Bus passenger re-identification control method and system

Abstract

The application belongs to the technical field of passenger re-identification, and provides a bus passenger re-identification control method and system. The method comprises the following steps: capturing a first video image of a boarding line area, determining each first passenger with a boarding intention according to the first video image, and determining and collecting first passenger features of each first passenger; capturing a second video image of an alighting line area, determining each second passenger with an alighting intention according to the second video image, and collecting second passenger features of each second passenger; performing passenger re-identification based on the second passenger features and the first passenger features to identify second passengers and corresponding first passengers belonging to the same passenger; and recording the boarding station and the alighting station of the same passenger. The application can realize accurate re-identification of passengers on a bus, thereby providing accurate basic data for bus passenger flow data analysis.

Description

Technical Field

[0001] This invention relates to the field of passenger re-identification technology, and more specifically, to a method and system for controlling the re-identification of bus passengers. Background Technology

[0002] With the acceleration of urbanization, public transportation systems are carrying an ever-increasing passenger volume. As a key component of urban transportation, accurate statistics and analysis of passenger flow data are of great significance to the operation and management of bus companies. By accurately grasping passenger boarding and alighting information at each stop, such as where someone boards and alights, bus companies can optimize route planning, rationally arrange departures, and improve vehicle dispatch efficiency, thereby effectively improving service quality, meeting the travel needs of citizens, and reducing operating costs.

[0003] Currently, public transport passenger flow data is mainly obtained by relying on card swipe data. The card swipe system automatically counts passenger flow by having passengers swipe their cards, providing some travel information. However, card swipe data cannot cover all passengers (such as those who purchase tickets with cash), and it cannot effectively identify non-swipe or abnormal swipe behavior, resulting in missing or inaccurate data. This significantly reduces the accuracy of current public transport passenger flow statistics and fails to provide comprehensive and reliable data support for public transport companies.

[0004] Video-based bus passenger flow analysis is a novel technological approach, its core being passenger re-identification technology. This involves re-identifying passengers boarding and alighting to determine their boarding and alighting trajectories, thereby analyzing passenger flow patterns. However, existing re-identification technologies primarily rely on continuous tracking of passengers throughout their journey, resulting in excessive computational demands. Furthermore, the selection of passenger features has not been optimized, making re-identification errors prone to occur when passenger characteristics change, leading to incorrect boarding and alighting trajectories or even the inability to generate them.

[0005] Therefore, there is an urgent need for a more efficient and accurate passenger re-identification scheme to solve the above-mentioned problems of existing technologies and provide reliable data support for public transportation operation management. Summary of the Invention

[0006] In response, the present invention provides a method, system, electronic device, computer storage medium, and computer program product for re-identifying bus passengers, in order to solve at least one of the above-mentioned technical problems.

[0007] In a first aspect, the present invention provides a method for controlling the re-identification of bus passengers, comprising the following steps: capturing a first video image of a boarding area; determining each first passenger with boarding intent based on the first video image; determining and collecting first passenger features for each first passenger; capturing a second video image of a disembarking area; determining each second passenger with disembarking intent based on the second video image; collecting second passenger features for each second passenger; wherein the first passenger features are determined based on the detection success rate of passenger features in the second video image; performing passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger belonging to the same passenger and the corresponding first passenger; and associating and recording the boarding and disembarking stations of the same passenger.

[0008] In a second aspect, the present invention provides a bus passenger re-identification control system, the system comprising a first capture unit, a second capture unit, a re-identification unit, and an association unit; wherein: the first capture unit captures a first video image of a boarding area, determines each first passenger with boarding intent based on the first video image, and determines and collects first passenger features for each first passenger; the second capture unit captures a second video image of a disembarking area, determines each second passenger with disembarking intent based on the second video image, and collects second passenger features for each second passenger; wherein the first passenger features are determined based on the detection success rate of passenger features in the second video image; the re-identification unit performs passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger belonging to the same passenger and the corresponding first passenger; the association unit records the boarding and disembarking stations of the same passenger in association.

[0009] In a third aspect, the present invention provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any of the preceding claims.

