processed, the predicted densities associated with different scenic spots within the routes to be processed are confirmed sequentially.
4. The method for personalized interaction of public landscapes based on multi-agent reinforcement learning according to claim 3, characterized in that, In step two, the specific method for selecting the optimal recommended route is as follows: The confirmed predicted densities are averaged again and recorded as the first feature of the corresponding route to be processed. The route length of the corresponding route to be processed is then determined and recorded as the second feature. The comprehensive feature is determined by the formula: Comprehensive Feature = First Feature × C1 + Second Feature × C2, where C1 and C2 are preset fixed coefficient factors. Different comprehensive features associated with different tour routes are determined sequentially, and the minimum value is selected from the determined sets of comprehensive features. The tour route associated with the minimum value is recorded as the optimal recommended route.
5. The method for personalized interaction of public landscapes based on multi-agent reinforcement learning according to claim 1, characterized in that, Also includes: Step 3: Based on the determined optimal recommended route, and combined with the real-time location of the visitors and the real-time personnel density associated with different scenic spots, determine the degree of deviation between the predicted density and the actual personnel density. If the deviation is too large, the optimal recommended route will be revised and the revised optimal recommended route will be displayed in real time. Otherwise, no processing is required.
6. The method for personalized interaction of public landscapes based on multi-agent reinforcement learning according to claim 5, characterized in that, In step three, the specific method for determining the degree of deviation between the predicted density and the actual population density is as follows: Within the optimal recommended route, determine the next scenic spot that the visitors are to reach, and confirm the actual personnel density change curve associated with the next scenic spot from the time of the visitors' visit to the current time. Simultaneously record the associated time period, and then simultaneously confirm the predicted density change curve of the associated time period from several sets of candidate curves. From the predicted density change curve and the actual personnel density change curve, determine the density difference at the same moment, and the density difference ≥ 0. From the confirmed density difference, select the maximum value, and record the density difference associated with the maximum value as the undetermined value. Record the actual personnel density at the same moment as the undetermined value as M1. Identify whether the undetermined value meets the following condition: undetermined value ≥ 0.3 × M1. If it meets the condition, it means that the deviation is too large, and a correction signal is generated, which requires correction of the optimal recommended route. If it does not meet the condition, no processing is required.
7. The method for personalized interaction of public landscapes based on multi-agent reinforcement learning according to claim 6, characterized in that, Step three, the process of correcting the optimal recommended route, includes: For the corrected signal, determine the next scenic spots that the visitors need to reach, and record the current location of the visitors as the starting point. Then, use the same processing method as in step two to redetermine the optimal recommended route. The revised optimal recommended route will then be displayed for visitors to choose from.
8. A public landscape personalized interaction system based on multi-agent reinforcement learning, wherein the system operates according to any one of claims 1-7, characterized in that, include: At the density confirmation end, the population density of the area is detected based on the monitoring sensors involved in each landscape location. The dynamic objects in the area are first identified by the lidar detection sensor, and then the dynamic objects are identified as people by the thermal imaging sensor. In this way, the population density of the corresponding landscape area is assessed. The optimal route selection process involves determining the starting point of the visitors and generating several tour routes based on the starting point and the location of different scenic spots. It also involves determining the predicted density of the corresponding scenic spot in the corresponding tour route by combining the population density and historical characteristics of the area associated with each scenic spot. Finally, it involves comprehensively evaluating the recommendation priority of each different tour route to select the optimal recommended route. The real-time route correction system determines the deviation between the predicted density and the actual density of visitors based on the determined optimal recommended route, combined with the real-time location of visitors and the real-time personnel density associated with different scenic spots. If the deviation is too large, the optimal recommended route is corrected again and the corrected optimal recommended route is displayed in real time; otherwise, no processing is required.