Subway station area passenger flow detection and analysis method based on multi-modal perception capability

Abstract

The application belongs to the technical field of passenger flow monitoring and management, and particularly relates to a subway station area passenger flow detection analysis method based on multi-modal sensing capability, which comprises the following steps: calculating a visual effectiveness coefficient according to the effective vertical distance of a pixel point in a video frame to a near-end reference line, and then dividing the video frame into a near-field anchoring area and a far-field area to be reconstructed; calculating a dynamic radio frequency penetration factor according to the number of passenger targets in the near-field anchoring area and the number of near-field radio frequency terminals; and obtaining a reconstructed passenger flow total amount by using the visual effectiveness coefficient of the far-field area to be reconstructed to perform weighted fusion according to the number of passenger targets in the near-field anchoring area, the dynamic radio frequency penetration factor, the number of far-field radio frequency terminals and the number of passenger targets in the far-field area to be reconstructed. The application can adaptively change the crowd structure, effectively solve the problem of missed detection caused by far-end visual occlusion and improve the passenger flow detection precision.

Description

Technical Field

[0001] This invention relates to the field of passenger flow monitoring and management technology. More specifically, this invention relates to a method for passenger flow detection and analysis in subway station areas based on multimodal sensing capabilities. Background Technology

[0002] With the rapid development of digital integration in rail transit, passenger flow situation awareness in subway stations has become a core component of smart station construction. Accurately grasping the real-time passenger flow distribution and quantity within the station is of great significance for ensuring operational safety during morning and evening peak hours, optimizing capacity scheduling, and implementing effective flow control.

[0003] Currently, the mainstream passenger flow detection method in subway stations mainly relies on video surveillance technology, which uses cameras installed high up on the platform or at the end of the station hall to capture and analyze live images. However, in high-density passenger flow scenarios such as morning and evening rush hours, due to the physical installation height of the cameras and the perspective imaging principle, the video image exhibits a significant geometric characteristic of near objects appearing larger and farther objects smaller. Passengers at the lower end of the frame can easily obstruct the view of passengers at the upper end, causing a sharp, non-linear decrease in the perception ability of vision-based detection algorithms in the far-end areas of the image. This results in a large number of missed detections, leading to serious distortion of passenger flow statistics and an inability to accurately reflect the congestion situation at the station.

[0004] Furthermore, while radio frequency (RF) signal monitoring boasts advantages such as strong penetration and immunity to line-of-sight obstructions, the number of RF terminals cannot be directly equated with the actual passenger flow. This is because the proportion of people carrying and activating smart devices fluctuates in real time across different time periods and areas, and there are instances where multiple devices belong to the same passenger or where passengers have no devices at all. Directly using RF data combined with fixed empirical coefficients to estimate passenger flow would introduce significant systematic errors, making it difficult to adapt to the penetration rate fluctuations caused by changes in population structure and failing to meet the needs of accurate passenger flow detection in subway station areas. Summary of the Invention

[0005] To address the technical problems of missed detections at distant locations due to perspective obstruction in video detection under dense passenger flow conditions, and statistical distortion caused by calculation errors due to penetration fluctuations in radio frequency detection, this invention provides a passenger flow detection and analysis method for subway station areas based on multimodal sensing capabilities, including:

[0006] The visual effectiveness coefficient is calculated based on the effective vertical distance from the pixel in the video frame to the near-end baseline; the video frame is divided into a near-field anchoring area and a far-field reconstruction area based on the visual effectiveness coefficient; the number of passenger targets and the number of near-field radio frequency terminals in the near-field anchoring area are counted, and the dynamic radio frequency penetration factor is calculated.

[0007] The number of passenger targets and far-field radio frequency terminals in the far-field reconstructing area are counted. Based on the number of passenger targets in the near-field anchoring area, the dynamic radio frequency penetration factor, the number of far-field radio frequency terminals, and the number of passenger targets in the far-field reconstructing area, the visual effectiveness coefficient of the pixels in the far-field reconstructing area is used for weighted fusion to obtain the total reconstructed passenger flow.

[0008] Preferably, the near-end baseline is the bottom edge of the video frame.

