Coal mine underground passenger car accurate number of people statistics and identity recognition method and system

Abstract

This invention relates to the field of coal mine safety monitoring and auxiliary transportation technology, and discloses a method and system for accurate passenger counting and identification in underground coal mine passenger carriages. The system uses an onboard fusion processor to simultaneously collect real-time video streams and directional RFID card reader event streams within the carriage, generating visually estimated passenger numbers and RFID passenger lists in parallel. The processor performs spatiotemporal alignment and consistency checks on the dual-modal data. When the difference between the two is within the allowable error range, the RFID passenger list is confirmed as a valid list. When the checks are inconsistent, the system executes an output strategy based on the maximum safety value or redundant filtering logic based on RSSI variance analysis according to the sign of the difference. It also has an RFID-dominated fallback mechanism in case of visual failure. This invention effectively solves the problems of large counting errors and mismatches between passengers and cards in underground mines, achieving accurate binding of passenger numbers and identities, and providing reliable support for mine safety monitoring and emergency rescue.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety monitoring and auxiliary transportation technology, specifically to a method and system for accurate passenger counting and identification in underground coal mine passenger carriages. Background Technology

[0002] The underground auxiliary transportation system in coal mines is a crucial link connecting the mine's underground yard with the working face, with passenger cars serving as the core carrier for personnel movement. Real-time monitoring of the precise number of passengers and their identities in each car is essential for preventing overloading risks, optimizing vehicle scheduling, and facilitating emergency rescue operations in disaster situations.

[0003] Currently, underground personnel monitoring mainly relies on machine vision or radio frequency identification (RFID) technology. However, single technologies have significant limitations in the complex underground environment. When using cameras for personnel counting, visual algorithms are prone to counting errors due to poor lighting conditions, high dust concentrations, and personnel blocking each other's view, and cannot obtain specific personnel identification information. When using RFID technology for monitoring, although it can identify personnel, the narrow underground tunnels and numerous metal supports cause severe multipath effects and reflections of radio frequency signals. This can easily lead to card readers misreading tags outside the vehicle or missing shielded tags, making it difficult to accurately determine whether personnel are actually inside the vehicle, resulting in frequent discrepancies between the person and the card.

[0004] While some existing solutions attempt to combine video and radio frequency (RF) technologies, most are merely simple overlay displays, lacking in-depth data fusion mechanisms. When visual headcount differs from RF reads, existing systems often lack effective logical means to determine whether someone is present without a card or has a card but no one, failing to effectively remove redundant tags caused by signal reflection, resulting in low reliability of monitoring data. Furthermore, the harsh operating environment of underground equipment means that when a single sensor malfunctions or crashes, existing systems typically lose monitoring capabilities entirely, lacking automatic degradation or fallback mechanisms, leading to monitoring blind spots.

[0005] Furthermore, in the event of emergencies such as vehicle derailment or gas overload, on-site communication and power supply are often at risk of interruption. Most existing monitoring systems employ a loop recording overlay mechanism, lacking critical data protection logic for the moment an accident is triggered. If the system fails to stop recording or lock the data promptly after an accident, crucial pre-accident personnel lists and video footage are easily overwritten by subsequent invalid data. This prevents rescue personnel from quickly obtaining the exact list of trapped personnel and their last known locations within the accident-affected carriages, impacting rescue decision-making. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides a method for accurate passenger counting and identification in underground coal mine passenger carriages. This method is applied to an onboard fusion processor and specifically includes the following steps:

