Scenic area passenger flow saturation monitoring method based on multi-channel monitoring video analysis

Abstract

The invention relates to the technical field of data processing, in particular to a scenic spot passenger flow saturation monitoring method based on multi-channel monitoring video analysis. The method is applied to ubiquitous travel places such as scenic spots, museums and travel complexes, and based on an existing scenic spot monitoring system, scenic spot passenger flow and saturation monitoring is achieved through monitoring video structured analysis. By introducing a computer vision analysis technology, a vision algorithm is added to an existing monitoring video for analysis, repeated utilization of scenic spot basic security and protection monitoring construction is realized, and problems caused by repeated investment of scenic spots are effectively reduced. The algorithm identification effective coefficient aiming at the specific scene is generated by combining a manual sampling correction algorithm identification process, so that the problems of difficulty in landing and mass production of a traditional computer vision algorithm and low identification accuracy of a general algorithm in the specific scene are solved, and the visual algorithm is enabled to land more accurately in the specific scene.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and specifically to a method for monitoring the saturation of tourist flow in a scenic spot based on multi-channel monitoring video analysis. The method is applied to scenic spots, museums, cultural and tourism complexes and other pan-cultural and tourism places, and is based on the existing monitoring system of the scenic spot, and realizes the method of monitoring the tourist flow and saturation of the scenic spot through structured analysis of monitoring videos. Background Art

[0002] With the rapid development of the tourism industry, scenic spot passenger flow management has become an urgent problem to be solved. Traditional passenger flow monitoring methods, such as operator signaling big data, scenic spot entrance and exit gate equipment, etc., have data granularity that is too coarse, cannot track the distribution of tourists within the scenic area, are costly, and difficult to achieve real-time and accurate. In addition, passenger flow analysis at a single monitoring point also has limitations and cannot fully reflect the passenger flow distribution of the entire scenic area.

[0003] With the continuous development and popularization of video surveillance technology, more and more scenic spots have begun to use video surveillance systems for security management. However, traditional video surveillance systems are mainly used for post-event tracing and evidence collection, and lack the ability to analyze and process real-time passenger flow data. Therefore, how to use video surveillance data to monitor and guide passenger flow has become a research hotspot in the current field of scenic spot management.

[0004] At present, although some passenger flow monitoring technologies based on video analysis have been proposed, these technologies often only focus on the detection and counting of pedestrian targets, lacking the analysis and processing of deeper information such as passenger flow distribution and movement speed. In addition, these technologies are often only applicable to passenger flow analysis at a single monitoring point, and cannot achieve comprehensive monitoring of passenger flow in the entire scenic area.

[0005] Therefore, developing a scenic spot passenger flow saturation monitoring method based on multi-channel surveillance video analysis to achieve real-time monitoring and accurate judgment of the scenic spot reception situation and the passenger flow distribution in various areas within the scenic spot is of great significance for improving the scenic spot management level and ensuring the safety of tourists. Summary of the invention

[0006] The main purpose of the present invention is to provide an efficient, accurate and real-time method for monitoring the passenger flow saturation of a scenic spot based on multi-channel monitoring video analysis. Through the implementation of the present invention, the shortcomings of traditional monitoring that cannot realize the linkage analysis of multiple regional-level data in a scenic spot are made up, and comprehensive and accurate monitoring of the passenger flow in each key area of ​​the scenic spot can be achieved, thereby accurately evaluating the passenger flow saturation of the scenic spot and providing a scientific and reliable decision-making basis for scenic spot managers.

