Scenic spot sightseeing vehicle empty seat display method

Abstract

The invention discloses an empty seat display method for a scenic spot sightseeing vehicle, particularly relates to the technical field of empty seat display, and discloses the empty seat display method for the scenic spot sightseeing vehicle. An array type thin film pressure sensor is embedded in a bottom plate of each seat, a narrow-angle ToF microspur radar is integrated above the seats, child label coordinates are obtained through UWB base stations at the four corners of a carriage, and a three-dimensional sensing data stream is formed. A feature vector is generated after normalization difference, a lightweight decision tree classifies the fine-grained states E0-E4 of the seat in real time, transient interference is filtered out through a finite state machine and a confidence model, and the fine-grained states E0-E4 are solidified into null / occupied two values. According to the method, weight, distance and position three-domain information is fused, and the real-time empty seat state can still be stably output under the scenes of vehicle driving, dense passenger flow, communication blind areas and the like.

Description

Technical Field

[0001] This invention relates to the field of empty seat display technology, and more specifically, to a method for displaying empty seats on a sightseeing vehicle in a scenic area. Background Technology

[0002] Sightseeing buses in scenic areas are a common mode of transportation for tourists traveling short distances during peak hours due to their flexible routes and frequent stops. The availability of empty seats on these buses typically relies on manual counting or reporting using a single sensor (such as photoelectric sensors, cameras, or seat pressure gauges). Manual methods are time-consuming and prone to errors; camera-based solutions show significantly reduced recognition rates in backlit, obstructed, or nighttime conditions; and relying solely on pressure or infrared sensors struggles to distinguish between transient scenarios such as children being picked up or passengers temporarily leaving their seats, leading to false reports of empty seats.

[0003] In addition, sightseeing vehicles often pass through 5G and LoRa blind spots while driving, and data cannot be transmitted back in seconds, causing cloud prediction models to fail and resulting in lag on the display end.

[0004] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method for displaying empty seats on a sightseeing vehicle in a scenic area to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for displaying empty seats on a sightseeing bus in a scenic area includes the following steps:

[0008] Collect the current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P. child (t);

[0009] The current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P are given. child (t) is normalized and differentially processed to generate a three-dimensional feature vector v(t); the feature vector v(t) is input into a pre-trained lightweight decision tree model to map v(t) to one of the five seat states;

[0010] The continuous state sequence is fed into a finite state machine with nodes E0–E4. Transient interference is filtered out according to the set state transition conditions and confidence scores to obtain the macroscopic state of the seat, State∈{Free,Occupied}.

[0011] When the macro state of a seat is Free and the duration is ≥τ_free, the edge node sends a <SeatID,State,Confidence> message to the display middleware via the MQTT protocol; when the macro state of a seat changes to Occupied, the empty seat mark is immediately revoked, where Confidence is the average confidence score of the most recent n frames; finally, it is displayed on the display end.

[0012] In a preferred embodiment, an array of thin-film pressure sensors installed on each seat base plate acquires the current instantaneous load W(t) in real time, a narrow-angle ToF micro-radar installed directly above the seat acquires the obstacle distance D(t) in real time, and the child tag spatial coordinates P... child (t) relies on a UWB tag worn on the collar or wristband and a two-way ranging network of three base stations deployed at the four corners of the carriage to obtain data, and all three data streams have a unified timestamp.

[0013] In a preferred embodiment, the generation of the three-dimensional feature vector v(t) is as follows: Where ΔW(t) = W(t) - W(t-1) represents the change in payload in the current frame relative to the previous frame; P seat W represents the geometric center coordinates of the seat in the local coordinate system. thr D represents the minimum threshold for identifying changes in a child's quality. seat D represents the radar height reference on the normal static seat surface. max R is the maximum allowable radar response range of the system. zone The radius of the "seat determination area" centered on the center of the seat.

[0014] In a preferred embodiment, the feature vector v(t) is input into a pre-trained lightweight decision tree model. The first layer of this model uses ΔW(t) as the splitting criterion to initially determine whether there is a payload and its possible level; the second layer combines D(t) and P... child (t), further distinguishing whether a child is being picked up or a passenger is only temporarily standing up; the third layer introduces the duration of short-term behavior to suppress short-term jumps caused by vibration or instantaneous movements;

[0015] Finally, v(t) is mapped to one of the five seat states.

[0016] In a preferred embodiment, the five seat states are: E0: seat vacant; E1: child sits alone; E2: adult sits; E3: child is picked up; E4: passenger is half-standing or twisting.