[0007] Step S100: Start data synchronization acquisition for a single trip: When the carriage enters the operating cycle or the door is open, control the panoramic camera to acquire real-time video stream data, and control the directional RFID card reader to generate RFID card reading event stream data when it senses the RFID signal of the positioning card; receive the real-time video stream data and the RFID card reading event stream data; Step S200: Parallel feature extraction of multimodal data: Frame-by-frame human target detection is performed on the video stream data using a pre-set visual analysis model to output visually estimated number of people; simultaneously, the radio frequency card reading event stream data is processed using a time window clustering algorithm to remove redundant reading data and extract unique identification identifiers to generate radio frequency passenger list data. Step S300: Perform consistency verification calculation for multimodal data: obtain the numerical value of the visually estimated number of people data and the element count value of the radio frequency passenger list data, calculate the absolute value of the difference between the two, and determine whether the absolute value of the difference is within the preset allowable error range. Step S400: Generate a verification result data packet: When the absolute value of the difference is within the allowable error range, the statistical result is determined to be valid, and the radio frequency passenger list data is used as the current passenger list; when the absolute value of the difference exceeds the allowable error range, the anomaly type is determined according to the positive or negative sign of the difference and an anomaly classification identifier code is generated; the final determined number of passengers, the radio frequency passenger list data, and the anomaly classification identifier code are encapsulated into a verification result data packet. Step S500, Data Upload and Platform-Side Linkage Response: The verification result data packet is sent to the ground application layer through the data transmission layer; the ground application layer parses the verification result data packet and triggers alarm logic when it contains the abnormal classification identifier code or the number of passengers exceeds the rated capacity.

[0008] Preferably, step S200, which uses a preset visual analysis model to output visually estimated number of people, specifically includes: The original video frame image is extracted from the video stream data, and geometric distortion correction and contrast-limited adaptive histogram equalization are performed on the original video frame image to generate an enhanced grayscale image. After scaling and normalizing the enhanced grayscale image, it is input into a target detection network using a lightweight convolutional neural network model, and outputs an original detection result set containing the confidence scores of candidate targets; non-maximum suppression processing is performed on the original detection result set, and a list of valid targets is generated according to a preset confidence threshold and an overlap threshold; The number of targets in the list of valid targets at the current time t is counted as the number of visually detected people. The arithmetic mean of the number of visually detected people in the most recent k frames is calculated using a sliding time window smoothing filter of length k, and the smoothed estimated number of visually detected people data is obtained.

[0009] Preferably, the step S200 of generating the radio frequency passenger list data specifically includes the following steps: During the time the door remains open and the subsequent preset delay period, the RF card reader event stream data is continuously cached, and the RF card reader event stream data consists of multiple discrete card reader tuples; The cached RFID card reader event stream data is grouped according to the unique identification code of the electronic tag to form multiple single card time series, and the time interval between two adjacent card reader tuples in the series is calculated. If there is a breakpoint in the single card time series where the time interval is greater than the preset maximum signal interruption threshold, the series is divided into multiple independent passage event clusters; Perform validity filtering on the segmented access event clusters: remove access event clusters containing fewer than the minimum number of valid reads or with a time span shorter than the minimum access duration; A cluster of passage events whose time span exceeds a preset lingering alarm threshold is designated as a lingering event at the entrance. The unique identification codes of all electronic tags that pass the validity filter are counted to construct the radio frequency passenger list data.

[0010] Preferably, before step S300, a step of performing spatiotemporal synchronization and alignment of the data stream is further included: A high-precision real-time clock module is used to maintain a unified system time base, and acquisition timestamps are applied to the video stream data and the RFID card reader event stream data, respectively. When the vehicle-mounted fusion processor receives the original video frame transmitted by the panoramic camera in the underlying driver, it captures the hardware interrupt moment and writes it as the acquisition timestamp into the video frame header to form a video frame sequence. Establish a visual data buffer for temporarily storing the visually estimated number of people data and an RF data buffer for temporarily storing the RF passenger list data; At each verification trigger moment, the acquisition timestamp of the latest processed video frame is read, and a backtracking search is performed in the radio frequency data buffer to find the radio frequency list state that is closest to the acquisition timestamp of the video frame that is less than or equal to the acquisition timestamp of the video frame. By employing a nearest neighbor matching strategy or a zero-order hold strategy, the visual detection result at that moment and the corresponding radio frequency list state are locked as a data pair at the same physical moment.