[0009] Preferably, the visual effectiveness coefficient satisfies the expression:

[0010] ;

[0011] In the formula, Represents pixels Visual effectiveness coefficient; Represents pixels Effective vertical distance to the near-end baseline; Indicates the vertical resolution of the video frame; This represents the perspective attenuation constant.

[0012] Preferably, the method for obtaining the effective vertical distance is as follows:

[0013] In response to a pixel being outside the bounding box of all passenger targets, the ordinate value of the pixel is used as the effective vertical distance from the pixel to the near-end baseline.

[0014] In response to a pixel being located within the bounding box of a passenger target, the ordinate of the bottom center point of the bounding box of the passenger target to which the pixel belongs is used as the effective vertical distance from the pixel to the near-end baseline.

[0015] In response to a pixel being located within the bounding boxes of different passenger targets simultaneously, the minimum value of the ordinate of the bottom center point of the bounding boxes of all passenger targets to which the pixel belongs is taken as the effective vertical distance from the pixel to the near-end baseline.

[0016] Preferably, the step of dividing the video frame into a near-field anchoring region and a far-field reconstruction region based on the visual effectiveness coefficient includes:

[0017] Pixels in the video frame whose visual effectiveness coefficient is greater than the preset effective threshold are taken as near-end pixels, and the area formed by all near-end pixels is taken as the near-field anchoring area.

[0018] Pixels in a video frame whose visual effectiveness coefficient is less than or equal to a preset effective threshold are designated as far-end pixels, and the region formed by all far-end pixels is designated as the far-field reconstruction area.

[0019] Preferably, the counting of passenger targets and near-field radio frequency terminals within the near-field anchoring area includes:

[0020] The far-end boundary line in the video frame is determined, where the far-end boundary line is the pixel ordinate corresponding to a visual effectiveness coefficient equal to a preset effective threshold. The far-end boundary line is mapped to physical space through camera calibration. A standard signal source is deployed at the corresponding position in the physical space, and the received signal strength is measured. The received signal strength is set as the radio frequency distance threshold. Terminal data with received signal strength greater than or equal to the radio frequency distance threshold are retained, and MAC address deduplication is performed to obtain the number of near-field radio frequency terminals. When the bounding box of a passenger target is completely within the near-field anchoring area, the passenger target is determined to belong to the near-field anchoring area, and the number of passenger targets within the near-field anchoring area is counted.

[0021] The number of passenger targets and the number of far-field radio frequency terminals within the far-field reconstructing area are counted, including:

[0022] Terminal data with received signal strength less than the radio frequency distance threshold are retained, and MAC address deduplication is performed to obtain the number of far-field radio frequency terminals. When the bounding box of a passenger target is located in the far-field reconstructing area, the passenger target is determined to belong to the far-field reconstructing area, and the number of passenger targets in the far-field reconstructing area is counted.

[0023] Preferably, the dynamic radio frequency penetration factor satisfies the expression:

[0024] ;

[0025] In the formula, Indicates the current moment. This represents the dynamic radio frequency penetration factor at the current moment; Indicates the number of near-field radio frequency terminals; Indicates the number of passenger targets within the near-field anchorage area; This represents the dynamic radio frequency penetration factor at the previous moment; This represents the smoothing coefficient.

[0026] Preferably, the calculation of the dynamic radio frequency penetration factor further includes:

[0027] In response to the number of passenger targets in the near-field anchoring zone being less than the preset minimum sample threshold, the dynamic radio frequency penetration factor is stopped from being updated, and the value of the previous moment remains unchanged.

[0028] Preferably, the reconstructed total passenger flow satisfies the expression:

[0029] ;

[0030] In the formula, This represents the total passenger flow during the reconstruction. Indicates the number of passenger targets within the near-field anchorage area; Indicates the number of far-field radio frequency terminals; This represents the dynamic radio frequency penetration factor at the current moment; This represents the arithmetic mean of the visual effectiveness coefficients of all far-end pixels within the far-field region to be reconstructed; This indicates the number of passenger targets within the far-field reconstruction area.

[0031] Preferably, it further includes:

[0032] The reconstructed total passenger flow is compared with the preset first-level large passenger flow threshold; in response to the reconstructed total passenger flow exceeding the first-level large passenger flow threshold, a flow restriction instruction is sent to the automatic ticketing system, and the broadcast system is driven to play evacuation prompts.

