Bus passenger flow statistical method based on deep learning

Abstract

The invention provides a bus passenger flow statistical method based on deep learning, and belongs to the technical field of intelligent traffic, and the method comprises the steps: obtaining a bus construction map, deploying a detection device, and carrying out the multi-source data collection based on the detection device. Classifying the multi-source data, determining vehicle data and passenger data, and detecting getting-on and getting-off behaviors of passengers according to the vehicle data and the passenger data to obtain a behavior detection result; performing magnitude analysis on the behavior detection result to obtain a structured magnitude report, performing face recognition model parameter association on the structured magnitude report to obtain an association result, and configuring a face recognition model based on the association result; historical passenger flow, a vehicle Beidou track and weather data are fused based on a face recognition model, a prediction model is formed, the predicted number of changed people is obtained, the predicted number of changed people is pushed to a dispatching center to optimize vehicle deployment, data-driven dynamic vehicle deployment is realized, the perception accuracy is improved, intelligent prediction of passenger flow is realized, and resource allocation is optimized.

Description

Technical Field

[0001] The present invention relates to the field of smart transportation technology, and in particular to a bus passenger flow statistics method based on deep learning. Background Art

[0002] With the acceleration of urbanization, public transportation systems face three major challenges: large passenger flow fluctuations, delayed scheduling, and wasted resources. Traditional scheduling relies on experience, with manual passenger flow statistics having an error rate of up to 20-30%, and is unable to cope with sudden surges. Vehicle monitoring, passenger flow counting, and weather data are distributed and independent, lacking the ability to integrate and analyze multi-source information, resulting in significant response delays.

[0003] Therefore, the present invention provides a bus passenger flow statistics method based on deep learning. Summary of the Invention

[0004] The present invention provides a deep learning-based bus passenger flow statistics method. It uses multi-source sensors to collect vehicle and passenger data in real time, and combines it with a behavior detection algorithm to identify boarding and alighting actions. It dynamically adjusts the parameters of the face recognition model based on passenger flow density, and integrates historical passenger flow, Beidou trajectory and weather data to build a prediction model. Finally, the prediction results are pushed to the dispatch center to realize data-driven dynamic vehicle deployment, improve perception accuracy, achieve intelligent prediction of passenger flow, and optimize resource allocation.

[0005] The present invention provides a bus passenger flow statistics method based on deep learning, comprising: Step 1: Obtain a bus structure diagram, deploy a detection device based on the bus structure diagram, and perform multi-source data collection based on the detection device; Step 2: Classify the multi-source data to determine vehicle data and passenger data, and detect the passenger boarding and alighting behavior based on the vehicle data and passenger data to obtain a behavior detection result; Step 3: Performing magnitude analysis on the behavior detection results to obtain a structured magnitude report, correlating the structured magnitude report with face recognition model parameters to obtain a correlation result, and configuring the face recognition model based on the correlation result; Step 4: Based on the face recognition model, historical passenger flow, vehicle Beidou trajectory and weather data are integrated to form a prediction model, and the predicted number of changes is obtained. The predicted number of changes is pushed to the dispatch center to optimize vehicle allocation.

[0006] The present invention provides a bus passenger flow statistics method based on deep learning, which obtains a bus structure map, deploys a detection device based on the bus structure map, and performs multi-source data collection based on the detection device, including: Determining a multi-source sensor array based on the bus structural diagram, performing digital twin verification on the bus using the multi-source sensor array, and deriving an actual arrangement of the multi-source sensor array based on the verification results; Based on the actual arrangement, corresponding detection devices are deployed to collect multi-source data.

[0007] The present invention provides a bus passenger flow statistics method based on deep learning. The method determines a multi-source sensor array based on the bus structure diagram, uses the multi-source sensor array to perform digital twin verification on the bus, and obtains the actual layout of the multi-source sensor array based on the verification results, including: Parsing the bus structural diagram, extracting key dimensional parameters, marking the positions of existing equipment in the bus, and determining available routes based on the key dimensional parameters and the position marks; Dividing the carriage into a plurality of detection areas based on the available lines, configuring a primary-secondary sensor pair in each detection area to form a spatial dimension matrix, establishing a sensor capability complementation table based on the primary-secondary sensor pairs to form a functional dimension matrix, and combining the spatial dimension matrix with the functional dimension matrix to derive a multi-source sensor array; Perform digital twin verification of the bus based on the multi-source sensor array to generate basic performance verification results, extreme scenario verification results, and multimodal fusion verification results; The first arrangement is derived based on basic performance verification, the second arrangement is derived based on extreme scenario verification results, and the third arrangement is derived based on multimodal fusion verification results; The first arrangement, the second arrangement and the third arrangement are comprehensively analyzed to perform Pareto optimal calculation to obtain the actual arrangement of the multi-source sensor array.

[0008] The present invention provides a bus passenger flow statistics method based on deep learning, which classifies the multi-source data, determines vehicle data and passenger data, and detects the passengers' boarding and alighting behaviors based on the vehicle data and passenger data to obtain behavior detection results, including: Preprocess multi-source data, determine the data type of the preprocessing results, and determine the corresponding feature extraction method and output dimension from the type-feature engineering table based on the data type; Determining vehicle data and passenger data based on the feature extraction method and output dimensions, and then determining a vehicle data classification model, a passenger data classification model, and multimodal collaboration; The passenger boarding and alighting behavior is detected based on the vehicle data classification model, the passenger data classification model and multimodal collaboration, thereby obtaining a behavior detection result.