[0007] S1: Deploy monitoring equipment and the corresponding servers and software systems in the target scenic area: The monitoring equipment should cover important channels and key nodes of the scenic area, such as entrances, exits, and major tourist areas; the technical requirements for servers and software systems are as follows: S1.1: Video stream processing capability: Deploy a video stream processing subsystem with motion detection frame capture and fixed time interval frame capture capabilities. Motion detection frame capture is achieved through inter-frame difference method and is used for the entrance and exit population algorithm preprocessing described below. The fixed time interval frame capture capability is based on the video frame rate and is used for the area population algorithm preprocessing described below. S1.2: Visual algorithm processing capability: Supports real-time video decoding and analysis, with a processing capacity of no less than 8 frames of high-definition 1080p video stream per second, and the ability to identify tourist targets in the video by deploying algorithm models on the server; S1.3: Data storage and retrieval capabilities: Redis cache and MySQL distributed database system are used to support PB-level data storage and high-speed retrieval; S1.4: Deploy the following software on the edge computing device or server: The monitoring video stream preprocessing tool OpenCV is used to process the video stream into an image stream that meets the requirements; the visual algorithm model, the pre-trained YOLOX model, is used to detect pedestrian targets in the image; the business application system is responsible for data storage and business indicator calculation.

[0008] S2: Configure the corresponding algorithm for each monitoring device, which is divided into two cases according to the regional attributes: S2.1: Configuration of the entrance and exit people counting algorithm. Two counting virtual lines are drawn in the video image, and the area between the two virtual lines is regarded as the counting area.

[0009] S2.2: Regional population algorithm: a virtual area is defined in the video image and the area is used as the counting area; Capture video images at regular intervals according to business requirements and analyze the number of people in the counting area; S3: Collect the video data collected by each monitoring device, use OpenCV to process the video stream, and then use the following visual algorithm model to extract pedestrian targets from the pre-processed image, and process the entrance and exit areas and the internal sightseeing area of ​​the scenic area separately: S3.1: Entrance and exit population recognition: OpenCV is used to process the video stream into an 8-frame-per-second image stream, and then YOLOX is used to perform target detection to obtain the detection results. The high-score frames and low-score frames are processed separately. The similarity between the low-score detection frames and the tracking trajectories is used to mine the real objects from the low-score frames and filter out the background. In the process of data association, Kalman filtering is used to predict the position of the tracking trajectory of the current frame in the next frame, so as to efficiently and accurately identify and count the number of tourists at the entrance and exit. S3.2: Regional population recognition: According to the configuration described in S2.2, there should be at least one virtual counting area. By setting a timer, use OpenCV to capture a picture at a fixed time interval, and then use deep learning to generate an image density map. Use the SANet network to extract multiple scales of head information for each image. Use convolution kernels of different sizes at the same time in each convolution layer, and finally obtain the final density map through deconvolution. Then use a series of Inception structures to extract features of different scales, and then use deconvolution Transposed CONV to generate a high-resolution density map. The obtained density maps are summed to obtain the specific number of people. S4: Sample and compare the collected video data, and compare the AI ​​recognition effect with manual recognition to obtain the effective coefficient β (retain two decimal places) of tourist recognition of each monitoring device. The calculation method is as follows: Where: SRI is the number of tourists obtained by manual comparison in the first time; SAI is the number of tourists identified by the AI ​​system for the first time; N is the number of sampling inspections; Through verification, the effective coefficient β of tourist identification is matched with the result of AI identification; S5: Calculate the daily reception capacity of the scenic spot through the entrance and exit number model: Take the surveillance video data of all entrances collected during the opening hours of the scenic spot, and use the AI ​​model to identify the number of tourists T passing through each entrance. The daily reception capacity of the scenic spot is C, and the calculation method is as follows: Where: β i Identify the effectiveness coefficient for the visitor of the i-th monitoring device; T i is the number of tourists passing through the i-th entrance; S5.1: Calculate the daily passenger flow saturation of the scenic spot based on the daily carrying capacity data provided by the scenic spot management agency. The calculation method is as follows: Where: η day is the daily tourist flow saturation of the scenic spot; C is the daily reception volume of the scenic spot; P is the daily carrying capacity provided by the scenic area management agency; S6: Calculate the instantaneous passenger flow saturation of the scenic spot by identifying the number of people in the area: Collect surveillance video data of all tourist areas in the scenic spot at a certain time point during the opening hours of the scenic spot. The AI ​​model identifies that the number of tourists contained in each surveillance device video frame is E, and the total number of tourists in all tourist attractions in the scenic spot (i.e. the number of people in the scenic spot in real time) is D, which is calculated as follows: Where: Ei is the number of tourists identified by the monitoring equipment at the i-th tourist attraction. If this value is an abnormal value (negative, decimal, or obviously deviates too far from the normal value), the geometric mean of other normal identification numbers is taken; βi is the tourist identification effectiveness coefficient of the i-th monitoring device; S6.1: Calculate the instantaneous passenger flow saturation η of the scenic spot based on the instantaneous carrying capacity data provided by the scenic spot management agency 瞬 , calculated as follows: Where: η instant is the instantaneous passenger flow saturation of the scenic spot D is the total number of real-time tourists in the scenic area to which step S7 belongs; P instant is the instantaneous carrying capacity of the scenic area; At this point, the calculation process of the instantaneous passenger flow saturation of the scenic spot has been described.