[0033] In response to the total passenger flow remaining below the preset energy-saving threshold during the reconstruction, instructions are sent to the environmental and equipment monitoring system to reduce the operating frequency of the fresh air handling unit.

[0034] The beneficial effects of this invention are as follows: By introducing an attenuation function that conforms to optical laws, this invention maps the ordinate of a pixel in a video frame or the ordinate of the bottom center point of the bounding box of the passenger target to which the pixel belongs to a visual validity coefficient that characterizes the reliability of the data, thereby achieving the distinction between the near-field high-confidence region and the far-field low-confidence region of visual data; This invention uses the visual count within the near-field anchoring area, which is minimally affected by occlusion, as a benchmark, and combines it with the number of radio frequency terminals in the same area to solve in real time the dynamic radio frequency penetration factor characterizing the smart terminal carrying rate of the crowd, enabling the system to adapt to changes in the crowd structure at different times; Based on this, this invention... Based on the visual effectiveness coefficient of the far-field area to be reconstructed, the visual detection values ​​of the far-field area are progressively weighted and fused with the theoretical number of people calculated based on the dynamic radio frequency penetration factor. In areas with low visual effectiveness, the weight of radio frequency data is automatically increased to fill the visual blind spots. In areas with acceptable visual effectiveness, some visual information is retained to correct the random fluctuation error of the radio frequency signal. This achieves the complementary advantages of visual-based near-end and radio frequency-based far-end, effectively solving the problem of missed detection at the far end caused by perspective occlusion in high-density passenger flow scenarios and the problem of passenger flow estimation distortion caused by fluctuation of radio frequency penetration rate, thus improving the accuracy and environmental adaptability of passenger flow detection in subway stations. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the passenger flow detection and analysis method for subway station areas based on multimodal sensing capabilities in this invention;

[0036] Figure 2 This is a diagram showing the comparison of passenger flow detection results. Detailed Implementation

[0037] 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, not all, of the embodiments of the present invention. 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.

[0038] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0039] This invention discloses a method for detecting and analyzing passenger flow in subway station areas based on multimodal sensing capabilities, referring to... Figure 1 This includes steps S1-S4:

[0040] S1. Calculate the visual effectiveness coefficient based on the effective vertical distance from the pixel point in the subway station video frame to the near-end baseline.

[0041] It should be noted that subway station surveillance cameras are typically installed high up on platforms or at the ends of station halls. Their images exhibit inherent perspective geometry characteristics: nearby objects appear large and clear, while distant objects appear small and blurry. During peak hours with high passenger density, this perspective characteristic leads to severe visual occlusion. Passengers positioned at the bottom of the frame easily obscure passengers at the top, causing a sharp, non-linear decrease in the detection capability of visual algorithms in distant areas, resulting in a large number of missed detections of distant passengers. Therefore, this invention introduces an attenuation function that conforms to optical laws, mapping the pixel's vertical coordinate to a physical indicator representing the reliability of the data, thereby accurately distinguishing between usable and unusable areas of visual data.

[0042] Specifically, video frames aligned with the target area of ​​the subway station are acquired, and the video frames are processed using a pedestrian detection algorithm to identify passenger targets in the video frames and extract the bounding boxes of each passenger target. In this embodiment, the pedestrian detection algorithm uses YOLOv8. In other embodiments, implementers may also choose other target detection models, such as Faster R-CNN, according to the actual situation.

[0043] Define the bottom edge of the video frame as the near-end baseline. For any pixel in the video frame, calculate its corresponding visual effectiveness coefficient:

[0044]

[0045] In the formula, Represents pixels The visual effectiveness coefficient ranges from 0 to 1. Represents pixels Effective vertical distance to the near-end baseline; This indicates the vertical resolution of a video frame, i.e., the number of rows in the video frame; This represents the perspective attenuation constant, used to reflect the effect of the camera's pitch angle on the rate of perspective attenuation. Its empirical range is 0.1 to 0.8. The smaller the camera's pitch angle, the more severe the perspective distortion. A smaller value should be chosen to speed up the process. The attenuation rate, in this embodiment, will The value is set to 0.3. In other embodiments, the implementer can set the perspective attenuation constant according to the actual implementation situation. Alternatively, it can be reverse-calibrated by measuring the height ratio of the same reference object at the near and far ends of the camera's coverage area. The specific value; It is a natural constant.