[0009] The present invention provides a bus passenger flow statistics method based on deep learning, which detects the passenger boarding and alighting behavior of passengers according to the vehicle data classification model, the passenger data classification model and multimodal collaboration, and then obtains the behavior detection results, including: Distinguishing vehicle features based on the vehicle data classification model, identifying passenger features based on the passenger data analysis model, and using the multimodal collaboration to formulate a feature fusion strategy corresponding to vehicle features and passenger features; The vehicle features and the passenger features are fused according to the feature fusion strategy to obtain a fusion result, and the fusion result is analyzed to detect the passenger getting on and off the vehicle behavior to obtain a behavior detection result.

[0010] The present invention provides a bus passenger flow statistics method based on deep learning, which performs magnitude analysis on behavior detection results to obtain a structured magnitude report, associates the structured magnitude report with face recognition model parameters to obtain an association result, and configures the face recognition model based on the association result, including: Aggregate behavior detection results in spatiotemporal dimensions, count the number of passengers getting on and off the bus within a unit time window, the distribution of behavior types, and regional density, divide the magnitude levels based on preset thresholds, and output a structured magnitude report; Magnitude features are extracted based on structured magnitude reports, and the magnitude features are associated with key parameters of the face recognition model to obtain association results. Hierarchical configuration is performed based on the association results, and the face recognition model is obtained by integrating all hierarchical configurations.

[0011] The present invention provides a bus passenger flow statistics method based on deep learning. Based on a face recognition model, it integrates historical passenger flow, vehicle Beidou trajectory and weather data to form a prediction model, derive the predicted number of changes, and push the predicted number of changes to the dispatch center to optimize vehicle allocation, including: Obtain real-time passenger flow from the facial recognition system, access the historical passenger flow database to extract historical passenger flow, simultaneously collect Beidou vehicle trajectories, and obtain weather data through the meteorological interface; Decompose the time dimension into periodic time features and obtain the topological features of the site at the same time; Extracting common time series feature vectors and key time series feature vectors from real-time passenger flow and historical passenger flow based on periodic time features, extracting static features from weather data, performing a first concatenation on the common time series feature vectors and the static features, and performing a second concatenation on the key time series feature vectors and the static features; The first splicing and the second splicing are connected based on the topological features to form a prediction model, and a predicted number of changes is obtained, which is then pushed to a dispatch center to optimize vehicle allocation.

[0012] The present invention provides a bus passenger flow statistics method based on deep learning, which pushes the predicted number of passengers to the dispatch center to optimize vehicle allocation, including: Match key causal chains based on the predicted number of changes, and trigger resource allocation plans based on the key causal chains; A sandbox simulation is conducted based on the resource allocation plan, and the plan evaluation results are output. The dispatch center determines the mobilization instructions based on the plan evaluation results to optimize vehicle allocation.

[0013] Compared with the existing technology, the beneficial effects of this application are as follows: real-time collection of vehicle and passenger data through multi-source sensors, combined with behavior detection algorithms to identify boarding and alighting actions; dynamic adjustment of face recognition model parameters based on passenger flow density, integration of historical passenger flow, Beidou trajectory and weather data to build a prediction model; and finally pushing the prediction results to the dispatching center to realize data-driven dynamic vehicle deployment, improve the accuracy of perception, realize intelligent prediction of passenger flow, and optimize resource allocation.

[0014] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a bus passenger flow statistics method based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention. Embodiment 1:

[0018] The embodiment of the present invention provides a bus passenger flow statistics method based on deep learning, such as Figure 1 As shown, including: Step 1: Obtain a bus structure diagram, deploy a detection device based on the bus structure diagram, and perform multi-source data collection based on the detection device; Step 2: Classify the multi-source data to determine vehicle data and passenger data, and detect the passenger boarding and alighting behavior based on the vehicle data and passenger data to obtain a behavior detection result; Step 3: Performing magnitude analysis on the behavior detection results to obtain a structured magnitude report, correlating the structured magnitude report with face recognition model parameters to obtain a correlation result, and configuring the face recognition model based on the correlation result; Step 4: Based on the face recognition model, historical passenger flow, vehicle Beidou trajectory and weather data are integrated to form a prediction model, and the predicted number of changes is obtained. The predicted number of changes is pushed to the dispatch center to optimize vehicle allocation.

[0019] In this embodiment, the bus structural drawing is a CAD design drawing or a three-dimensional model of the bus, which includes engineering data such as the car structure and equipment layout, for example, the door position and step height marking of a low-floor bus, and the stair position and upper-level seat distribution map of a double-decker bus.

[0020] In this embodiment, the detection device is a deployed physical sensor and supporting equipment, such as the Hikvision DS-2CD3 series vehicle-mounted camera.

[0021] In this embodiment, the initial layout of multi-source sensors (such as cameras, radars, and pressure sensors) is designed based on the bus structure diagram. Real operating scenarios (such as passenger flow, occlusion, and extreme weather) are simulated in a digital twin environment to verify sensor performance. The installation location and parameters are optimized based on the simulation results, and ultimately physical deployment is guided to ensure data collection coverage without blind spots, achieve precise deployment, and improve robustness.

[0022] In this embodiment, the vehicle data refers to data related to the vehicle, such as the speed, acceleration, position, etc. of the vehicle, for example, GPS position data and speedometer data of the vehicle.

[0023] In this embodiment, passenger data refers to data related to the passenger, such as the passenger's age, gender, behavior, etc., for example, sensor data of the passenger passing through the door and the passenger image captured by the camera.