[0010] Beneficial Effects of the Invention The scenic spot passenger flow saturation monitoring method based on multi-channel surveillance video analysis provided by the present invention has shown significant beneficial effects in practical applications. These effects are mainly reflected in the following aspects: 1. By introducing computer vision analysis technology, visual algorithms are added to existing surveillance videos for analysis, enabling the reuse of basic security monitoring construction in scenic spots and effectively reducing the problems caused by repeated investment in scenic spots.

[0011] 2. By introducing the concept of tourist identification effectiveness coefficient and combining it with manual sampling to correct the algorithm identification process, an algorithm identification effectiveness coefficient for specific scenarios is generated, which solves the problems of difficult implementation and mass production of traditional computer vision algorithms and low recognition accuracy of general algorithms in specific scenarios, making the vision algorithm more accurate in specific scenarios.

[0012] 3. Through the linkage analysis of multi-channel surveillance videos and combined with the monitoring mechanism specifically formulated by the cultural and tourism industry itself, the present invention can reflect the overall passenger flow situation of the scenic spot in real time, solving the pain point that the single-point local analysis of traditional monitoring technology cannot reflect the overall passenger flow situation. When the passenger flow saturation reaches or exceeds the preset threshold, the early warning mechanism is automatically triggered so that the scenic area managers can take timely response measures.

[0013] 4. By accurately monitoring the passenger flow saturation, the accuracy problem of traditional big data monitoring is solved. The traditional operator signaling and Internet manufacturer APP location data have the problem that the data range is not well defined and the data only includes their own users. The overall passenger flow of the scenic spot is estimated by the market share of their own users. In this way, the elderly, children, and non-smartphone users cannot be fully counted. The use of computer vision can effectively count the entire population, especially the multi-camera linkage monitoring proposed in this solution, which can more accurately count the overall passenger flow, so that managers can allocate resources more reasonably, optimize tourist flow lines, avoid tourist crowding and long waiting time, thereby improving tourist satisfaction and experience quality.

[0014] 5. The method of the present invention can realize real-time monitoring of the passenger flow of the scenic spot by means of multi-point combination and multi-channel monitoring linkage, ensuring that managers can grasp the passenger flow dynamics in the scenic spot in real time. Once the passenger flow saturation exceeds the preset threshold, the system will immediately trigger the early warning mechanism to provide timely decision support for managers, so that they can quickly take countermeasures to ensure the safety of the scenic spot and the satisfaction of tourists.

[0015] 6. The implementation of the present invention helps scenic spots to achieve intelligent management, reduces reliance on a large number of manual workers, and reduces management costs. At the same time, through the automated and intelligent passenger flow monitoring and early warning system, scenic spot managers can handle various emergencies more efficiently, thus improving overall management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the monitoring process; Figure 2 Schematic diagram of the configuration principle for counting tourists crossing the line. DETAILED DESCRIPTION