[0046] The method for obtaining the effective vertical distance is as follows: when pixel point When located outside the bounding box of all passenger targets, the pixel Effective vertical distance to the near-end baseline Take pixel points The ordinate, i.e., the pixel point The vertical distance to the near-end baseline, when the pixel When located within the bounding box of any passenger target, the pixel Effective vertical distance to the near-end baseline Take pixel points The y-coordinate of the bottom center point of the bounding box of the passenger target, i.e., pixel number. The vertical distance from the bottom of the bounding box of the passenger target to the near-end baseline. It should be noted that when a pixel is located within the bounding boxes of different passenger targets, the minimum value of the y-coordinate of the bottom center point of the bounding boxes of all passenger targets to which the pixel belongs is taken as the effective vertical distance from the pixel to the near-end baseline. That is, the target closest to the near end is given priority because it has the lowest probability of occlusion and the highest visual reliability.

[0047] When the effective vertical distance from a pixel to the near-end baseline is smaller, it indicates that the pixel is located at the bottom of the video frame, or the bounding box of the passenger target to which the pixel belongs is located at the bottom of the video frame. At this time, the visual effectiveness coefficient approaches 1, indicating that the visual detection in this area is minimally affected by occlusion and the data is highly reliable. When the pixel or the passenger target to which the pixel belongs moves upward to the top of the video frame, the visual effectiveness coefficient decays exponentially, reflecting that the probability of detection failure in distant areas due to perspective occlusion increases significantly.

[0048] S2. Obtain the near-field anchoring area in the video frame based on the visual effectiveness coefficient, count the number of passenger targets and the number of near-field radio frequency terminals in the near-field anchoring area, and calculate the dynamic radio frequency penetration factor.

[0049] It should be noted that although radio frequency (RF) signals have the advantages of strong penetration and are not affected by line-of-sight obstruction, they cannot be directly equated with passenger flow. This is because the proportion of subway passengers carrying and activating smart devices fluctuates in real time at different times. If a fixed ratio coefficient is used to convert RF counts into passenger flow, it will introduce a huge system error. Since the visual data at the near end of the video frame is not obstructed, the number of people detected therein can be used as an approximate true value for calibration. Therefore, this invention uses the near-end area as a calibration anchor point, and utilizes accurate near-end visual counts and RF counts to solve the conversion relationship between RF and number of people at the current moment in real time, thereby enabling the system to adapt to changes in crowd structure.

[0050] Specifically, the video frames with a visual validity coefficient greater than a preset effective threshold are selected. The pixels are taken as near-end pixels, and the region formed by all near-end pixels is defined as the near-field anchoring region. In this embodiment, the effective threshold is... In other embodiments, the effective threshold can be set to 0.9, depending on the actual implementation situation. .

[0051] The far-end boundary of the near-field anchoring region in the video frame is determined by the expression for the visual effectiveness coefficient. It can be seen that when equal to the effective threshold When, the corresponding effective vertical distance This is the ordinate of the far boundary line. Taking the logarithm of both sides of the expression for the visual effectiveness coefficient and simplifying, we get:

[0052]

[0053] In the formula, The ordinate of the far boundary line is represented. This indicates the vertical resolution of a video frame, i.e., the number of rows in the video frame; Represents the perspective attenuation constant. Represents a logarithmic function with the natural constant as its base; This is the effective threshold.

[0054] To achieve alignment between image space and physical space, camera calibration needs to be completed during the system initialization phase: using a calibration board of known size or on-site landmarks, such as platform edges or ground marking lines, to establish a mapping relationship between video frame pixel coordinates and physical space coordinates.

[0055] Based on the calibration results, the vertical coordinate in the video frame is calculated as follows: The pixel row is projected onto the physical ground, forming a physical boundary line on the platform or concourse floor. A standard signal source, such as a fixed-power Wi-Fi or Bluetooth beacon, is deployed at this boundary line. The received signal strength (RSSI) within the monitored area is measured, and this value is set as the radio frequency distance threshold. The radio frequency distance threshold is used to cut out an area in the radio frequency signal space that is consistent with the physical range of the near-field anchoring area in the video frame.