[0024] In this embodiment, the behavior detection result refers to the passenger behavior judgment derived from the fusion result analysis. This is typically based on the output of a machine learning model or rule engine. For example, if the feature fusion result for passenger A indicates that the vehicle is decelerating and the passenger is standing near the door, the behavior detection result indicates that passenger A is preparing to get off the vehicle. If the feature fusion result for passenger B indicates that the vehicle is accelerating and the passenger has just passed through the door, the behavior detection result indicates that passenger B has just boarded the vehicle.

[0025] In this embodiment, multi-source heterogeneous data are processed through intelligent classification and feature engineering, and classification models for vehicle and passenger data are constructed respectively. The multimodal collaborative algorithm is combined with spatiotemporal features to achieve high-precision boarding and alighting behavior detection. The system automatically matches data types and feature extraction methods, and dynamically optimizes the detection process.

[0026] In this embodiment, the structured magnitude report includes density level, behavior distribution characteristics and abnormal markers, which are organized and standardized statistical results, usually in JSON or database records. For example, the report includes the time period 09:00-10:00, area A1 exit, 200 boarding passengers, and high density level.

[0027] In this embodiment, the association result is a matching rule between features and parameters. For example, when "red density + secondary behavior warning" appear at the same time: the face capture frame rate is increased from 15fps to 30fps, and the similarity threshold is raised from 0.85 to 0.92.

[0028] In this embodiment, the facial recognition model is a complete recognition system that is dynamically adjusted. For example, during the Spring Festival travel rush, the train station system automatically enables "peak mode" from 6:00 to 9:00, reducing resolution but doubling the analysis frequency. When a fugitive is identified, it instantly switches to "precision mode": the highest definition image quality + strict comparison.

[0029] In this embodiment, passenger behavior data is analyzed through spatiotemporal aggregation to generate a structured magnitude report, and facial recognition model parameters (such as detection frame rate and ROI area) are dynamically adjusted based on passenger flow density to achieve closed-loop optimization of "passenger flow perception-model adaptation" and ensure optimal recognition performance in different scenarios.

[0030] In this embodiment, the historical passenger flow is the passenger flow record of the same time period in the past (such as the same day of the same period last year / last week). For example, the total passenger flow of the station on the first day of National Day last year was 100,000.

[0031] In this embodiment, the vehicle Beidou trajectory is the real-time position and movement path of the bus / shuttle bus obtained through Beidou satellite positioning. For example, the trajectory point sequence of bus A is: [(longitude X1, latitude Y1, time T1), (X2, Y2, T2)...].

[0032] In this embodiment, the weather data is real-time weather information (such as temperature, rainfall, and wind speed) obtained through an API. For example, the current weather data is: {"temperature": 28°C, "rainfall": 5mm, "wind speed": level 3}.

[0033] In this embodiment, the input of the prediction model is the spliced ​​comprehensive features, and the output is the passenger flow forecast value for the future time period. For example, the input is: ["morning rush hour", "transfer station", 500, "+50%", "heavy rain"], and the output is: 650 people are predicted to enter the station in the next hour (an increase of 30% over the usual number).

[0034] In this implementation, a hybrid prediction model with enhanced spatiotemporal features is constructed by integrating real-time facial recognition passenger flow, historical passenger flow patterns, Beidou vehicle trajectories, and weather data. This model combines periodic passenger flow trends with station topology to dynamically generate passenger flow forecasts, driving intelligent scheduling decisions and enabling data-driven, precise capacity allocation. Subsequently, resource allocation plans are dynamically generated by analyzing passenger flow forecast data and key causal chains (e.g., weather, passenger flow, and resource demand). After verification through digital twin simulations, optimal scheduling instructions are output, achieving closed-loop optimization from prediction to execution.

[0035] The working principle and beneficial effects of the above technical solution are: real-time collection of vehicle and passenger data through multi-source sensors, combined with behavior detection algorithms to identify boarding and alighting actions; dynamic adjustment of facial recognition model parameters based on passenger flow density, integration of historical passenger flow, Beidou trajectory and weather data to build a prediction model; and finally pushing the prediction results to the dispatch center to realize data-driven dynamic vehicle deployment, improve perception accuracy, achieve intelligent prediction of passenger flow, and optimize resource allocation. Example 2:

[0036] An embodiment of the present invention provides a bus passenger flow statistics method based on deep learning, which obtains a bus structure map, deploys a detection device based on the bus structure map, and performs multi-source data collection based on the detection device, including: Determining a multi-source sensor array based on the bus structural diagram, performing digital twin verification on the bus using the multi-source sensor array, and deriving an actual arrangement of the multi-source sensor array based on the verification results; Based on the actual arrangement, corresponding detection devices are deployed to collect multi-source data.

[0037] In this embodiment, the multi-source sensor array is a combination of heterogeneous sensors deployed in the vehicle cabin to collect multi-dimensional data. For example, the visual sensor is a 1080P binocular camera on the top of the door (to detect passenger flow), the radar sensor is a 60GHz millimeter-wave radar on the side wall of the vehicle cabin (to penetrate obstructions for detection), and the environmental sensor is a temperature, humidity, and light intensity composite probe near the cab.

[0038] In this embodiment, digital twin verification is a testing process that simulates sensor performance in a virtual environment. For example, Unity3D simulates the camera recognition rate in heavy rain weather, and ANSYS simulates the impact of electromagnetic interference on radar point clouds.

[0039] In this embodiment, the verification results are quantitative performance indicators output by the digital twin test, for example, the camera's missed detection rate in backlit scenes is ≤3%, and the radar's multi-target resolution accuracy in crowded conditions is ±5cm.

[0040] In this embodiment, the actual layout is an optimized physical installation scheme of the sensors. For example, the front door camera is installed at a height of 2.4 meters from the ground with a depression angle of 12°. The pressure sensor is arranged at three equally spaced measurement points under the step.