[0017] A method for monitoring the passenger flow saturation of a scenic spot based on multi-channel surveillance video analysis comprises the following steps: S1: Deploy monitoring equipment and the corresponding servers and software systems in the target scenic area: The monitoring equipment should cover important channels and key nodes in the scenic area, such as entrances, exits, and major tourist areas. The technical requirements for servers and software systems are as follows: S1.1: Video stream processing capability: Deploy the video stream processing subsystem to implement motion detection frame capture and fixed time interval frame capture capabilities as required. Motion detection frame capture is implemented through the inter-frame difference method and is used for the entrance and exit population algorithm preprocessing described below. The fixed time interval frame capture capability is based on the video frame rate (e.g., 8 frames per second) and is used for the area population algorithm preprocessing described below. S1.2: Visual algorithm processing capability: The server is equipped with a high-performance GPU, specifically NVDIA Tesla T4, to support real-time video decoding and analysis, with a processing capacity of no less than 8 frames of high-definition (1080p) video stream per second. By deploying an algorithm model on the server, it is able to identify tourist targets in the video. The specific identification process is described below; S1.3: Data storage and retrieval capabilities: Use Redis cache and MySQL distributed database system to support PB-level data storage and high-speed retrieval, and meet the following requirements: It supports automatic data sharding and multi-copy data replication to ensure high availability and reliability of data. It adopts a strategy that combines the eventual consistency model and the strong consistency model to meet real-time requirements and data accuracy. It supports horizontal expansion and can add nodes as needed to improve the system's storage and computing capabilities. The system adopts a microservice architecture design and supports on-demand expansion of computing and storage resources to meet the needs of scenic spots of different sizes.

[0018] S1.4: Specifically, based on the technical indicators, the following software needs to be deployed on the edge computing device or server: The monitoring video stream preprocessing tool (OpenCV) is used to process the video stream into an image stream that meets the requirements; the visual algorithm model (pre-trained YOLOX model) is used to detect pedestrian targets in the image; the business application system is responsible for data storage and business indicator calculation.

[0019] S2: Configure the corresponding algorithm for each monitoring device, which is divided into two cases according to the regional attributes: S2.1: Configuration of entrance and exit counting algorithm (refer to Appendix Figure 2 According to the principle of counting the number of visitors crossing the entrance and exit, two virtual counting lines are drawn in the video image, and the area between the two virtual lines is regarded as the counting area.

[0020] S2.2: Regional population algorithm: a virtual area is defined in the video image and the area is used as the counting area; Capture video images at regular intervals according to business requirements, and analyze the number of people in the counting area. This business requirement refers to the scheduled tasks set according to the requirements of scenic area management and control, supporting intervals from 0 seconds (uninterrupted) to 30 minutes; S3: Collect the video data collected by each monitoring device, use OpenCV to process the video stream, and then use the following visual algorithm model to extract pedestrian targets from the pre-processed image. The entrance and exit areas and the internal sightseeing areas of the scenic area are processed separately: S3.1: Recognition of the number of people at the entrance and exit: OpenCV is used to process the video stream into an 8-frame-per-second image stream, and then YOLOX is used for target detection to obtain the detection results. The high-score frame and the low-score frame are processed separately. The similarity between the low-score detection frame and the tracking trajectory is used to mine the real object from the low-score frame and filter out the background. In the process of data association, Kalman filtering is used to predict the position of the tracking trajectory of the current frame in the next frame, and the number of tourists at the entrance and exit is efficiently and accurately identified and counted. The specific process is as follows: According to the configuration described in S2.1, there should be two counting virtual lines to track the tourist objects in the picture and combine the virtual lines to determine the number of tourists received. If the position of the tracked object in the previous frame does not belong to the counting area, but the position in the current frame belongs to the counting area, the counting process starts, and the position of the previous frame is regarded as the starting position. If the position of the tracked object in the previous frame belongs to the counting area, but the position of the current frame is not within the counting range, the counting process ends, and the position of the current frame is used as the end position of the target. If the end position and the starting position of the tracked object are on both sides of the counting area, the number of people going up or down will be increased by one. When a person enters the picture, the algorithm will start tracking the human target and obtain the number of tourists in the direction of the line by counting the number of people crossing the line.