[0017] In a preferred embodiment, the confidence score is derived by weighted summation and normalization based on feature stability weights, transition consistency scores, and perceptual channel synergy.

[0018] In a preferred embodiment, the feature stability weight is the covariance of v(t) within the most recent T = 5 frames; the transition consistency score is the similarity between the previous state transition path and the current path; and the perceptual channel synergy is W(t), D(t), and P. child (t) The three parties have a consistent judgment on the trend of state transition.

[0019] In a preferred embodiment, the SeatID is a unique identifier for the seat, taken from a fixed number inside the vehicle; the State represents the current business-related "empty / occupied" state, allowing only two strings: Free for "empty seat" and Occupied for "occupied seat"; the Confidence gives the confidence level of this state judgment, ranging from 0 to 1; it is not a single-frame p edge It is not the average confidence score obtained from the "most recent three window votes".

[0020] In a preferred embodiment, the interaction rules of the display terminal revolve around State and Confidence:

[0021] When State = Free and Confidence ≥ 0.7, the display shows an empty seat;

[0022] When State = Free but Confidence < 0.7, the display shows "Pending Confirmation".

[0023] If any subsequent frame is fixed (Ψ≥0.75) and republished with State=Free and Confidence≥0.7, the display will immediately cancel the "Pending Confirmation" display and instead display an empty seat; if the fixed result is "Occupied", it will switch to displaying a reserved seat.

[0024] In a preferred embodiment, the formula for calculating Ψ is as follows: Ψ=α·p edge +(1-α)·ρ seq α is the adjustment coefficient, and ρ seq The calculation is as follows: Where n major N represents the number of times the main state appears in the current sliding window, and N is the window length.

[0025] The technical effects and advantages of this invention are as follows:

[0026] This invention provides a method for displaying empty seats on scenic sightseeing buses that integrates multi-source sensing, state recognition, and confidence modeling, significantly improving the accuracy and robustness of empty seat detection. The system acquires 3D sensing data with a unified timestamp using a seat pressure array, narrow-angle ToF micro-radar, and UWB spatial positioning technology. It generates feature vectors in real time and inputs them into a lightweight five-class decision tree for state classification. A state transition graph is constructed using a finite state machine to effectively identify fine-grained dynamic behaviors, including "a child being picked up" and "a passenger half-standing up." Through a confidence scoring mechanism and a state consistency index Ψ for joint decision-making, it can robustly output reliable empty seat states even in complex environments such as vehicle vibration and frequent boarding and alighting. The system also publishes state information to the display middleware via the MQTT protocol, achieving end-to-end real-time updates.

[0027] Furthermore, this invention supports intelligent switching between three display states: "Free," "Occupied," and "Pending Confirmation," enhancing tourists' intuitive perception of seat availability. The system possesses edge computing capabilities, eliminating reliance on cloud inference and adapting to operational requirements in 5G / LoRa communication blind spots. It also incorporates a cloud-based self-learning mechanism, supporting dynamic model updates and parameter optimization, automatically adapting to different vehicle types, routes, and tourist behavior characteristics. The UWB tags also feature location and alarm functions, addressing the need to prevent children from getting lost. The overall system structure is lightweight, highly responsive, easy to deploy and integrate, and suitable for visual management of sightseeing bus seats in high-traffic, highly variable environments, demonstrating significant practical value and promising prospects for wider adoption. Attached Figure Description

[0028] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0029] Figure 1 This is a flowchart illustrating a method for displaying empty seats on a sightseeing bus in a scenic area, according to the present invention.

[0030] Figure 2 Flowchart for edge nodes to publish seat status to the middleware. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1

[0033] This invention provides a method for displaying empty seats on sightseeing vehicles in scenic areas, such as... Figure 1 As shown, it includes the following steps:

[0034] Collect the current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P. child (t);

[0035] The current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P are given. child (t) is normalized and differentially processed to generate a three-dimensional feature vector v(t); the feature vector v(t) is input into a pre-trained lightweight decision tree model to map v(t) to one of the five seat states;

[0036] The continuous state sequence is fed into a finite state machine with nodes E0–E4. Transient interference is filtered out according to the set state transition conditions and confidence scores to obtain the macroscopic state of the seat, State∈{Free,Occupied}.