[0011] Preferably, the processing in step S400 when the absolute value of the difference is within the allowable error range specifically includes: Read the tolerance threshold set based on the detection accuracy error of the panoramic camera and the rated passenger capacity of the carriage; If the absolute value of the difference is less than or equal to the tolerance threshold, it is determined that the multimodal data at the current moment meets the consistency requirements. Data acceptance is performed based on data dimension priority: the radio frequency passenger list data containing the identity of the person is confirmed as the final valid passenger list, and the number of elements in the radio frequency passenger list data is output as the current accurate number of passengers.

[0012] Preferably, the processing in step S400 when the absolute value of the difference exceeds the allowable error range specifically includes: Calculate the direction indicator of the difference between the visually estimated number of people data and the number of elements in the radio frequency passenger list data; If the difference direction indicator is greater than the tolerance threshold, it is determined to be a visual count overflow state, and an output strategy based on the maximum safety value is executed: the visually estimated number of people with the larger value is confirmed as the current actual number of passengers, and a missing person location card alarm indicator is generated. If the difference direction indicator is less than the negative tolerance threshold, it is determined to be an RF count overflow state, and the redundancy screening logic based on signal characteristics is initiated: analyze the time series variance of the received signal strength indication of each electronic tag in the RF passenger list data, and mark tags with variance less than the preset static threshold or abnormal reading timestamps as suspected redundant tags. A dual output strategy is implemented: the visually estimated number of people is output to reflect the physical occupancy situation, while all card numbers in the radio frequency passenger list data are retained and an anomaly verification suffix is ​​added to the suspected redundant tags.

[0013] Preferably, the method further includes equipment status monitoring and automatic degradation steps: The device status monitoring subroutine is executed in parallel to monitor the hardware interrupt signal of the video input interface and calculate the grayscale histogram and inter-frame difference coefficient of the current video frame. When a video signal loss fault or sensor logic crash is detected, an attempt is made to send a power reset command to the panoramic camera. If the fault persists after resetting, the system operating mode will be switched to RF-dominated fallback mode: the operation of the visual analysis algorithm module will be suspended, the consistency check will be stopped, the number of unique electronic tags contained in the RF passenger list data will be used as the estimated number of passengers, and a visual blocking fault code will be generated, encapsulated in a degraded state data packet, and uploaded.

[0014] Preferably, the processing of the ground application layer in step S500 specifically includes: Communication is established with the mine personnel / vehicle positioning system server through a standard data interface, and the absolute coordinate data of each car in the underground roadway is obtained in real time using the unique number of the car as the index key. The absolute coordinate data and the verification result data packet are spatiotemporally fused and correlated to drive the large screen display system to dynamically refresh the carriage location icon on the mine electronic map; When the abnormal classification code is received or overcrowding is detected, the position icon of the corresponding carriage is turned to flash, and a voice synthesis command is sent to the underground broadcasting system.

[0015] Preferably, step S500 further includes emergency rescue auxiliary processing: The ground application layer continuously monitors the status data of the third-party downhole environmental monitoring system or vehicle operation monitoring system; when an accident trigger signal or emergency rescue instruction is received, the verification result data packet last uploaded by the accident-related carriage is retrieved from the historical database server; Write protection is applied to the verification result data packet to prevent it from being overwritten by the loop recording mechanism. The personnel identity list within the packet is then parsed to generate an emergency rescue list, which is then sent to the rescue team's handheld terminal via a private wireless network message push service.

[0016] Preferably, a system for accurate passenger counting and identification in underground coal mine passenger carriages includes: The underground sensing layer includes an on-board terminal installed in the underground passenger compartment, the on-board terminal including a panoramic camera, a directional radio frequency identification card reader and an on-board fusion processor; The panoramic camera is used to collect real-time video stream data inside the carriage; the directional RFID card reader is used to generate card reading event data containing card number and timestamp. The vehicle-mounted fusion processor is connected to the panoramic camera and the directional radio frequency identification card reader respectively, and is configured to execute the method for accurate number of passengers and identity recognition in underground coal mine passenger compartments. The data transmission layer, including a mining industrial ring network or a wireless private network, is connected to the vehicle-mounted fusion processor and is used to upload the verification result data packet; The ground application layer includes a security monitoring platform, a dispatch and command center, and a large-screen display system. The security monitoring platform is connected to the data transmission layer and is used to parse the verification result data packets and perform overload judgment and alarm logic.