[0021] S3.2: Regional population recognition: According to the configuration described in S2.2, there should be at least one virtual counting area. By setting a timer, use OpenCV to capture a picture at a fixed time interval (the fixed time is set according to the scenic area management requirements, and supports 0 seconds (uninterrupted) to 30 minutes). Then, generate an image density map based on deep learning, extract multiple scales of head information of each image through the SANet network, use different sizes of convolution kernels at the same time in each convolution layer, and finally obtain the final density map through deconvolution. Then use a series of Inception structures to extract features of different scales, and then use deconvolution (Transposed CONV) to generate a high-resolution density map. The obtained density maps are summed to obtain the specific number of people.

[0022] S4: Further, the number of tourists identified by the above two models is verified, and the collected video data is sampled and compared. The AI ​​recognition effect is compared with manual recognition to obtain the tourist recognition effectiveness coefficient β (retain two decimal places) of each monitoring device. The calculation method is as follows: Where: SRI is the number of tourists obtained by manual comparison in the first time; SAI is the number of tourists identified by the AI ​​system for the first time; N is the number of sampling inspections.

[0023] Through verification, matching the tourist identification effectiveness coefficient β with the AI ​​identification results can effectively optimize the model's landing accuracy in specific scenarios, and based on this, the subsequent steps are carried out.

[0024] S5: Calculate the daily reception capacity of the scenic spot through the entrance and exit number model: Take the surveillance video data of all entrances collected during the opening hours of the scenic spot, and use the AI ​​model to identify the number of tourists T passing through each entrance. The daily reception capacity of the scenic spot is C, and the calculation method is as follows: Where: β i Identify the effectiveness coefficient for the visitor of the i-th monitoring device; T i is the number of tourists passing through the i-th entrance.

[0025] S5.1: Calculate the daily passenger flow saturation of the scenic spot based on the daily carrying capacity data provided by the scenic spot management agency. The calculation method is as follows: Where: η day is the daily tourist flow saturation of the scenic spot; C is the daily reception volume of the scenic spot; P is the daily carrying capacity provided by the scenic area management agency, which is formulated in accordance with the "LB / T 034-2014 Guidelines for Determining the Maximum Carrying Capacity of Scenic Spots"; At this point, the calculation process of the scenic spot's daily passenger flow saturation has been described.

[0026] Next, we start to calculate the instantaneous passenger flow saturation of the scenic spot.

[0027] S6: Calculate the instantaneous passenger flow saturation of the scenic spot by identifying the number of people in the area: collect the surveillance video data of all tourist areas in the scenic spot at a certain time point during the opening hours of the scenic spot (excluding the surveillance video data of all entrances and exits). The AI ​​model identifies that the number of tourists contained in each surveillance device video frame is E, and the total number of tourists in all tourist attractions in the scenic spot (that is, the number of people in the scenic spot in real time) is D, which is calculated as follows: Where: Ei is the number of tourists identified by the monitoring equipment at the i-th tourist attraction. If this value is an abnormal value (negative, decimal, or obviously deviates too far from the normal value), the geometric mean of other normal identification numbers is taken; βi is the tourist identification effectiveness coefficient of the i-th monitoring device.

[0028] S6.1: Calculate the instantaneous passenger flow saturation of the scenic spot based on the instantaneous carrying capacity data provided by the scenic spot management agency , calculated as follows: Where: The instantaneous passenger flow saturation of the scenic spot D is the total number of real-time tourists in the scenic area to which step S7 belongs; The instantaneous carrying capacity of the scenic spot is formulated in accordance with the "LB / T 034-2014 Guidelines for Determining the Maximum Carrying Capacity of Scenic Spots"; At this point, the calculation process of the instantaneous passenger flow saturation of the scenic spot has been described.

[0029] The following is an explanation of this technical solution through a specific implementation case: During a holiday tourist peak, a scenic spot has a total of 5 cameras connected, of which cam1 and cam2 are cameras in the entrance and exit area, and cam3, cam4, and cam5 are cameras in the sightseeing area. It is known that the daily capacity of the scenic spot is 5,000 people. The instantaneous carrying capacity of the scenic area is 1,250 people; The data from each camera is analyzed through the deployed server software, as follows: Entrance and exit population counting: Use OpenCV to process the video streams of cameras cam1 and cam2 in the entrance and exit area into image streams of 8 frames per second, and then use YOLOX to detect and track pedestrian targets and record the number of targets; Regional population counting: Use OpenCV to capture images from the tour area cameras cam3, cam4, and cam5 every 1 minute, input the YOLOX model for head and shoulder detection, and record the number of targets.