[0041] The working principle and beneficial effects of the above technical solution are as follows: the initial layout of multi-source sensors (such as cameras, radars, and pressure sensors) is designed through the bus structure diagram, and real operating scenarios (such as passenger flow, occlusion, and extreme weather) are simulated in a digital twin environment to verify sensor performance; the installation location and parameters are optimized based on the simulation results, and ultimately physical deployment is guided to ensure data collection coverage without blind spots, achieve precise deployment, and improve robustness. Example 3:

[0042] An embodiment of the present invention provides a bus passenger flow statistics method based on deep learning. The method determines a multi-source sensor array based on a bus structural diagram, performs digital twin verification of the bus using the multi-source sensor array, and obtains the actual layout of the multi-source sensor array based on the verification results, including: Parsing the bus structural diagram, extracting key dimensional parameters, marking the positions of existing equipment in the bus, and determining available routes based on the key dimensional parameters and the position marks; Dividing the carriage into a plurality of detection areas based on the available lines, configuring a primary-secondary sensor pair in each detection area to form a spatial dimension matrix, establishing a sensor capability complementation table based on the primary-secondary sensor pairs to form a functional dimension matrix, and combining the spatial dimension matrix with the functional dimension matrix to derive a multi-source sensor array; Perform digital twin verification of the bus based on the multi-source sensor array to generate basic performance verification results, extreme scenario verification results, and multimodal fusion verification results; The first arrangement is derived based on basic performance verification, the second arrangement is derived based on extreme scenario verification results, and the third arrangement is derived based on multimodal fusion verification results; The first arrangement, the second arrangement and the third arrangement are comprehensively analyzed to perform Pareto optimal calculation to obtain the actual arrangement of the multi-source sensor array.

[0043] In this example, key dimensional parameters refer to dimensions in the bus's structural drawings that are crucial for design and analysis, such as the length, width, and height of the bus compartment, seat dimensions, and aisle width. These parameters are crucial for determining the location and layout of the bus's internal equipment, for example, the length (10 meters), width (2.5 meters), and height (3 meters).

[0044] In this embodiment, the parsing process refers to the action of extracting these key dimensional parameters from the bus structural drawing, for example, using CAD software to measure and record the various dimensions of the bus compartment.

[0045] In this embodiment, the available lines refer to lines inside the bus that can be used to install sensors or other devices, determined based on device locations and key dimensional parameters, such as aisles on the floor of the bus and spaces under seats.

[0046] In this embodiment, position marking refers to marking the positions of existing equipment in the bus (such as seats, handrails, emergency exits, etc.) so that these positions can be taken into account in subsequent designs. For example, a point is marked for each seat position on the drawing.

[0047] In this embodiment, the process of dividing the detection area refers to dividing the carriage into multiple areas based on available lines to facilitate the arrangement of sensors. For example, the carriage is divided into three detection areas: front, middle and rear.

[0048] In this embodiment, the primary-auxiliary sensor pair refers to a primary sensor and an auxiliary sensor configured in the detection area. The primary sensor is usually responsible for the main detection task, and the auxiliary sensor provides supplementary information. For example, the primary sensor is a temperature sensor and the auxiliary sensor is a humidity sensor.

[0049] In this embodiment, the spatial dimension matrix refers to a matrix representing the distribution of sensors in space, which helps determine the positions and coverage of the sensors. For example, a 3x3 matrix represents the positions of 9 sensors in the vehicle compartment.

[0050] In this embodiment, the sensor capability complement table refers to a table showing the complementary relationship between the capabilities of different sensors, and is used to optimize sensor configuration. For example, the table shows the detection range and accuracy of the temperature sensor and the humidity sensor.

[0051] In this embodiment, the functional dimension matrix refers to a matrix representing the functional characteristics of the sensor, combined with the spatial dimension matrix, and is used to evaluate the functionality of the entire sensor network, for example, a matrix showing the environmental parameters that each sensor can detect.

[0052] In this embodiment, the multi-source sensor array refers to an array composed of multiple sensors of different types, which can work together to provide comprehensive monitoring data, for example, an array composed of temperature, humidity, and carbon dioxide sensors.

[0053] In this embodiment, the basic performance verification result is to verify the performance of the sensor under normal conditions, such as the accuracy of the temperature sensor at 25°C; the extreme scenario verification result is to verify the performance of the sensor under extreme conditions, such as the accuracy of the temperature sensor at -10°C or 50°C; the multimodal fusion verification result is to verify the performance after the fusion of multiple sensor data, such as combining temperature and humidity data to predict the comfort level in the vehicle cabin.

[0054] In this embodiment, the first arrangement is a sensor arrangement based on basic performance verification results; the second arrangement is a sensor arrangement based on extreme scenario verification results; and the third arrangement is a sensor arrangement based on multimodal fusion verification results.

[0055] In this embodiment, Pareto optimality refers to the configuration in which the performance of a particular sensor cannot be improved without compromising the performance of other sensors. This is an optimization method used to find the best sensor placement, for example, minimizing the number of sensors while ensuring sufficient sensor coverage across all detection areas.

[0056] In this embodiment, the actual arrangement refers to the arrangement result of the optimal sensor arrangement scheme derived from the Pareto optimality calculation. For example, 20 sensors, including temperature, humidity, and carbon dioxide sensors, are installed in the carriage to achieve the best monitoring effect. ,in, represents the optimal sensor layout scheme finally selected; represents the weight of the i-th target; Represents the value of the i-th objective function under the sensor layout scheme x represents the maximum value of target i in the frontier; represents the minimum value of target i in the frontier; γ represents the penalty coefficient of the implementation function; C(x) represents the total implementation function of the sensor layout scheme x; x represents the candidate sensor layout scheme; P represents the Pareto front solution set; i represents the index of the objective function.