[0037] When the macro state of a seat is Free and the duration is ≥τ_free, the edge node sends a <SeatID,State,Confidence> message to the display middleware via the MQTT protocol; when the macro state of a seat changes to Occupied, the empty seat mark is immediately revoked, where Confidence is the average confidence score of the most recent n frames; finally, it is displayed on the display end.

[0038] Specifically,

[0039] The edge node first acquires the load W(t) in real time through an array of thin-film pressure sensors embedded in the seat base. The output of each unit pressure gauge is amplified by a strain gauge bridge and a low-noise instrumentation amplifier, and then pushed to the microcontroller at a fixed sampling rate of 50Hz. During the power-on self-test phase, the system performs a one-time multi-point linearization of each sensor using an unloaded, standard weight calibration curve. Subsequently, a temperature-compensated LUT (LUT stands for Look-Up Table; it stores the temperature-zero drift correction coefficient, which the MCU can quickly look up and compensate for the thermal drift of the pressure sensor after reading the current temperature) is used to correct thermal drift. First-order IIR (IIR stands for Infinite Impulse Response Filter; here, a first-order IIR is used for online denoising, balancing real-time performance and resource consumption) denoising is implemented on the MCU side. Multiple sensing units are weighted and fused to obtain a single-channel load W(t), which is then compared with the previous 20 frames of moving average to remove abnormal spikes, and finally written into a unified timestamp buffer.

[0040] The radar range D(t) is obtained by a narrow-beam 60GHz FMCW (FMCW stands for Frequency-Modulated Continuous-Wave, a radar system that calculates target range by transmitting linear frequency modulated pulses and measuring the echo frequency difference) micro radar mounted 20cm above the seat. The radar sequentially transmits linear frequency modulated pulses and receives echoes. The FPGA (Field-Programmable Gate Array, responsible for high-speed radar sampling and FFT acceleration, reducing the computational burden on the MCU) front end performs a fast FFT (Fast Fourier Transform, used to transform the radar intermediate frequency signal to the range-Doppler spectral domain and extract echo peaks) to obtain the range-dimensional energy spectrum. The microcontroller then extracts the range corresponding to the strongest target peak as D(t). To avoid multipath interference from the vehicle's internal metal structure, the system performs a 30-second static environmental scan during initial deployment, constructing a background template using constant false alarm rate (CFAR) detection and subtracting it in real time during operation. Then, it smooths the output using a three-frame median filter and an α-β predictor (an α-β predictor is a second-order linear predictive filter, also known as an Alpha-Beta Filter, used for smoothing target range sequences and short-time velocity estimation; its α and β are adjustable gains). The radar mission cycle starts synchronously with the ADC (Analog-to-Digital Converter, used to quantize the continuous voltage output from the pressure plate into a digital signal). After continuous processing, D(t) and a timestamp of the same period are written to a buffer.

[0041] Child tag spatial coordinates P child (t) relies on a UWB (Ultra-Wideband) tag worn on the collar or wristband, which uses nanosecond-level pulses in the 3–10 GHz band to achieve centimeter-level precession time difference ranging, and a two-way ranging network of three base stations deployed at the four corners of the carriage. Every 20 ms, the tag sends a single pulse Chirp to each base station. After the base station responds, two-way time difference ranging is completed in the 3 GHz ultra-wideband channel. The edge node uses a precisely synchronized TCXO (Temperature-Compensated Crystal Oscillator; provides a stable time base to UWB and MCU to ensure nanosecond-level ranging accuracy) clock to input the three-way ranging results into the least squares solver to obtain the three-dimensional spatial coordinates P of the child tag. child (t). Then, a unified global timestamp is immediately sent to the cache.

[0042] The child wristband also includes location tracking and alarm functions to prevent children from getting lost in crowded scenic areas.

[0043] All data streams are timestamped and written to a circular buffer to provide time-consistent basic data for subsequent state identification.

[0044] Data from three sensing channels are simultaneously acquired during each sampling period (50Hz): the current instantaneous load W(t) output by the seat pressure sensing array, the distance D(t) to the nearest obstacle in the overhead space output by the narrow-angle ToF radar, and the spatial coordinates P of the child tag detected by the UWB positioning module. child (t). Simultaneously, the following set of internally preset static reference parameters are: W thr D represents the minimum threshold for identifying changes in a child's quality. seat D represents the radar height reference on the normal static seat surface. max R is the maximum allowable radar response range of the system. zone The radius of the "seat determination area" centered on the center of the seat.