[0010] In a fourth aspect, the present invention provides a computer storage medium storing a computer program that can be executed by a processor to implement the method as described in any of the preceding claims.

[0011] In a fifth aspect, the present invention provides a computer program product comprising a computer program executable by a processor to implement the method as described in any of the preceding claims.

[0012] This method captures video images of the boarding and alighting areas using a camera, employs object detection and tracking algorithms to determine passenger intent and collect features, filters key features based on detection success rate, and combines this with a re-identification algorithm to achieve accurate matching, ultimately associating and recording boarding and alighting stations. This effectively solves the problem of missing traditional passenger flow statistics, improves passenger identification accuracy, helps bus companies understand travel patterns, optimize routes and vehicle allocation, and enhance operational efficiency and service quality. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart illustrating a bus passenger re-identification control method disclosed in an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of the structure of a bus passenger re-identification control system disclosed in an embodiment of the present invention. Detailed Implementation

[0016] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0018] like Figure 1 As shown, this embodiment of the invention discloses a method for re-identifying and controlling bus passengers, including the following steps: S10, capturing a first video image of the boarding line area, determining each first passenger with boarding intent based on the first video image, and determining and collecting the first passenger features of each first passenger.

[0019] This invention employs a vision-based solution, including monocular vision and binocular stereo vision, and uses a dual-line approach to determine passengers' boarding and alighting intentions. Specifically, boarding lines and corresponding cameras are deployed at the bus's boarding door area, and disembarking lines and corresponding cameras are deployed at the bus's alighting door area. This eliminates the need for continuous tracking of passengers on the bus, reducing computational power consumption.

[0020] When a bus stops at a station, a camera captures the first video image of the boarding area. The boarding area is a pre-defined region located near the bus doors, specifically designed to determine if passengers intend to board. After acquiring the first video image, object detection and tracking algorithms are used to analyze the people in the image and identify the first passengers who intend to board. For example, if a person is detected moving towards the boarding door and their trajectory matches the boarding direction, it is determined that the person has a clear intention to board.

[0021] After identifying each first passenger, their first passenger features are collected, including but not limited to distinctive information such as hairstyle, clothing color, style, accessories, and items carried, for subsequent passenger re-identification. For example, an unsupervised deep learning model can be used to train a feature extractor to extract robust passenger features (e.g., 512 dimensions).

[0022] S20, capture a second video image of the drop-off area, determine each second passenger who intends to get off the bus based on the second video image, and collect the second passenger features of each second passenger; wherein, the first passenger features are determined based on the detection success rate of passenger features in the second video image.

[0023] A second video image is captured by a camera in the designated area near the exit doors to determine passengers' intention to disembark. This second video image is then analyzed using object detection and tracking technology to identify passengers who intend to disembark, such as those moving towards the exit doors.

[0024] The first passenger feature is determined based on its detection success rate in the second video image. That is, features easily detected in the disembarkation video image are prioritized as the first passenger feature to improve the accuracy of subsequent re-identification. For example, if a passenger's distinctive hat is easily identifiable in the disembarkation image, then the passenger's hat feature in the first video image will be used as the first passenger feature.

[0025] It is understandable that the second passenger feature includes more passenger features than the first passenger feature, but only the passenger features corresponding to the first passenger feature are used during re-identification.

[0026] S30, based on the second passenger features and the first passenger features, perform passenger re-identification to identify the second passenger belonging to the same passenger group and the corresponding first passenger.

[0027] The collected features of the second passenger are compared and analyzed with those of the first passenger. A passenger re-identification algorithm is then used to identify the second passenger and its corresponding first passenger who belong to the same person. Because passengers may experience occlusion or changes in posture in public transportation scenarios, the re-identification algorithm needs to consider multiple factors, such as feature similarity and feature stability across different images. By calculating the similarity score between features and setting an appropriate threshold, it determines which second and first passengers belong to the same person, thus achieving accurate matching of passengers across images from different stations and at different times.

[0028] S40, associate the boarding and alighting stations of the same passenger into a record.