[0057] The working principle and beneficial effects of the above technical solution are: by analyzing the bus structure diagram to divide the detection area, designing a primary-auxiliary sensor collaborative network, verifying basic performance, extreme scenarios and multimodal fusion effects in the digital twin, and finally combining the three verification results based on the Pareto optimal algorithm to output the global optimal sensor layout plan, achieve accurate sensor coverage, and improve the adaptability of bus information collection. Embodiment 4:

[0058] An embodiment of the present invention provides a bus passenger flow statistics method based on deep learning, which classifies the multi-source data, determines vehicle data and passenger data, and detects the boarding and alighting behavior of passengers based on the vehicle data and passenger data to obtain behavior detection results, including: Preprocess multi-source data, determine the data type of the preprocessing results, and determine the corresponding feature extraction method and output dimension from the type-feature engineering table based on the data type; Determining vehicle data and passenger data based on the feature extraction method and output dimensions, and then determining a vehicle data classification model, a passenger data classification model, and multimodal collaboration; The passenger boarding and alighting behavior is detected based on the vehicle data classification model, the passenger data classification model and multimodal collaboration, thereby obtaining a behavior detection result.

[0059] In this embodiment, the preprocessing result refers to the data after the original data is cleaned, formatted, and standardized, for example, the timestamp format is unified, missing values ​​are filled, and outliers are removed.

[0060] In this embodiment, the type-feature engineering table is a mapping table that maps different data types to corresponding feature extraction methods and output dimensions. It is constructed based on domain knowledge and data characteristics. For example, a table lists data types (such as text, numbers, time series, etc.), corresponding feature extraction methods (such as bag-of-words model, normalization, Fourier transform, etc.), and output dimensions (such as 100-dimensional vector, 20-dimensional vector, etc.).

[0061] In this embodiment, the data type refers to the type of data, such as text, numerical data, image, time series, etc. Determining the data type is a prerequisite for selecting an appropriate feature extraction method, for example, numerical data, text data.

[0062] In this embodiment, the feature extraction method refers to the process of converting raw data into a format acceptable to the model, which involves extracting key information from the data and representing it in the form of a vector. For example, for text data, a bag-of-words model can be used to extract features; for numerical data, normalization or standardization can be used.

[0063] In this embodiment, the output dimension refers to the dimension of the feature vector obtained after feature extraction. This dimension determines the size of the model input, for example, a 100-dimensional vector.

[0064] In this embodiment, the input and output of the vehicle data classification model refer to the format of the vehicle data used by the model during training and prediction and the results of the model output. For example, the input is the acceleration and speed of the vehicle, and the output is whether the vehicle is braking.

[0065] In this embodiment, the input and output of the passenger data classification model refer to the format of the passenger data used by the model during training and prediction and the results of the model output. For example, the input is the passenger's image and the output is whether the passenger is wearing a mask.

[0066] In this embodiment, multimodal collaborative classification refers to combining data from different modalities (such as text, images, and audio) for classification. This approach can improve classification accuracy and robustness. For example, combining vehicle GPS location data and passenger image data can be used to detect passenger boarding and alighting behaviors.

[0067] In this embodiment, key features are extracted through vehicle and passenger classification models respectively, and a multimodal collaborative algorithm (such as the attention mechanism) is used to dynamically fuse vehicle status and passenger behavior characteristics. Combined with spatiotemporal correlation analysis, high-precision boarding and alighting behavior detection is achieved, forming a closed-loop analysis chain of "data perception-feature fusion-behavior judgment".

[0068] The working principle and beneficial effects of the above technical solution are: through intelligent classification and feature engineering to process multi-source heterogeneous data, classification models for vehicle and passenger data are constructed separately, and multimodal collaborative algorithms are combined to fuse spatiotemporal features to achieve high-precision boarding and alighting behavior detection. The system automatically matches data types and feature extraction methods to dynamically optimize the detection process. Example 5:

[0069] The embodiment of the present invention provides a bus passenger flow statistics method based on deep learning. The method detects the boarding and alighting behavior of passengers based on the vehicle data classification model, the passenger data classification model, and multimodal collaboration, and then obtains the behavior detection results, including: Distinguishing vehicle features based on the vehicle data classification model, identifying passenger features based on the passenger data analysis model, and using the multimodal collaboration to formulate a feature fusion strategy corresponding to vehicle features and passenger features; The vehicle features and the passenger features are fused according to the feature fusion strategy to obtain a fusion result, and the fusion result is analyzed to detect the passenger getting on and off the vehicle behavior to obtain a behavior detection result.

[0070] In this embodiment, vehicle features refer to attributes extracted from vehicle data that describe the vehicle's state or behavior. These features can be the vehicle's physical characteristics, operating status, or other related information, such as the vehicle's current speed, acceleration, braking status, steering angle, etc.

[0071] In this embodiment, passenger characteristics refer to information extracted from passenger data to describe passenger attributes. These characteristics can be the passenger's physiological characteristics, behavioral habits, or other relevant information, such as the passenger's age, gender, height, whether he or she is carrying a bag, etc.

[0072] In this embodiment, the feature fusion strategy refers to how to combine features from different data sources into a comprehensive feature representation. This typically involves selecting which features to use, how to handle the correlation between different features, and how to fuse them. For example, combining the vehicle's acceleration and speed features with the passenger's body posture and position features to predict whether the passenger is ready to get on or off the vehicle.