[0045] During the feature preprocessing stage, the system processes the current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P. child (t) Perform normalization and feature difference operations to calculate the three-dimensional normalized feature vector at the current time step: Where ΔW(t) = W(t) - W(t-1) represents the change in payload in the current frame relative to the previous frame; P seat This represents the geometric center coordinates of the seat in the local coordinate system. This three-dimensional vector integrates three key pieces of information: the load change on the seat at the current moment, whether the seat is directly above a person, and the child's relative spatial position. It can effectively distinguish behavioral patterns such as sitting down, being picked up, standing, and getting up from the seat.

[0046] Subsequently, the system invokes a pre-trained lightweight five-class decision tree model (depth ≤ 3) to classify the current feature vector v(t) in real time. The first layer of this model uses ΔW(t) as the splitting criterion to initially determine whether there is a load and its possible level (e.g., significant heavy pressure indicates an adult, light pressure indicates a child, and zero load indicates an empty seat); the second layer combines D(t) and P... child (t) further distinguishes whether a child is being picked up or a passenger is only temporarily getting up; the third layer introduces short-term behavior duration to suppress short-term jumps caused by vibration or instantaneous actions, ensuring that the state output has physical coherence.

[0047] Ultimately, the system will map v(t) to one of five seat states within a 1-frame sampling period:

[0048] E0: Seat vacant, indicating no significant pressure, no obstruction, and the UWB label is not near the seat;

[0049] E1: A child is sitting alone, characterized by light load, low radar height, and the child's tag being close;

[0050] E2: An adult is seated with a noticeable change in weight, but no valid UWB tag is nearby;

[0051] E3: When a child is picked up, it is characterized by a sudden drop in pressure, the child tag moving away from the child, and the radar height not yet returning to normal.

[0052] E4: Passengers are half-asleep or twisting, pressure fluctuations are moderate, radar fluctuates intermittently, and the UWB tag is still in the near field.

[0053] States are connected by state edges, representing possible behavioral paths (e.g., "child sits down" → "child is picked up"). Each state edge has a transition condition (v(t), Δt) and a confidence score p. edge This ensures that the state does not jump due to a single frame anomaly.

[0054] Each state transition edge is accompanied by a confidence score p. edge ∈[0,1], used to measure whether the current transition is worthy of acceptance. This value is dynamically calculated based on the following factors:

[0055] Feature stability weight: Take the covariance of v(t) within the most recent T=5 frames. The smaller the covariance, the smaller the feature fluctuation, which means the state judgment is considered more stable, thus increasing the confidence.

[0056] Transition Consistency Score: The similarity between the previous state transition path and the current path. A higher similarity score increases the confidence level.

[0057] Perceptual channel synergy: If W(t), D(t), P child (t) If the three parties have the same judgment on the trend of state transition (e.g., pressure decreases, radar altitude increases, tag moves away from synchronization), then the confidence level is further increased.

[0058] The specific formula is as follows: P = λ0TZ + λ1ZY + λ2GZ; where P, after normalization, becomes the confidence score p. edge TZ represents the feature stability weight; ZY represents the transfer consistency score; GZ represents the perceptual channel synergy; λ0, λ1, and λ2 are the feature stability weight, transfer consistency score, and perceptual channel synergy weight, respectively, all of which are greater than zero and are designed as needed.

[0059] A finite state machine is introduced to drive state transitions. The FSM (Finite State Machine) forms a state diagram with nodes E0–E4. The transition between each pair of nodes is triggered by the current feature vector v(t) and the state dwell time Δt.

[0060] Taking the typical path E1→E3 as an example, its transition condition is a sudden change in load, and the child's UWB tag spatial coordinates P child The distance between (t) and the parent label is less than 15cm, and the state transition must be completed within 0.5 seconds; while the transition from E3 to E0 requires D(t) ≥ D seat +8cm, and maintain this condition for 0.3 seconds. To mitigate interference from vehicle body vibration, lighting disturbances, etc., a confidence score p is associated with each transfer edge. edge .

[0061] Construct a dynamic feedback loop: the confidence score p of the FSM output. edge The state consistency index ρ in the sliding window seq (Consistency ratio) Both factors interact in the strategy for whether a state is fixed:

[0062] If the confidence score p of a certain transfer path edge If the value is relatively high, but the state consistency index in its sliding window is low (e.g., one-tenth), further observation is needed before immediate solidification.

[0063] If the state consistency index in the sliding window is high (e.g., 4 / 5), but its confidence score p edge If the value remains consistently low (e.g., below 0.4), the solidification process should be postponed or a reassessment should be requested from the upstream authority.