[0029] Once the first and second passengers belonging to the same passenger group are successfully identified, their boarding and alighting points are linked and recorded. This recorded data can be used by the bus company to analyze passenger travel patterns, optimize bus routes, and rationally allocate vehicle resources. For example, by recording a large amount of passenger boarding and alighting point data, the bus company can identify areas with concentrated passenger flow on certain routes during specific time periods, thereby increasing the number of buses during those periods and improving the quality of bus service.

[0030] This method captures video images of the boarding and alighting areas using a camera, employs object detection and tracking algorithms to determine passenger intent and collect features, filters key features based on detection success rate, and combines this with a re-identification algorithm to achieve accurate matching, ultimately associating and recording boarding and alighting stations. This effectively solves the problem of missing traditional passenger flow statistics, improves passenger identification accuracy, helps bus companies understand travel patterns, optimize routes and vehicle allocation, and enhance operational efficiency and service quality.

[0031] As an example, the first passenger feature is determined based on the success rate of passenger feature detection in the second video image, specifically including: determining the first congestion level of passengers in real time inside the bus, and the median value of the historical number of passengers getting off at subsequent bus stops; predicting the second congestion level of the disembarkation line area when the bus stops at subsequent bus stops based on the first congestion level and the median value; when the second congestion level is higher than the congestion level threshold, determining a height marker based on the shooting angle height of the second video image, and identifying each passenger feature on the passenger's body that is higher than the height marker as the first passenger feature.

[0032] Traditional passenger re-identification technologies typically employ fixed feature collection strategies, such as using passengers' hairstyles, clothing colors and styles, accessories, and carried items as passenger features for re-identification. This fixed feature collection strategy, which does not consider the dynamic changes in the public transportation scenario, is prone to obscuring certain passenger features (such as trouser color and crossbody bag style) when the alighting area is crowded during peak hours (i.e., when many people alight). This leads to a significant decrease in the detection success rate in the second video image. Moreover, indiscriminately collecting all features increases computational costs, and features with low stability may interfere with matching accuracy.

[0033] To address this, the present invention designs a dynamic feature acquisition strategy, which predicts the congestion level of the subsequent bus stop drop-off area by combining dual data. Specifically, it predicts the second congestion level based on the real-time first congestion level of passengers inside the bus and the median value of the historical number of passengers getting off at the stop at subsequent bus stops.

[0034] In public transportation operations, relying solely on real-time passenger congestion levels inside the bus cannot fully account for the unique boarding and alighting characteristics of each stop. For example, even if the bus is not currently crowded at some transfer stations, the number of passengers getting off later (i.e., the number of people in the alighting area) may be relatively high. On the other hand, simply relying on the historical median number of passengers getting off at that stop at subsequent bus stops is insufficient to cope with sudden surges in passenger flow and other abnormal situations. Therefore, this invention combines these two key data points to more accurately predict the congestion level of the alighting area at subsequent stops.

[0035] The second level of congestion is calculated, for example, as follows: ;in, The second level of congestion. The highest level of congestion. This represents the median number of people who disembarked in the past. It is over time The dynamically changing weight coefficients are determined using the following reinforcement learning algorithm: [The algorithm uses the current time...] First level of congestion Median number of passengers disembarking in history The error between the predicted second congestion level and the actual second congestion level at the previous moment, together constitute the state-to-state curve. .

[0036] At every moment The agent can choose to adjust and The value of , the action space is defined as a combination of weights within a certain range.

[0037] Set reward function If the predicted second congestion level deviates from the actual level by less than a set threshold, a positive reward is given; otherwise, a negative reward is given. For example... .

[0038] The agent continuously interacts with the environment, optimizes its strategy based on reward feedback, and updates weights using algorithms such as Deep Q-Network (DQN) to maximize long-term cumulative rewards.

[0039] When the predicted second level of congestion exceeds a pre-set congestion threshold, it indicates a high probability of severe obstruction in the disembarkation area. In this case, a suitable height marker is calculated based on the shooting angle of the cameras deployed in the disembarkation area. This height marker will be virtually projected onto passengers in the boarding area. Areas below this height marker are highly likely to be obstructed in the second video image. Therefore, features on the passenger's body above this height marker, such as hairstyle, distinctive hats, or high-positioned backpacks (e.g., shoulder bags), are identified as the first passenger features because these features have a high detection success rate in the second video image.