[0073] In this embodiment, the fusion result refers to the result obtained by combining vehicle features and passenger features according to the feature fusion strategy. This result is typically a dataset containing more information, which can be used for more complex analysis and modeling. For example, a comprehensive feature vector containing vehicle speed, acceleration, and passenger posture and position can be generated.

[0074] The working principle and beneficial effects of the above technical solution are: extracting key features through vehicle and passenger classification models respectively, using multimodal collaborative algorithms (such as attention mechanism) to dynamically fuse vehicle status and passenger behavior characteristics, combining spatiotemporal correlation analysis to achieve high-precision boarding and alighting behavior detection, forming a closed-loop analysis chain of "data perception-feature fusion-behavior judgment", and realizing accurate detection and dynamic adaptation. Example 6:

[0075] An embodiment of the present invention provides a bus passenger flow statistics method based on deep learning. The method performs magnitude analysis on behavior detection results to obtain a structured magnitude report, associates the structured magnitude report with face recognition model parameters to obtain an association result, and configures the face recognition model based on the association result, including: Aggregate behavior detection results in spatiotemporal dimensions, count the number of passengers getting on and off the bus within a unit time window, the distribution of behavior types, and regional density, divide the magnitude levels based on preset thresholds, and output a structured magnitude report; Magnitude features are extracted based on structured magnitude reports, and the magnitude features are associated with key parameters of the face recognition model to obtain association results. Hierarchical configuration is performed based on the association results, and the face recognition model is obtained by integrating all hierarchical configurations.

[0076] In this embodiment, spatiotemporal aggregation is to divide and count behavioral data by both time segments and spatial regions. For example, a subway station divides the monitoring area into three spatial units: "gate", "platform", and "channel". With a time window of 15 minutes, data for each area from 8:00 to 8:15 are counted.

[0077] In this embodiment, the number of passengers getting on and off is the number of passengers entering and leaving a specific area during a fixed time period. For example, during the period from 7:30 to 8:00 at the east gate of the bus station, there are 85 boarding card swiping records and 72 getting off code scanning records.

[0078] In this embodiment, the behavior type distribution is the proportion of each type of behavior in the total detection volume. For example, the waiting area monitoring analysis shows: standing waiting 68%, walking 25%, luggage left behind 3%, and physical conflict 4%.

[0079] In this embodiment, the regional density is the real-time number of people per unit area. For example, the infrared sensor in the middle of the carriage detects 2.3 people / m2, which exceeds the warning threshold of 2.0 people / m2.

[0080] In this embodiment, the preset thresholds and magnitude levels are pre-set critical values ​​for classification, for example, density classification: below 1.5 people / ㎡ = green, 1.5-2.5 people / ㎡ = yellow, and above 2.5 people / ㎡ = red; abnormal behavior classification: 3 falls / hour = level 1 warning, 5 falls / hour = level 2 emergency.

[0081] In this embodiment, the structured magnitude report is a statistical analysis summary in a standardized format, for example, "2023-12-01 08:15:00 Report: Area: Middle of Platform 2, Density Level: Red (2.8 people / ㎡), Behavior Alert: Fall Behavior Triggered (3 times within 5 minutes), Recommended Measures: Activate the flow control plan."

[0082] In this embodiment, magnitude feature extraction is to extract key decision indicators from the report. For example, the core feature is extracted: "continuous high density + frequent abnormal behavior" and the related factors are: morning rush hour + rainy and snowy weather.

[0083] In this embodiment, the key parameters of face recognition are core adjustable parameters that affect the recognition effect, such as image acquisition: resolution (1080P / 720P), analysis speed: detection frame rate (10fps / 25fps), and comparison accuracy: similarity threshold (0.8 / 0.9).

[0084] In this embodiment, the hierarchical configuration is a multi-level parameter combination scheme, for example, basic mode: sunny and weekdays, 1080P@15fps, enhanced mode: rainy and snowy weather, 720P@20fps, emergency mode: large passenger flow + events, 480P@30fps.

[0085] The working principle and beneficial effects of the above technical solution are: through spatiotemporal aggregation analysis of passenger behavior data, generating structured magnitude reports, and dynamically adjusting facial recognition model parameters (such as detection frame rate and ROI area) based on passenger flow density, to achieve closed-loop optimization of "passenger flow perception-model adaptation" and ensure optimal recognition performance in different scenarios. Example 7:

[0086] The present invention provides a deep learning-based bus passenger flow statistics method. This method integrates historical passenger flow, vehicle Beidou trajectory, and weather data based on a face recognition model to form a prediction model, derive a predicted passenger flow change, and pushes the predicted passenger flow change to a dispatch center to optimize vehicle allocation. The method includes: Obtain real-time passenger flow from the facial recognition system, access the historical passenger flow database to extract historical passenger flow, simultaneously collect Beidou vehicle trajectories, and obtain weather data through the meteorological interface; Decompose the time dimension into periodic time features and obtain the topological features of the site at the same time; Extracting common time series feature vectors and key time series feature vectors from real-time passenger flow and historical passenger flow based on periodic time features, extracting static features from weather data, performing a first concatenation on the common time series feature vectors and the static features, and performing a second concatenation on the key time series feature vectors and the static features; The first splicing and the second splicing are connected based on the topological features to form a prediction model, and a predicted number of changes is obtained, which is then pushed to a dispatch center to optimize vehicle allocation.

[0087] In this embodiment, the real-time passenger flow is the current passenger flow counted in real time by the face recognition system. For example, the gate camera of the subway station detects 200 people entering the station and 150 people leaving the station in the current 5 minutes.

[0088] In this embodiment, the historical passenger flow database is a structured database that stores historical passenger flow data, for example, a station passenger flow table stored hourly in a MySQL database, including fields such as "date, time period, and number of people entering and exiting".