[0017] This invention provides a method and system for accurate passenger counting and identification in underground coal mine passenger carriages. It has the following beneficial effects: (1) This invention calculates the absolute value of the difference between the visually estimated number of people data and the number of elements in the radio frequency passenger list data, and determines whether the absolute value of the difference is within the preset allowable error range. This effectively solves the counting error problem of a single sensor in the complex environment of the mine. When the consistency requirements are met, the system can confirm the radio frequency passenger list data as the final valid passenger list based on the data dimension priority. While ensuring accurate passenger count, it also achieves accurate identification of the identity of the people in the carriage.

[0018] (2) This invention has a complete anomaly attribution and handling procedure and equipment status monitoring mechanism, which improves the robustness of the system. When inconsistencies occur, the system can distinguish the anomaly type based on the direction of the difference: for visual count overflow, a personnel missing location card alarm is generated to prevent overcrowding risk; for radio frequency count overflow, a redundancy screening logic based on signal characteristics is initiated, and suspected redundant tags are eliminated by analyzing the time series variance of the received signal strength indication. In addition, when a panoramic camera malfunction is detected and reset is ineffective, the system can automatically switch to radio frequency-dominated fallback mode, directly using the radio frequency passenger list data as an estimate of the number of passengers, ensuring the continuity of monitoring data under extreme conditions.

[0019] (3) The present invention integrates emergency rescue auxiliary logic at the ground application layer, providing key data support for mine safety accident handling. The safety monitoring platform is linked with the underground environmental monitoring system through a standard data interface. When it receives an accident trigger signal, it immediately performs the last known state extraction operation, retrieves the last uploaded verification result data packet of the accident-related carriage and performs a write protection operation to prevent it from being covered by the loop recording mechanism. The emergency rescue list generated by the system through parsing the data packet contains the name, employee number and last known location of the person in distress, eliminating rescue blind spots and improving the efficiency and accuracy of emergency rescue. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0021] 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.

[0022] Please see Figure 2 This invention provides a system for accurate passenger counting and identification in underground coal mine passenger carriages, comprising: an underground sensing layer, a data transmission layer, and a surface application layer, specifically: The underground sensing layer includes vehicle-mounted terminals installed in the underground passenger cars and an underground roadway network distributed in the underground transport roadways. The vehicle-mounted terminals serve as independent data acquisition and edge computing nodes, and are configured separately for each passenger car.

[0023] The vehicle-mounted terminal specifically includes: a panoramic camera, a directional RFID card reader, and a vehicle-mounted fusion processor. The panoramic camera is connected to the vehicle-mounted fusion processor via a video data transmission cable to collect real-time video stream data covering the interior area of ​​the vehicle. The directional RFID card reader is connected to the vehicle-mounted fusion processor via a data communication interface to generate card reading event data containing the card number and timestamp when a person wearing a positioning card passes through the vehicle door, and sends it to the vehicle-mounted fusion processor. The vehicle-mounted fusion processor adopts mining-grade intrinsically safe or explosion-proof protective packaging, and is connected to the panoramic camera and the directional RFID card reader respectively to receive video stream data and card reading event data.

[0024] The underground roadway network includes a roadway global positioning base station network, which is used to track the real-time location of personnel wearing positioning cards in the roadway and generate a global personnel location data stream. The vehicle-mounted fusion processor communicates with the underground roadway network to obtain the global personnel location data stream.

[0025] The vehicle-mounted fusion processor runs a multimodal data fusion algorithm to perform spatiotemporal alignment and verification calculations on video stream data, card reading event data, and global personnel location data stream, and outputs verified data containing the accurate number of passengers and a list of their identities.

[0026] The data transmission layer includes a mining industrial ring network or a wireless private network, which is connected to the vehicle-mounted fusion processor to upload the verified data via the TCP / IP protocol.