[0040] Through such a dynamic feature selection mechanism, it can proactively adapt to complex public transportation scenarios, prioritizing the collection and identification of features that are less affected by occlusion and have high stability, thus avoiding wasting computing resources on low-value features that are easily occluded.

[0041] It is understandable that the real-time passenger density on a bus can be calculated by counting the number of passengers boarding and alighting using cameras placed at the boarding and alighting locations, or by installing a panoramic camera inside the bus to extract the number of passengers and then assess the density. No specific limitations are imposed on this method. Historical alighting data at bus stops can be stored in the bus's memory or retrieved from the bus management system; no specific limitations are imposed on this method either.

[0042] As an example, the method further includes: determining several third passengers who are companions of the first passenger based on the first video image; then performing passenger re-identification based on the features of the second passenger and the features of the first passenger to identify the second passenger who belongs to the same passenger and the corresponding first passenger, including: performing passenger re-identification based on the features of the second passenger and the features of the first passenger to obtain a first probability that the second passenger and the corresponding first passenger belong to the same passenger; if the first probability is lower than a specified threshold and the second probability that the fourth passenger in the second video image and the third passenger belong to the same passenger is higher than a specified threshold, then: extracting the conversation features and / or the assistive features of the second passenger and the fourth passenger, and evaluating the companion probability accordingly; when the companion probability is higher than the probability threshold, identifying the second passenger and the corresponding first passenger as belonging to the same passenger.

[0043] During the process of re-identifying passengers on public transport, it is common for passengers to change their external characteristics after boarding the bus, such as taking off their hats or changing their coats. This makes it easy for feature matching to fail when relying on the external characteristics of a single individual for re-identification, resulting in the inability to accurately associate the boarding and alighting information of the same passenger.

[0044] To address the aforementioned technical issues, this invention designs a fellow passenger relationship reasoning mechanism, as follows: First, during the boarding phase of the bus, based on the first video image, and based on the conversation characteristics (e.g., face-to-face communication, gesture interaction, etc.) and the assistance characteristics of the items carried (e.g., jointly carrying, watching over items) of the first passenger, at least one third passenger who is a fellow passenger of the first passenger is identified, and the third passenger characteristics of each third passenger are collected.

[0045] When performing passenger re-identification, following the re-identification method of the aforementioned embodiment, the features of the second passenger are first matched with the recorded features of the first passenger to obtain a first probability that the second passenger and the corresponding first passenger belong to the same passenger. If the first probability is lower than a specified threshold (but still higher than the lower limit threshold, indicating that the first probability is in a gray area that cannot be directly excluded), and the second probability that the fourth passenger in the second video image and the recorded third passenger belong to the same passenger is higher than the specified threshold, then the conversation features and / or accessory assistance features of the second and fourth passengers are extracted to assess the probability that the second and fourth passengers are traveling together.

[0046] When the probability of being a passenger is higher than a certain threshold, it can be determined that the second and fourth passengers are the first and third passengers who were identified as passengers when they boarded the bus. At this point, the second passenger, whose probability is lower than a certain threshold, is considered to be the same passenger as the corresponding first passenger. In this way, even if the external characteristics of the first passenger change, the second passenger can still be identified as the same passenger as the corresponding first passenger through their association with their fellow passengers.

[0047] As an example, the probability threshold is determined by: calculating the number of third passengers and taking it as the number of companions; matching the number of companions to obtain a preliminary probability threshold; the preliminary probability threshold is negatively correlated with the number of companions; a correction coefficient is obtained based on the second congestion level; the preliminary probability threshold is corrected using the correction coefficient to obtain the probability threshold; the correction coefficient is negatively correlated with the second congestion level.