[0056] The number of passenger targets within the near-field anchoring area is counted, and radio frequency signals in the environment are collected simultaneously. Signals with received signal strength less than the radio frequency distance threshold are filtered out. Signal data, retaining only those with signal strength greater than or equal to the radio frequency distance threshold. The terminal data is used to ensure that its physical location is within the near-field anchoring zone. Then, the remaining terminal data is deduplicated by MAC address to obtain the number of near-field radio frequency terminals. It should be noted that when counting the number of passenger targets within the near-field anchoring zone, a passenger target is considered to belong to the near-field anchoring zone if its bounding box is completely within the zone, and not if its bounding box is partially within the zone.

[0057] Furthermore, based on the number of passenger targets and the number of near-field radio frequency terminals within the near-field anchoring zone, the dynamic radio frequency penetration factor is determined:

[0058]

[0059] In the formula, Indicates the current moment. The dynamic radio frequency penetration factor represents the current moment and is used to characterize the amount of radio frequency signal per unit number of people in the current population. Indicates the number of near-field radio frequency terminals; Indicates the number of passenger targets within the near-field anchorage area; This represents the dynamic radio frequency penetration factor at the previous moment; This represents the smoothing coefficient, used to control the weighting of current and historical values ​​during the update process. Its value ranges from 0 to 1.

[0060] When the number of near-field radio frequency terminals Relative to the number of passenger targets within the near-field anchorage area When the number of devices increases, it indicates an increase in the number of radio frequency terminals per unit of population, meaning an increase in the smart terminal penetration rate among the population. At this point, the dynamic radio frequency penetration factor... The smoothing coefficient increases accordingly; conversely, it decreases. The smaller the value, the lower the dynamic radio frequency penetration factor. The more dependent on historical values The stronger the system's anti-interference capability against instantaneous changes in radio frequency signals, the better. In this embodiment, the smoothing coefficient is... In other embodiments, the smoothing coefficient is set to 0.1. The implementer can adjust the smoothing coefficient based on the fluctuation frequency of the on-site radio frequency signal, according to the actual implementation situation. Size.

[0061] It should be noted that in the process of calculating the dynamic radio frequency penetration factor, the number of passenger targets within the near-field anchoring zone... When the sample size is less than the preset minimum sample threshold, to prevent calculation divergence due to division by zero or sample bias, the system stops updating the dynamic radio frequency penetration factor and retains the value from the previous moment. In this embodiment, the minimum sample threshold is set to 2. In other embodiments, the implementer can set the minimum sample threshold according to the actual implementation situation.

[0062] During the system initialization phase, the initial value of the dynamic radio frequency penetration factor needs to be preset. This ensures that the system has a calculable baseline value when it is first run, and avoids the inability to calculate the dynamic radio frequency penetration factor or the divergence of results due to the lack of historical data. This initial value can be obtained from historical operational data statistics, such as the average terminal carrying rate during the morning peak hours of the past week.

[0063] S3. Obtain the far-field reconstructed area in the video frame based on the visual effectiveness coefficient, count the number of passenger targets and far-field radio frequency terminals in the far-field reconstructed area, and perform weighted fusion based on the number of passenger targets in the near-field anchoring area, dynamic radio frequency penetration factor, number of far-field radio frequency terminals and number of passenger targets in the far-field reconstructed area, using the visual effectiveness coefficient of the far-field reconstructed area to obtain the total reconstructed passenger flow.

[0064] Specifically, pixels in a video frame whose visual effectiveness coefficient is less than or equal to a preset effective threshold are designated as far-end pixels. The region formed by all far-end pixels is defined as the far-field reconstruction area, and the number of passenger targets within this area is counted. For the acquired radio frequency signals from the environment, pixels with received signal strength greater than or equal to a radio frequency distance threshold are filtered out. Signal data, only retaining signal strength less than the radio frequency distance threshold. The terminal data is used to ensure that its physical location is within the far-field reconstruction area. Then, the remaining terminal data is deduplicated by MAC address to obtain the number of far-field radio frequency terminals. It should be noted that when counting the number of passenger targets within the far-field reconstruction area, a passenger target is considered to belong to the far-field reconstruction area if its bounding box is completely or partially within the far-field reconstruction area, and not if its bounding box is completely outside the far-field reconstruction area.