[0064] If p edge and ρ seq At the same time, if the temperature is high, the state will solidify quickly;

[0065] If the two conflict, the system should maintain its current stable state and mark it as "pending confirmation" for further confirmation by subsequent frames.

[0066] A confidence stability index (Ψ) is defined to jointly evaluate the FSM output and sequence stability, as follows: Ψ = α·p edge +(1-α)·ρ seq Frames with Ψ≥0.75 are used as high-confidence samples for online fine-tuning; frames with 0.45≤Ψ<0.75 are only used for semi-supervised consistency distillation; frames with Ψ<0.45 are ignored. α is an adjustment coefficient, set as needed.

[0067] Among them, the state consistency index (ρ) seq ): Where n major N represents the number of times the main state appears in the current sliding window (maximum frequency), and N is the window length.

[0068] Based on the above processing, the system's rules for issuing empty seats are clarified. The system will only push a "SeatFree" event to the in-vehicle display middleware when the current fixed state is E0 or E3 and this state has been persisting for at least τ_free = 0.2 seconds; once the state switches back to E1 or E4, the empty seat mark will be immediately removed to avoid misleading passengers; if the system enters E2 (adult seated), it will block any empty seat reports for that seat for the next 30 seconds to prevent multiple false alarms caused by an adult temporarily leaving the seat.

[0069] Furthermore, considering the dynamic changes in the vehicle's operating environment, an intelligent correction mechanism for external context factors is introduced. When GPS detects that the vehicle is about to stop at a station (≤2 minutes) or the passenger density in the carriage exceeds 0.9 people / seat, the system automatically increases the W... thr The system uses the τ_free threshold to suppress the risk of false alarms; if the vehicle is on a sharp bend or a bumpy road, the system actively tightens the E4→E0 transfer logic to mitigate false alarms caused by vibration.

[0070] Furthermore, to adapt to differences in routes, vehicle types, and user behavior, an online self-learning and parameter adaptation mechanism is activated in the cloud. During off-peak hours each day, the system uploads the FSM operation logs to the cloud, re-estimates the Gaussian distribution model for each state using the expectation-maximization algorithm, and generates an updated decision tree; simultaneously, the transition confidence p is adjusted. edge and threshold parameter (W) thr D seat ,τ_free) performs a Bayesian update to ensure that the model parameters issued the next day are closer to the real scene.

[0071] Finally, the edge nodes continuously publish <SeatID, State, Confidence> messages to the presentation middleware via the MQTT protocol; such as Figure 2 As shown;

[0072] SeatID is a unique identifier for a seat, used to tell the display client "which physical seat this status update belongs to". In actual deployment, SeatID is usually a fixed number inside the vehicle (e.g., "Car-12 / Seat-05"), or a UUID burned into the device at the factory. The display client relies on this key to map the backend state to the corresponding seat icon on the interface, so it must remain unchanged and not be repeated throughout the entire communication link.

[0073] The State classifies a seat as either "empty" or "occupied" in the current business context, allowing only two strings: Free, meaning "seat available." This corresponds to the underlying finite state machine set {E0, E3}—either a completely empty seat is detected (E0), or a child has just been picked up and is not yet occupied (E3). Occupied, meaning "seat occupied." This corresponds to the state set {E1, E2, E4}, indicating a child sitting alone, an adult sitting down, or a passenger only partially adjusting their posture. This approach compresses five fine-grained seating postures into a single empty / occupied value that only the front-end is concerned with, reducing communication volume and front-end logic complexity.

[0074] Confidence gives the confidence level of this state judgment, ranging from 0 to 1. It is not a single-frame p. edge It is not the average confidence score obtained from the "most recent three window votes".

[0075] The interaction rules on the display side revolve around State and Confidence:

[0076] When State = Free and Confidence ≥ 0.7, it means that the system is sufficiently confident in the "empty seat" judgment, and the front end immediately renders the corresponding seat as a completely opaque green (or the "empty seat" color block agreed upon by the developers).

[0077] When State = Free but Confidence < 0.7, the system still tends to assume the seat is empty, even though the threshold for confirmation has not been reached. The front end then uses a semi-transparent or flashing "Pending Confirmation" style to remind passengers that the chair "appears empty, but is still being verified."

[0078] If any subsequent frame is fixed (Ψ≥0.75) and republished with State=Free and Confidence≥0.7, the display will immediately stop flickering and switch to a formal "empty seat"; if the fixed result is Occupied, it will switch to a seat-occupying display.