[0030] Combined with the number of manually identified tourists, the tourist identification effectiveness coefficient β of each camera was determined. After comparing the number of manually identified tourists 5 times, the following conclusions were drawn: Based on this, the daily passenger flow saturation and instantaneous passenger flow saturation of the scenic spot are calculated respectively. The case steps are as follows: First, calculate the number of people received at the entrance and exit based on the two entrance and exit cameras cam1 and cam2, and substitute them into the formula of step S5: The number of visitors per day in the scenic area C = int{Σ(Ti / βi)} = int{2500 / 0.99+2200 / 0.97}=4794 Further, the daily passenger flow saturation of the scenic spot is calculated by substituting it into the formula of step S5.1: η 日 =C / P × 100%=4794 / 5000× 100%=95.88% According to the above steps, the daily passenger flow saturation of the scenic spot is 95.88%.

[0031] Next, based on the three tourist area cameras cam3, cam4, and cam5, the instantaneous passenger flow saturation of the scenic spot is calculated. First, the total number of tourists in all tourist attractions in the scenic spot (that is, the number of people in the scenic spot in real time) D is counted and substituted into the formula of step S6: Total number of tourists to scenic spots D = int{Σ(E i / β i )}= int{192 / 1.03+67 / 1.0+132 / 0.88}=403 Furthermore, substitute into formula S6.1 to calculate the instantaneous passenger flow saturation of the scenic spot: Instantaneous passenger flow saturation of scenic spot η 瞬 =D / P 瞬 × 100%=403 / 1250× 100%=32.24% According to the above steps, the instantaneous passenger flow saturation of the scenic spot is 32.24%.

[0032] It can be seen from this embodiment that the method can be effectively used for monitoring the passenger flow of scenic spots, especially the application of multi-channel monitoring linkage analysis to form the passenger flow judgment of the overall scenic spot; Furthermore, the scenic spot load threshold is set in combination with the scenic spot’s own hardware conditions and relevant legal provisions, and corresponding measures are taken to restrict the entry of tourists and divert and guide tourists in each area to ensure the personal safety of tourists and their sightseeing experience.