[0089] In this embodiment, the periodic time feature is a periodic pattern extracted from time (such as hour / week / season), for example, the morning rush hour (7:00-9:00), weekend, and holiday signs.

[0090] In this embodiment, the topological features are features that describe the spatial relationship between stations (such as adjacent stations and hub levels). For example, station B is a transfer station for three subway lines, and its adjacent stations are C and D.

[0091] In this embodiment, the common time series feature vector is a regular time series statistical value (such as the mean value and the number of people in a sliding window). For example, the average passenger flow in the past hour is 500 people, which is a 10% increase from the previous month.

[0092] In this embodiment, the key time series feature vector is a key time series indicator (such as a peak value, a mutation point) that affects the passenger flow. For example, the passenger flow in the same period yesterday suddenly increased by 50% (due to an emergency).

[0093] In this embodiment, the static features are weather / environmental attributes that remain unchanged in the short term, such as the current weather label "heavy rain" and the temperature "30°C".

[0094] In this embodiment, the first splicing and the second splicing are to combine different feature vectors by dimension. For example, the first splicing is: common time series features (mean 500 people) + static features (heavy rain) → [500, heavy rain], and the second splicing is: key time series features (sudden change +50%) + static features (heavy rain) → [+50%, heavy rain].

[0095] In this embodiment, the predicted number of changed passengers is the passenger flow change (absolute number or percentage) predicted by the model. For example, it is predicted that the passenger flow from 15:00 to 16:00 will increase by 200 people compared to normal days.

[0096] The working principle and beneficial effects of this technical solution are as follows: By integrating real-time facial recognition passenger flow, historical passenger flow patterns, Beidou vehicle trajectory data, and weather data, a hybrid prediction model with enhanced spatiotemporal features is constructed. This model combines periodic passenger flow trends with station topology to dynamically generate passenger flow forecasts, driving intelligent scheduling decisions and achieving data-driven, precise capacity allocation. Example 8:

[0097] The present invention provides a deep learning-based bus passenger flow statistics method that pushes the predicted passenger flow changes to a dispatch center to optimize vehicle allocation, including: Match key causal chains based on the predicted number of changes, and trigger resource allocation plans based on the key causal chains; A sandbox simulation is conducted based on the resource allocation plan, and the plan evaluation results are output. The dispatch center determines the mobilization instructions based on the plan evaluation results to optimize vehicle allocation.

[0098] In this embodiment, the key causal chain is the core factor that affects passenger flow changes and its logical relationship chain. For example, heavy rain (cause) → increased demand for taxis (effect) → the number of people entering the subway station decreased by 20% (final effect); the concert ended (cause) → 500 people gathered at the bus station (effect) → 3 additional emergency vehicles were needed (final effect).

[0099] In this embodiment, the resource allocation plan is a specific resource adjustment plan formulated in response to the predicted change. For example, when the predicted passenger flow increases by 200 people: add two spare buses, extend the last bus by 30 minutes, and open an emergency channel to divert passengers.

[0100] In this embodiment, the sandbox simulation is to simulate the effect of the plan execution in a virtual environment. For example, the input conditions are: it is predicted that the passenger flow will increase by 300 people during the evening peak hour, and 5 spare vehicles are available. The simulation process is: after simulating the dispatch of 3 more vehicles: the waiting time is reduced from 15 minutes to 8 minutes, and the platform congestion is reduced from 90% to 60%.

[0101] In this embodiment, the scheme evaluation result is a quantitative score of each scheme after deduction. For example, Scheme A (adding 3 vehicles): cost: 5,000 yuan, efficiency improvement: waiting time reduced by 47%, comprehensive score: 85 points; Scheme B (adding 2 vehicles + diversion): cost: 3,000 yuan, efficiency improvement: waiting time reduced by 35%, comprehensive score: 78 points.

[0102] In this embodiment, the dispatch center is the central control unit for decision-making on resource allocation. For example, the large screen of a city's traffic command center displays: real-time monitoring of passenger flow at 12 hub stations, receiving the top 3 allocation plans recommended by the system, and manual final confirmation of the execution of plan B.

[0103] In this embodiment, the mobilization instruction is an operational command issued to the execution unit, for example, the instruction content is: "Dispatch vehicle Beijing B12345 to Guomao Station before 18:00", "Activate Emergency Channel No. 2 of Xidan Station", and the execution feedback is: the vehicle has arrived and the GPS has confirmed that the channel monitoring shows the open status.

[0104] In this embodiment, vehicle deployment is the adjustment of transport resources implemented according to instructions. For example, in a regular scenario, the frequency of departures is increased by 10% during peak hours in the morning and evening. In an emergency scenario, 20 reserve buses are urgently dispatched from the garage in the event of an emergency.

[0105] The working principle and beneficial effects of the above technical solution are: by analyzing passenger flow forecast data and key causal chains (such as weather-passenger flow-resource demand), a resource allocation plan is dynamically generated. After the effect is verified through digital twin sandbox simulation, the optimal scheduling instructions are output to achieve closed-loop optimization from prediction to execution.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A bus passenger flow statistics method based on deep learning, characterized in that: include: Step 1: Obtain a bus structure diagram, deploy a detection device based on the bus structure diagram, and perform multi-source data collection based on the detection device; Step 2: Classify the multi-source data to determine vehicle data and passenger data, and detect the passenger boarding and alighting behavior based on the vehicle data and passenger data to obtain a behavior detection result; Step 3: Performing magnitude analysis on the behavior detection results to obtain a structured magnitude report, correlating the structured magnitude report with face recognition model parameters to obtain a correlation result, and configuring the face recognition model based on the correlation result; Step 4: Based on the face recognition model, historical passenger flow, vehicle Beidou trajectory and weather data are integrated to form a prediction model, and the predicted number of changes is obtained. The predicted number of changes is pushed to the dispatch center to optimize vehicle allocation.