[0048] In public transportation passenger re-identification scenarios, a single, fixed probability threshold is insufficient to adapt to varying sizes of passenger groups and complex, crowded environments, leading to an increased risk of misjudgment. When there are many passengers traveling together, the second passenger may need to interact with more fourth passengers (i.e., third passengers) within a limited time (i.e., the short period after disembarking). This results in the second passenger's interactions with their fellow passengers becoming more dispersed, meaning the intensity of their interactions with each of the fourth passengers is likely to decrease. In this case, using a higher probability threshold can easily miss genuine passenger relationships. Furthermore, in highly crowded scenarios, the probability of accidental interactions between the second passenger and other non-passenger groups also increases, further reducing the intensity of the second passenger's interactions with each of the fourth passengers.

[0049] To address the aforementioned situation, this invention employs a dynamic probability threshold determination mechanism. Specifically: First, the number of third passengers is calculated, which is used as the number of companions. Since a larger number of companions increases the probability that the interaction intensity between the second passenger and each of the fourth passengers will significantly decrease during the disembarkation period, a negative correlation is established between the number of companions and the initial probability threshold; that is, the larger the number of companions, the lower the initial probability threshold.

[0050] For example, when the number of companions is 2, the initial probability threshold is set to 0.9; if the number increases to 5, the initial probability threshold is lowered to 0.8 accordingly, thereby reducing the judgment standard in large group scenarios and avoiding missing real companion relationships.

[0051] Next, the initial probability threshold is adjusted based on the second level of crowding. The second level of crowding reflects the density of people in the disembarkation area. The higher the crowding, the higher the probability of accidental interactions between the second passenger and other non-passenger passengers, leading to a further reduction in the interaction intensity between the second passenger and each of the fourth passengers. Therefore, a correction coefficient is set to be negatively correlated with the second level of crowding. When the second level of crowding is high, the correction coefficient becomes smaller, resulting in a smaller adjustment to the initial probability threshold.

[0052] For example, in high-crowding scenarios, the correction coefficient is set to 0.9, further lowering the initial probability threshold from 0.8 to 0.72 to strictly distinguish between real peer relationships and accidental behavior; while in low-crowding scenarios, the correction coefficients are 1.0, 0.95, 0.96, etc., that is, reducing the reduction of the initial probability threshold or not adjusting the initial probability threshold at all, to ensure the accuracy and inclusiveness of the identification.

[0053] By combining the dynamic threshold determination method of the number of passengers traveling together and the congestion of the disembarkation area, the system can adapt to different public transport operation scenarios. In complex situations such as large groups traveling together and high congestion, it can effectively balance the accuracy and completeness of identification, reduce the probability of false positives and false negatives, and significantly improve the reliability and effectiveness of passenger re-identification based on travel companion relationships.

[0054] As an example, determining each first passenger with the intention to board the bus based on the first video image includes: acquiring the human posture of each passenger in the first video image, extracting passengers whose tilt angle towards the door is greater than an angle threshold and whose foot movement trajectory extends towards the door for N consecutive frames based on the human posture, and determining these passengers as each first passenger with the intention to board the bus; wherein, N is dynamically determined based on the peak period characteristics of bus operation.

[0055] In public transportation scenarios, traditional methods for determining boarding intent often rely on single visual features (such as a person facing the door), which are easily influenced by behaviors such as passengers pausing or lingering, leading to misjudgments. For example, a passenger asking for directions near the door might be misjudged as boarding.

[0056] To address this, the present invention utilizes human pose estimation technology to perform three-dimensional pose modeling of the passenger in the first video image, extracts the angle between the torso's central axis and the direction of the car door as the tilt angle, and simultaneously tracks the movement trajectory of the passenger's feet in consecutive video frames. When the following dual conditions are met, it is determined that the passenger has the intention to board the vehicle: (1) Angle threshold constraint: the tilt angle is greater than the angle threshold (e.g., 30°), indicating that the passenger is significantly facing the car door in spatial orientation; (2) Temporal continuity constraint: the foot movement trajectory stably points towards the car door area in consecutive N frames of images, avoiding misjudging the occasional behavior of turning towards the car door as the intention to board the vehicle.

[0057] In addition, the N value can be dynamically determined based on the peak-hour characteristics of bus operations. For example, during morning and evening peak hours, the N value can be appropriately increased (e.g., set to 150 frames) to improve judgment accuracy; during off-peak hours, the N value can be decreased (e.g., set to 100 frames or 50 frames) to improve response speed.