[0065] Based on the number of passenger targets in the near-field anchoring zone, the number of passenger targets in the far-field reconstructing zone, the dynamic radio frequency penetration factor, and the number of far-field radio frequency terminals, the total passenger flow is reconstructed:

[0066]

[0067] In the formula, This represents the total passenger flow during the reconstruction. Indicates the number of passenger targets within the near-field anchorage area; Indicates the number of far-field radio frequency terminals; This represents the dynamic radio frequency penetration factor at the current moment; The arithmetic mean of the visual validity coefficients of all far-end pixels in the far-field region to be reconstructed is used to characterize the overall confidence of the visual detection results in the far-field region to be reconstructed. This indicates the number of passenger targets within the far-field reconstruction area. To avoid... This can cause the denominator to be zero. In actual calculations, if... Then let This ensures that the total passenger flow can be calculated stably under any circumstances. To preset a minimum non-zero threshold, in this embodiment, the minimum non-zero threshold is... The value is set to 0.01. In other embodiments, implementers can set the value according to the actual implementation situation. .

[0068] When the far field distance is extremely long, it leads to When it approaches 0, the weight term When the value approaches 1, the calculation result is mainly determined by... The decision is made by using the theoretical number of people inferred from radio frequency data to fill in visual blind spots; when the far-field distance is appropriate... When it has a certain value, The increased weight of the item indicates that the system corrects the error caused by radio frequency fluctuations by retaining some visual detection values. This invention realizes a progressive fusion reconstruction with vision as the main factor at the near end and radio frequency as the main factor at the far end.

[0069] For example, Figure 2 The diagram shows a comparison of passenger flow detection results. It can be seen that during peak passenger flow periods, due to visual failure caused by perspective occlusion, the pure visual detection results are significantly lower than the actual passenger flow, resulting in a large data gap. However, the passenger flow total curve reconstructed by this invention closely matches the actual passenger flow curve. Especially in the peak areas where the pure visual algorithm misses detections, this invention effectively compensates for the lost passenger flow by utilizing radio frequency data.

[0070] S4. Implement tiered alarm and linkage control based on the reconstructed total passenger flow.

[0071] Specifically, the reconstructed total passenger flow is compared in real time with preset tiered alarm thresholds: in response to the total passenger flow exceeding the first-level high passenger flow threshold, the first-level control plan is automatically triggered, sending a flow restriction instruction to the automatic fare collection system and driving the broadcast system to play evacuation prompts; in response to the total passenger flow continuously falling below the energy-saving threshold, an instruction is sent to the environmental and equipment monitoring system to reduce the operating frequency of the fresh air handling units. The first-level high passenger flow threshold can be set by the implementers based on the station's design carrying capacity or historical peak passenger flow; the energy-saving threshold can be set by the implementers based on the station's average passenger flow level during daily off-peak hours.

[0072] Meanwhile, the total passenger flow and its corresponding spatiotemporal labels are stored in a data warehouse for subsequent passenger flow pattern analysis and model iteration.

Claims

1. A method for passenger flow detection and analysis in subway station areas based on multimodal sensing capabilities, characterized in that, include: The visual effectiveness coefficient is calculated based on the effective vertical distance of a pixel in a video frame to the near-end baseline. In the formula, Represents pixels Visual effectiveness coefficient; Represents pixels Effective vertical distance to the near-end baseline; Indicates the vertical resolution of the video frame; Represents the perspective attenuation constant; Based on the visual effectiveness coefficient, video frames are divided into near-field anchoring areas and far-field reconstruction areas; The method for counting the number of passenger targets and near-field radio frequency terminals within the near-field anchoring zone includes: determining the far-end boundary line in the video frame, where the far-end boundary line is the pixel ordinate corresponding to a visual effectiveness coefficient equal to a preset effective threshold; mapping the far-end boundary line to physical space through camera calibration, deploying a standard signal source at the corresponding position in the physical space and measuring the received signal strength, setting the received signal strength as the radio frequency distance threshold; retaining terminal data with received signal strength greater than or equal to the radio frequency distance threshold, and performing MAC address deduplication to obtain the number of near-field radio frequency terminals; when the bounding box of a passenger target is completely within the near-field anchoring zone, the passenger target is determined to belong to the near-field anchoring zone, and the number of passenger targets within the near-field anchoring zone is counted. Calculate the dynamic radio frequency penetration factor: In the formula, Indicates the current moment. This represents the dynamic radio frequency penetration factor at the current moment; Indicates the number of near-field radio frequency terminals; Indicates the number of passenger targets within the near-field anchorage area; This represents the dynamic radio frequency penetration factor at the previous moment; Indicates the smoothing coefficient; The number of passenger targets and far-field radio frequency terminals in the far-field reconstructing area is counted, including: retaining terminal data with received signal strength less than the radio frequency distance threshold and performing MAC address deduplication to obtain the number of far-field radio frequency terminals; when the bounding box of a passenger target is located in the far-field reconstructing area, the passenger target is determined to belong to the far-field reconstructing area, and the number of passenger targets in the far-field reconstructing area is counted. Based on the number of passenger targets in the near-field anchoring area, the dynamic radio frequency penetration factor, the number of far-field radio frequency terminals, and the number of passenger targets in the far-field reconstructing area, the total reconstructed passenger flow is obtained by weighted fusion using the visual effectiveness coefficient of the pixels in the far-field reconstructing area.