[0027] The ground application layer includes a safety monitoring platform, a dispatch and command center, and a large-screen display system. The safety monitoring platform is deployed on a ground server and connected to the data transmission layer. It is used to parse and verify the data and execute the over-capacity judgment logic. The dispatch and command center and the large-screen display system are both connected to the safety monitoring platform and are used to display visualized carriage monitoring images and alarm status information.

[0028] Please see Figure 1 Based on the above, this invention provides a method for accurate passenger counting and identification in underground coal mine passenger carriages, specifically including the following steps: Step S100: Start data synchronization acquisition for a single trip. When the vehicle fusion processor detects that the carriage has entered a running cycle or the door is open, it controls the panoramic camera and the directional RFID card reader to enter the working state. The panoramic camera acquires continuous optical images inside the carriage, generates video stream data with time information and transmits it to the vehicle fusion processor. The directional RFID card reader synchronously monitors the door area. When it senses the RFID signal of the positioning card, it generates RFID card reading event stream data containing the card number and the precise reading time and transmits the RFID card reading event stream data to the vehicle fusion processor.

[0029] In step S200, parallel feature extraction of multimodal data is performed. The vehicle-mounted fusion processor receives video stream data, uses a pre-set visual analysis model to perform frame-by-frame human target detection on the video stream data, and outputs visually estimated number of people data. At the same time, the vehicle-mounted fusion processor receives RFID card reading event stream data, uses a time window clustering algorithm to remove redundant reading data and extract unique identity identifiers, and generates RFID passenger list data containing all valid card numbers.

[0030] Step S300: Perform consistency verification calculation for multimodal data. The vehicle fusion processor obtains the value of the visually estimated number of people data and the element count value of the radio frequency passenger list data. The vehicle fusion processor calculates the absolute value of the difference between the visually estimated number of people data and the element count value of the radio frequency passenger list data, and determines whether the absolute value of the difference is within the preset allowable error range.

[0031] In step S400, a verification result and status data packet are generated. When the absolute value of the difference is within the allowable error range, the vehicle fusion processor determines that the statistical result is valid and uses the radio frequency passenger list data as the current passenger list. When the absolute value of the difference exceeds the allowable error range, the vehicle fusion processor determines the anomaly type according to the positive or negative sign of the difference and generates the corresponding anomaly classification identifier code. The vehicle fusion processor encapsulates the finally determined number of passengers, the radio frequency passenger list data, and the anomaly classification identifier code into a verification result data packet (including normal verification data packet or abnormal verification data packet).

[0032] In step S500, data upload and platform-side linkage response: the vehicle-mounted fusion processor sends the verification result data packet to the ground application layer through the data transmission layer. The safety monitoring platform parses the verification result data packet and writes it into the historical database. When the verification result data packet contains an abnormal classification identifier code or the displayed number of passengers exceeds the rated passenger capacity, the safety monitoring platform triggers the alarm logic, outputs an alarm signal on the large screen display system, and automatically retrieves the video stream data associated with the current moment for manual review.

[0033] The vehicle-mounted terminal establishes a visual perception domain inside the vehicle cabin through a panoramic camera.

[0034] The panoramic camera is fixed to the central area of ​​the ceiling inside the vehicle using a metal bracket and shock-absorbing pads. The optical axis of the panoramic camera lens points vertically to the floor of the vehicle. The installation position of the panoramic camera is kept at a horizontal distance from the door installation area to form an overhead view of the interior space of the vehicle.

[0035] The panoramic camera includes a wide-angle lens module or a fisheye lens module. The effective focal length of the wide-angle lens module or fisheye lens module is adapted according to the vertical height from the top surface of the carriage to the floor of the carriage, so that the field of view (FOV) of the panoramic camera is greater than or equal to 90 degrees. The coverage of the field of view coincides with the physical boundary of the carriage interior wall, forming a personnel activity area. The personnel activity area covers the seating area, the central aisle area, and the area inside the doors of the carriage.

[0036] The panoramic camera integrates an infrared illumination unit (IRLED) and an infrared cutoff filter switcher (IR-Cut). In low-light environments underground, the infrared illumination unit emits infrared auxiliary light into the personnel activity area, which, together with the infrared cutoff filter switcher, ensures image clarity. The top-down view layout and the cooperation of the infrared illumination unit are used to collect the head and shoulder contour features of the personnel to be detected in the personnel activity area.