[0058] like Figure 2 As shown, this embodiment of the invention also provides a bus passenger re-identification control system 100, the system including a first capture unit 11, a second capture unit 12, a re-identification unit 13, and an association unit 14; wherein: the first capture unit 11 captures a first video image of the boarding line area, determines each first passenger with boarding intention based on the first video image, and determines and collects the first passenger features of each first passenger; the second capture unit 12 captures a second video image of the alighting line area, determines each second passenger with alighting intention based on the second video image, and collects the second passenger features of each second passenger; wherein, the first passenger features are determined based on the detection success rate of passenger features in the second video image; the re-identification unit 13 performs passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger belonging to the same passenger and the corresponding first passenger; the association unit 14 associates and records the boarding and alighting stations of the same passenger.

[0059] As an example, the first passenger feature is determined based on the success rate of passenger feature detection in the second video image, specifically including: determining the first congestion level of passengers in real time inside the bus, and the median value of the historical number of passengers getting off at subsequent bus stops; predicting the second congestion level of the disembarkation line area when the bus stops at subsequent bus stops based on the first congestion level and the median value; when the second congestion level is higher than the congestion level threshold, determining a height marker based on the shooting angle height of the second video image, and identifying each passenger feature on the passenger's body that is higher than the height marker as the first passenger feature.

[0060] As an example, the re-identification unit 13 is specifically used to: determine several third passengers who are companions of the first passenger based on the first video image; then perform passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger who belongs to the same passenger and the corresponding first passenger, including: performing passenger re-identification based on the second passenger features and the first passenger features to obtain a first probability that the second passenger and the corresponding first passenger belong to the same passenger; if the first probability is lower than a specified threshold and the second probability that the fourth passenger in the second video image and the third passenger belong to the same passenger is higher than a specified threshold, then: extract the conversation features and / or the assistance features of the second passenger and the fourth passenger, and evaluate the companion probability accordingly; when the companion probability is higher than the probability threshold, the second passenger and the corresponding first passenger are identified as belonging to the same passenger.

[0061] As an example, the probability threshold is determined by: calculating the number of third passengers and taking it as the number of companions; matching the number of companions to obtain a preliminary probability threshold; the preliminary probability threshold is negatively correlated with the number of companions; a correction coefficient is obtained based on the second congestion level; the preliminary probability threshold is corrected using the correction coefficient to obtain the probability threshold; the correction coefficient is negatively correlated with the second congestion level.

[0062] As an example, the first capture unit 11 is specifically used to: acquire the human posture of each passenger in the first video image, extract passengers whose tilt angle towards the door is greater than an angle threshold and whose foot movement trajectory extends towards the door for N consecutive frames based on the human posture, and identify these passengers as first passengers who intend to board the bus; wherein, N is dynamically determined according to the peak period characteristics of bus operation.

[0063] This invention also provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any of the foregoing embodiments.

[0064] This invention also provides a computer storage medium storing a computer program that can be executed by a processor to implement the methods described in any of the foregoing claims.

[0065] This invention also provides a computer program product comprising a computer program that can be executed by a processor to implement the method as described in any of the foregoing embodiments.

[0066] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0067] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for re-identifying passengers on a bus, characterized in that: The method includes the following steps: capturing a first video image of the boarding line area, determining each first passenger with boarding intention based on the first video image, and determining and collecting the first passenger features of each first passenger; A second video image of the drop-off area is captured. Based on the second video image, each second passenger with the intention to get off is identified, and the second passenger features of each second passenger are collected. The first passenger features are determined based on the detection success rate of passenger features in the second video image. Passenger re-identification is performed based on the second passenger features and the first passenger features to identify the second passenger belonging to the same passenger and the corresponding first passenger. The boarding station and drop-off station of the same passenger are associated and recorded. The first passenger feature is determined based on the success rate of passenger feature detection in the second video image. Specifically, it includes: determining the first level of passenger congestion in real time inside the bus and the median value of the historical number of passengers getting off at subsequent bus stops; predicting the second level of congestion in the disembarkation line area when the bus stops at subsequent bus stops based on the first level of congestion and the median value; when the second level of congestion is higher than the congestion threshold, determining a height marker based on the shooting angle height of the second video image, and identifying each passenger feature on the passenger's body that is higher than the height marker as the first passenger feature. The method further includes: determining several third passengers who are companions of the first passenger based on the first video image; then performing passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger who belongs to the same passenger and the corresponding first passenger, including: performing passenger re-identification based on the second passenger features and the first passenger features to obtain a first probability that the second passenger and the corresponding first passenger belong to the same passenger; if the first probability is lower than a specified threshold and the second probability that the fourth passenger in the second video image and the third passenger belong to the same passenger is higher than a specified threshold, then: extracting the conversation features and / or the assist features of the items carried by the second passenger and the fourth passenger, and evaluating the companion probability accordingly; when the companion probability is higher than the probability threshold, the second passenger and the corresponding first passenger are identified as belonging to the same passenger.