[0079] 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, or a combination of computer software and electronic hardware. 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 implementation should not be considered beyond the scope of this application.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0081] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0083] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for displaying empty seats on a sightseeing vehicle in a scenic area, characterized in that, Includes the following steps: Collect the current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P. child (t); The current instantaneous load W(t), obstacle distance D(t), and child tag spatial coordinates P are given. child (t) is normalized and differentially processed to generate a three-dimensional feature vector v(t); the feature vector v(t) is input into a pre-trained lightweight decision tree model to map v(t) to one of the five seat states; The continuous state sequence is fed into a finite state machine with nodes E0–E4. Transient interference is filtered out according to the set state transition conditions and confidence scores to obtain the macroscopic state of the seat, State∈{Free,Occupied}. When the macro state of a seat is Free and the duration is ≥τ_free, the edge node sends a <SeatID,State,Confidence> message to the display middleware via the MQTT protocol; when the macro state of a seat changes to Occupied, the empty seat mark is immediately revoked, where Confidence is the average confidence score of the most recent n frames; finally, it is displayed on the display end.

2. The method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 1, characterized in that: An array of thin-film pressure sensors installed on each seat base plate acquires the current instantaneous load W(t) in real time, while a narrow-angle ToF micro-radar installed directly above the seat acquires the obstacle distance D(t) in real time, along with the child tag's spatial coordinates P. child (t) relies on a UWB tag worn on the collar or wristband and a two-way ranging network of three base stations deployed at the four corners of the carriage to obtain data, and all three data streams have a unified timestamp.

3. The method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 1, characterized in that: The generated three-dimensional feature vector v(t): Where ΔW(t)=W(t)-W(t-1) represents the change in load of the current frame relative to the previous frame; P seat W represents the geometric center coordinates of the seat in the local coordinate system. thr D represents the minimum threshold for identifying changes in a child's quality. seat D represents the radar height reference on the normal static seat surface. max R is the maximum allowable radar response range of the system. zone The radius of the "seat determination area" centered on the center of the seat.

4. The method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 1, characterized in that: The feature vector v(t) is input into a pre-trained lightweight decision tree model. The first layer of this model uses ΔW(t) as the splitting criterion to initially determine whether there is a payload and its possible level; the second layer combines D(t) and P... child (t), further distinguishing whether a child is being picked up or a passenger is only temporarily standing up; the third layer introduces the duration of short-term behavior to suppress short-term jumps caused by vibration or instantaneous movements; Finally, v(t) is mapped to one of the five seat states.

5. A method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 4, characterized in that: The five seat states are: E0: Seat vacant; E1: A child sits alone; E2: Adults take their seats; E3: The child is being picked up; E4: Passenger is half-standing up or twisting.

6. The method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 1, characterized in that: The confidence score is derived by weighted summation and normalization based on feature stability weights, transition consistency scores, and perceptual channel synergy.

7. A method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 6, characterized in that: The feature stability weight is the covariance of v(t) within the most recent T=5 frames; the transition consistency score is the similarity between the previous state transition path and the current path; the perceptual channel synergy is W(t), D(t), and P. child (t) The three parties have a consistent judgment on the trend of state transition.

8. A method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 1, characterized in that: The SeatID is a unique identifier for the seat, taken from a fixed number inside the vehicle; the State represents the current business-related "empty / occupied" state, allowing only two strings: Free for "empty seat" and Occupied for "occupied seat"; the Confidence gives the confidence level of this state judgment, ranging from 0 to 1; it is not a single-frame p edge It is not the average confidence score obtained from the "most recent three window votes".

9. A method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 1, characterized in that: The interaction rules for the display end revolve around State and Confidence: When State = Free and Confidence ≥ 0.7, the display shows an empty seat; When State = Free but Confidence < 0.7, the display shows "Pending Confirmation". If any subsequent frame is fixed (Ψ≥0.75) and republished with State=Free and Confidence≥0.7, the display will immediately cancel the "Pending Confirmation" display and instead display an empty seat; if the fixed result is "Occupied", it will switch to displaying a reserved seat.

10. A method for displaying empty seats on a sightseeing vehicle in a scenic area according to claim 9, characterized in that: The formula for calculating Ψ is as follows: Ψ=α·p edge +(1-α)·ρ seq α is the adjustment coefficient, and ρ seq The calculation is as follows: Where n major N represents the number of times the main state appears in the current sliding window, and N is the window length.

Images