[0037] The panoramic camera is physically connected to the vehicle-mounted fusion processor via shielded twisted-pair cable or coaxial cable. The panoramic camera converts the captured optical images into digital video signals and transmits them to the video input interface of the vehicle-mounted fusion processor via shielded twisted-pair cable or coaxial cable.

[0038] The directional RFID reader uses a combination of narrow beam antenna technology and signal strength filtering to construct an RFID electronic door curtain centered on the door frame plane in the door area.

[0039] The directional RFID reader is mounted on the top center of the door frame inside the vehicle door via a metal bracket. The antenna radiation surface of the directional RFID reader is vertically oriented towards the door entrance area on the vehicle floor. The installation angle of the directional RFID reader is parallel to the door plane to ensure that the main lobe of the radio frequency signal covers the door passage vertically downward.

[0040] The directional RFID reader is equipped with a low-gain directional flat panel antenna. The half-power beamwidth (HPBW) of the directional flat panel antenna is limited to within 60 degrees to reduce the spread of radio frequency signals into the seating area inside the carriage. The radio frequency transmission power of the directional RFID reader is set to a low-power mode, so that the effective reading distance is controlled within the vertical projection area of ​​the door frame and within 0.5 meters to 1.0 meters in front of and behind the door frame.

[0041] To further eliminate misreading caused by metal reflections inside the carriage, the directional RFID reader has a preset Received Signal Strength Indication (RSSI) filtering threshold. The directional RFID reader only recognizes a valid identification when the signal strength fed back by the positioning card is higher than the RSSI filtering threshold. The combination of the physical limitation of the narrow beam antenna and the logical limitation of the RSSI filtering threshold ensures that the card reading event is triggered only when the person to be tested wearing the positioning card passes through the door.

[0042] The directional RFID reader is physically connected to the vehicle-mounted fusion processor via a data communication cable. When the directional RFID reader detects a tag signal that meets the RSSI filtering threshold, it generates RFID reading event stream data containing a specific card number and a precise reading timestamp, and sends it to the vehicle-mounted fusion processor via the data communication cable.

[0043] Furthermore, the vehicle-mounted fusion processor adopts a fanless passive heat dissipation architecture design and is packaged inside a mining explosion-proof control box or a mining intrinsically safe control box.

[0044] The explosion-proof or intrinsically safe control box for mining is fixed to the equipment integration area inside the car body via a spring-damped shock-absorbing mounting bracket. The equipment integration area is located in the protected space between the side walls of the car body or under the seats. The outer shell of the explosion-proof or intrinsically safe control box is thermally coupled to the heat dissipation backplate of the on-board fusion processor via thermally conductive silicone grease, and the metal walls of the car body are used for auxiliary heat dissipation.

[0045] The in-vehicle fusion processor includes an embedded computing motherboard, a deep learning hardware acceleration unit, a high-precision real-time clock module (RTC), and a multi-protocol communication interface group at the hardware level. The embedded computing motherboard is equipped with an industrial-grade central processing unit and non-volatile storage media. The non-volatile storage media internally stores the embedded operating system, object detection algorithm model files, and system control programs. The deep learning hardware acceleration unit is connected to the central processing unit through a high-speed bus and is used to execute convolutional neural network operations on video stream data in parallel. The high-precision real-time clock module is equipped with an independent backup battery to provide a unified millisecond-level system time base for video stream data and RFID card reader event stream data.

[0046] The multi-protocol communication interface group includes a video input interface, a sensor data interface, and an uplink communication interface. The video input interface uses an M12 shockproof aviation plug to connect the video data transmission cable of the panoramic camera. The sensor data interface uses an opto-isolated RS485 interface to physically connect to the directional RFID reader. The uplink communication interface is used to access the mining industrial ring network or wireless private network.