Claims

1. A method for monitoring the passenger flow saturation of a scenic spot based on multi-channel surveillance video analysis, characterized in that The steps include: S1: Deploy monitoring equipment and the corresponding servers and software systems in the target scenic area: The monitoring equipment should cover important channels and key nodes of the scenic area, such as entrances, exits, and major tourist areas; the technical requirements for servers and software systems are as follows: S1.1: Video stream processing capability: Deploy a video stream processing subsystem with motion detection frame capture and fixed time interval frame capture capabilities. Motion detection frame capture is achieved through inter-frame difference method and is used for the entrance and exit population algorithm preprocessing described below. The fixed time interval frame capture capability is based on the video frame rate and is used for the area population algorithm preprocessing described below. S1.2: Visual algorithm processing capability: Supports real-time video decoding and analysis, with a processing capacity of no less than 8 frames of high-definition 1080p video stream per second, and the ability to identify tourist targets in the video by deploying algorithm models on the server; S1.3: Data storage and retrieval capabilities: Redis cache and MySQL distributed database system are used to support PB-level data storage and high-speed retrieval; S1.4: Deploy the following software on the edge computing device or server: The surveillance video stream preprocessing tool OpenCV is used to process the video stream into a picture stream that meets the requirements; the visual algorithm model, the pre-trained YOLOX model, is used to detect pedestrian targets in the image; the business application system is responsible for data storage and business indicator calculation; S2: Configure the corresponding algorithm for each monitoring device, which is divided into two cases according to the regional attributes: S2.1: Configuration of the entrance and exit people counting algorithm, demarcating two counting virtual lines in the video image, and treating the area between the two virtual lines as the counting area; S2.2: Regional population algorithm: a virtual area is defined in the video image and the area is used as the counting area; Capture video images at regular intervals according to business requirements and analyze the number of people in the counting area; S3: Collect the video data collected by each monitoring device, use OpenCV to process the video stream, and then use the following visual algorithm model to extract pedestrian targets from the pre-processed image, and process the entrance and exit areas and the internal sightseeing area of ​​the scenic area separately: S3.1: Entrance and exit population recognition: OpenCV is used to process the video stream into an 8-frame-per-second image stream, and then YOLOX is used to perform target detection to obtain the detection results. The high-score frames and low-score frames are processed separately. The similarity between the low-score detection frames and the tracking trajectories is used to mine the real objects from the low-score frames and filter out the background. In the process of data association, Kalman filtering is used to predict the position of the tracking trajectory of the current frame in the next frame, so as to efficiently and accurately identify and count the number of tourists at the entrance and exit. S3.2: Regional population recognition: According to the configuration described in S2.2, there should be at least one virtual counting area. By setting a timer, use OpenCV to capture a picture at a fixed time interval, and then use deep learning to generate an image density map. Use the SANet network to extract multiple scales of head information for each image. Use convolution kernels of different sizes at the same time in each convolution layer, and finally obtain the final density map through deconvolution. Then use a series of Inception structures to extract features of different scales, and then use deconvolution Transposed CONV to generate a high-resolution density map. The obtained density maps are summed to obtain the specific number of people. S4: Sample and compare the collected video data, and compare the AI ​​recognition effect with manual recognition to obtain the effective coefficient β (retain two decimal places) of tourist recognition of each monitoring device. The calculation method is as follows: Where: SRI is the number of tourists obtained by manual comparison in the first time; SAI is the number of tourists identified by the AI ​​system for the first time; N is the number of sampling inspections; Through verification, the effective coefficient β of tourist identification is matched with the result of AI identification; S5: Calculate the daily reception capacity of the scenic spot through the entrance and exit number model: Take the surveillance video data of all entrances collected during the opening hours of the scenic spot, and use the AI ​​model to identify the number of tourists T passing through each entrance. The daily reception capacity of the scenic spot is C, and the calculation method is as follows: Where: β i Identify the effectiveness coefficient for the visitor of the i-th monitoring device; T i is the number of tourists passing through the i-th entrance; S5.1: Calculate the daily passenger flow saturation of the scenic spot based on the daily carrying capacity data provided by the scenic spot management agency. The calculation method is as follows: Where: η day is the daily tourist flow saturation of the scenic spot; C is the daily reception volume of the scenic spot; P is the daily carrying capacity provided by the scenic area management agency; S6: Calculate the instantaneous passenger flow saturation of the scenic spot by identifying the number of people in the area: Collect surveillance video data of all tourist areas in the scenic spot at a certain time point during the opening hours of the scenic spot. The AI ​​model identifies that the number of tourists contained in each surveillance device video frame is E, and the total number of tourists in all tourist attractions in the scenic spot (that is, the number of people in the scenic spot in real time) is D, which is calculated as follows: Where: Ei is the number of tourists identified by the monitoring equipment at the i-th tourist attraction. If this value is an abnormal value (negative, decimal, or obviously deviates too far from the normal value), the geometric mean of other normal identification numbers is taken; βi is the tourist identification effectiveness coefficient of the i-th monitoring device; S6.1: Calculate the instantaneous passenger flow saturation η of the scenic spot based on the instantaneous carrying capacity data provided by the scenic spot management agency 瞬 , calculated as follows: η 瞬 =D / P 瞬 × 100% Where: η instant is the instantaneous passenger flow saturation of the scenic spot D is the total number of real-time tourists in the scenic area to which step S7 belongs; P instant is the instantaneous carrying capacity of the scenic area; At this point, the calculation process of the instantaneous passenger flow saturation of the scenic spot has been described.

Images