2. The bus passenger re-identification control method according to claim 1, characterized in that: The probability threshold is determined by calculating the number of third passengers and taking it as the number of companions, and matching the number of companions to obtain a preliminary probability threshold. The initial probability threshold is negatively correlated with the number of companions. The correction coefficient is obtained based on the second congestion matching, and the initial probability threshold is corrected using the correction coefficient to obtain the probability threshold; The correction coefficient is negatively correlated with the second congestion level.

3. The bus passenger re-identification control method according to claim 1, characterized in that: Determining each first passenger with the intention to board the bus based on the first video image includes: acquiring the human posture of each passenger in the first video image, extracting passengers whose tilt angle towards the door is greater than an angle threshold and whose foot movement trajectory extends towards the door for N consecutive frames based on the human posture, and identifying these passengers as each first passenger with the intention to board the bus; wherein, N is dynamically determined based on the peak period characteristics of bus operation.

4. A bus passenger re-identification control system, characterized in that, The system includes a first capture unit, a second capture unit, a re-identification unit, and an association unit; wherein: the first capture unit captures a first video image of the boarding area, determines each first passenger with boarding intent based on the first video image, and determines and collects the first passenger features of each first passenger; the second capture unit captures a second video image of the alighting area, determines each second passenger with alighting intent based on the second video image, and collects the second passenger features of each second passenger; wherein the first passenger features are determined based on the detection success rate of passenger features in the second video image; the re-identification unit performs passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger belonging to the same passenger and the corresponding first passenger; the association unit records the boarding and alighting stations of the same passenger in association. The first passenger feature is determined based on the success rate of passenger feature detection in the second video image. Specifically, it includes: determining the first level of passenger congestion in real time inside the bus and the median value of the historical number of passengers getting off at subsequent bus stops; predicting the second level of congestion in the disembarkation line area when the bus stops at subsequent bus stops based on the first level of congestion and the median value; when the second level of congestion is higher than the congestion threshold, determining a height marker based on the shooting angle height of the second video image, and identifying each passenger feature on the passenger's body that is higher than the height marker as the first passenger feature. The re-identification unit is specifically used for: determining several third passengers who are companions of the first passenger based on the first video image; then performing passenger re-identification based on the second passenger features and the first passenger features to identify the second passenger who belongs to the same passenger and the corresponding first passenger, including: performing passenger re-identification based on the second passenger features and the first passenger features to obtain a first probability that the second passenger and the corresponding first passenger belong to the same passenger; if the first probability is lower than a specified threshold and the second probability that the fourth passenger in the second video image belongs to the third passenger is higher than a specified threshold, then: extracting the conversation features and / or the assistive features of the second passenger and the fourth passenger, and evaluating the companion probability accordingly; when the companion probability is higher than the probability threshold, identifying the second passenger and the corresponding first passenger as belonging to the same passenger.

5. An electronic device, characterized in that: The electronic device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1-3.

6. A computer storage medium, characterized in that: The computer storage medium stores a computer program that can be executed by a processor to implement the method as described in any one of claims 1-3.

7. A computer program product, characterized in that: The computer program product includes a computer program that can be executed by a processor to implement the method as described in any one of claims 1-3.

Images