2. The method for passenger flow detection and analysis in subway station areas based on multimodal perception capability according to claim 1, characterized in that, The near-end baseline is the bottom edge of the video frame.

3. The method for passenger flow detection and analysis in subway station areas based on multimodal sensing capability according to any one of claims 1-2, characterized in that, The method for obtaining the effective vertical distance is as follows: In response to a pixel being outside the bounding box of all passenger targets, the ordinate value of the pixel is used as the effective vertical distance from the pixel to the near-end baseline. In response to a pixel being located within the bounding box of a passenger target, the ordinate of the bottom center point of the bounding box of the passenger target to which the pixel belongs is used as the effective vertical distance from the pixel to the near-end baseline. In response to a pixel being located within the bounding boxes of different passenger targets simultaneously, the minimum value of the ordinate of the bottom center point of the bounding boxes of all passenger targets to which the pixel belongs is taken as the effective vertical distance from the pixel to the near-end baseline.

4. The method for passenger flow detection and analysis in subway station areas based on multimodal perception capability according to claim 1, characterized in that, The process of dividing video frames into near-field anchoring regions and far-field reconstruction regions based on visual effectiveness coefficients includes: Pixels in the video frame whose visual effectiveness coefficient is greater than the preset effective threshold are taken as near-end pixels, and the area formed by all near-end pixels is taken as the near-field anchoring area. Pixels in a video frame whose visual effectiveness coefficient is less than or equal to a preset effective threshold are designated as far-end pixels, and the region formed by all far-end pixels is designated as the far-field reconstruction area.

5. The method for detecting and analyzing passenger flow in subway station areas based on multimodal perception capability according to claim 1, characterized in that, The calculation of the dynamic radio frequency penetration factor also includes: In response to the number of passenger targets in the near-field anchoring zone being less than the preset minimum sample threshold, the dynamic radio frequency penetration factor is stopped from being updated, and the value of the previous moment remains unchanged.

6. The method for passenger flow detection and analysis in subway station areas based on multimodal perception capability according to claim 1, characterized in that, The reconstructed total passenger flow satisfies the expression: ； In the formula, This represents the total passenger flow during the reconstruction. Indicates the number of passenger targets within the near-field anchorage area; Indicates the number of far-field radio frequency terminals; This represents the dynamic radio frequency penetration factor at the current moment; This represents the arithmetic mean of the visual effectiveness coefficients of all far-end pixels within the far-field region to be reconstructed; This indicates the number of passenger targets within the far-field reconstruction area.

7. The method for passenger flow detection and analysis in subway station areas based on multimodal perception capability according to claim 1, characterized in that, Also includes: The reconstructed total passenger flow is compared with the preset threshold for the highest passenger flow. In response to the total passenger flow exceeding the first-level large passenger flow threshold, a flow restriction instruction is sent to the automatic ticketing system, and the broadcast system is driven to play evacuation prompts. In response to the total passenger flow remaining below the preset energy-saving threshold during the reconstruction, instructions are sent to the environmental and equipment monitoring system to reduce the operating frequency of the fresh air handling unit.

Images