2. The bus passenger flow statistics method based on deep learning according to claim 1 is characterized in that: Obtaining a bus structural diagram, deploying a detection device based on the bus structural diagram, and performing multi-source data collection based on the detection device, including: Determining a multi-source sensor array based on the bus structural diagram, performing digital twin verification on the bus using the multi-source sensor array, and deriving an actual arrangement of the multi-source sensor array based on the verification results; Based on the actual arrangement, corresponding detection devices are deployed to collect multi-source data.

3. The bus passenger flow statistics method based on deep learning according to claim 2 is characterized in that: Determining a multi-source sensor array based on the bus structural diagram, performing digital twin verification on the bus using the multi-source sensor array, and obtaining an actual arrangement of the multi-source sensor array based on the verification results, including: Parsing the bus structural diagram, extracting key dimensional parameters, marking the positions of existing equipment in the bus, and determining available routes based on the key dimensional parameters and the position marks; Dividing the carriage into a plurality of detection areas based on the available lines, configuring each detection area with a primary-secondary sensor pair to form a spatial dimension matrix, establishing a sensor capability complementation table based on the primary-secondary sensor pairs to form a functional dimension matrix, and combining the spatial dimension matrix with the functional dimension matrix to derive a multi-source sensor array; Perform digital twin verification of the bus based on the multi-source sensor array to generate basic performance verification results, extreme scenario verification results, and multimodal fusion verification results; The first arrangement is derived based on basic performance verification, the second arrangement is derived based on extreme scenario verification results, and the third arrangement is derived based on multimodal fusion verification results; The first arrangement, the second arrangement and the third arrangement are comprehensively analyzed to perform Pareto optimal calculation to obtain the actual arrangement of the multi-source sensor array.

4. The bus passenger flow statistics method based on deep learning according to claim 1 is characterized in that: Classifying the multi-source data to determine vehicle data and passenger data, performing boarding and alighting behavior detection on passengers based on the vehicle data and passenger data, and obtaining behavior detection results, including: Preprocess multi-source data, determine the data type of the preprocessing results, and determine the corresponding feature extraction method and output dimension from the type-feature engineering table based on the data type; Determining vehicle data and passenger data based on the feature extraction method and output dimensions, and then determining a vehicle data classification model, a passenger data classification model, and multimodal collaboration; The passenger boarding and alighting behavior is detected based on the vehicle data classification model, the passenger data classification model and multimodal collaboration, thereby obtaining a behavior detection result.

5. The bus passenger flow statistics method based on deep learning according to claim 4 is characterized in that: The passenger boarding and alighting behavior is detected based on the vehicle data classification model, the passenger data classification model, and multimodal collaboration, thereby obtaining behavior detection results, including: Distinguishing vehicle features based on the vehicle data classification model, identifying passenger features based on the passenger data analysis model, and using the multimodal collaboration to formulate a feature fusion strategy corresponding to vehicle features and passenger features; The vehicle features and the passenger features are fused according to the feature fusion strategy to obtain a fusion result, and the fusion result is analyzed to detect the passenger getting on and off the vehicle behavior to obtain a behavior detection result.

6. The bus passenger flow statistics method based on deep learning according to claim 1 is characterized in that: Performing magnitude analysis on the behavior detection results to obtain a structured magnitude report, correlating the structured magnitude report with face recognition model parameters to obtain a correlation result, and configuring the face recognition model based on the correlation result, including: Aggregate behavior detection results in spatiotemporal dimensions, count the number of passengers getting on and off the bus within a unit time window, the distribution of behavior types, and regional density, divide the magnitude levels based on preset thresholds, and output a structured magnitude report; Magnitude features are extracted based on structured magnitude reports, and the magnitude features are associated with key parameters of the face recognition model to obtain association results. Hierarchical configuration is performed based on the association results, and the face recognition model is obtained by integrating all hierarchical configurations.

7. The bus passenger flow statistics method based on deep learning according to claim 1 is characterized in that: Based on the facial recognition model, historical passenger flow, vehicle Beidou trajectory and weather data are integrated to form a prediction model to obtain the predicted number of changes. The predicted number of changes is pushed to the dispatch center to optimize vehicle allocation, including: Obtain real-time passenger flow from the facial recognition system, access the historical passenger flow database to extract historical passenger flow, simultaneously collect Beidou vehicle trajectories, and obtain weather data through the meteorological interface; Decompose the time dimension into periodic time features and obtain the topological features of the site at the same time; Extracting common time series feature vectors and key time series feature vectors from real-time passenger flow and historical passenger flow based on periodic time features, extracting static features from weather data, performing a first concatenation on the common time series feature vectors and the static features, and performing a second concatenation on the key time series feature vectors and the static features; The first splicing and the second splicing are connected based on the topological features to form a prediction model, and a predicted number of changes is obtained, which is then pushed to a dispatch center to optimize vehicle allocation.

8. The bus passenger flow statistics method based on deep learning according to claim 7 is characterized in that: Push the predicted number of changes to the dispatch center to optimize vehicle allocation, including: Match key causal chains based on the predicted number of changes, and trigger resource allocation plans based on the key causal chains; A sandbox simulation is conducted based on the resource allocation plan, and the plan evaluation results are output. The dispatch center determines the mobilization instructions based on the plan evaluation results to optimize vehicle allocation.

Images