[0047] The vehicle-mounted fusion processor integrates a wide-voltage isolated power supply module, which has surge and overvoltage protection functions. Its input is connected to the vehicle's main power system, and its output provides stable DC operating voltage to the vehicle-mounted fusion processor, panoramic camera, and directional RFID reader.

[0048] Furthermore, the vehicle-mounted fusion processor invokes the visual analysis algorithm module residing in its non-volatile storage medium to process the video stream data from the panoramic camera.

[0049] The visual analysis algorithm module first performs image preprocessing. It extracts original video frame images from the video stream data in a time-series manner. To address the wide-angle distortion caused by the panoramic camera, the module uses a preset camera intrinsic parameter matrix to perform geometric distortion correction on the original video frame images, unfolding them into distortion-free planar projection images. Subsequently, the module applies the Limit Contrast Adaptive Histogram Equalization (CLAHE) algorithm to the corrected images to enhance texture details in dark areas and generate enhanced grayscale images.

[0050] The visual analysis algorithm module performs tensor transformation on the enhanced grayscale image. It scales the enhanced grayscale image to the input resolution set by the lightweight convolutional neural network model (e.g., 640×640 pixels) and normalizes the pixel values ​​to generate tensor data that meets the model's input requirements.

[0051] The visual analysis algorithm module inputs tensor data into a lightweight convolutional neural network model. The lightweight convolutional neural network model uses CSPDarknet or MobileNet as the feature extraction backbone to adapt to the edge computing resources of the vehicle fusion processor. The lightweight convolutional neural network model performs convolution operations on the tensor data and outputs a raw detection result set containing multiple candidate targets.

[0052] The original detection result set includes the center point coordinates, width and height dimensions, class probability, and confidence score of each candidate target. The visual analysis algorithm module performs non-maximum suppression (NMS) processing on the original detection result set, eliminating targets with confidence scores below a preset threshold. The candidate targets are identified, and the intersection-over-union ratio (IoU) between the remaining candidate targets is calculated. When the IoU of two candidate targets exceeds the overlap threshold, the threshold is determined. At that time, the visual analysis algorithm module retains candidate targets with higher confidence scores and generates a list of valid targets.

[0053] The visual analysis algorithm module calculates the instantaneous number of people in the current frame based on the list of valid targets. The number of targets in the list of valid targets is recorded as the number of people at the current moment. Number of visual inspection personnel To eliminate count jumps caused by video jitter, the visual analysis algorithm module introduces a length of... The sliding time window smoothing filter, the visual analysis algorithm module calculates the nearest... Frames The arithmetic mean of the numbers is taken and rounded down to obtain the smoothed visual estimate of the number of people (denoted as ). ).

[0054] Furthermore, the vehicle-mounted fusion processor invokes the radio frequency signal analysis program residing in its non-volatile storage medium to process the radio frequency card reading event stream data from the directional radio frequency identification card reader.

[0055] The radio frequency signal analysis program uses a dynamic time window clustering algorithm to generate a determined occupant list. The vehicle fusion processor responds to the door opening signal and starts the radio frequency data receiving process. During the time the door remains open and the subsequent preset delay time (e.g., within 30 seconds after the door closes), the vehicle fusion processor continuously buffers the radio frequency card reading event stream data.

[0056] The RFID card reader event stream data consists of multiple discrete card reader tuples, and the data structure of each card reader tuple is defined as follows: ,in This is a unique identifier for the electronic tag. For the time of reading, This represents the signal strength value.

[0057] The vehicle-mounted fusion processor performs cluster analysis at the end of the receiving process. The vehicle-mounted fusion processor first... The algorithm groups all cached read tuples into multiple single-card time series. For each single-card time series, the algorithm... Sort the sequence in ascending order and calculate the time interval between two adjacent tuples. .

[0058] The in-vehicle fusion processor uses time discreteness discrimination logic, and the system presets a maximum signal interruption threshold. (For example, 2 seconds), if there is a single card time series At the breakpoint, the vehicle fusion processor divides the sequence into multiple independent access event clusters, each representing an independent physical behavior of a person passing through the vehicle door area.

[0059] The vehicle-mounted fusion processor performs spatiotemporal feature validity filtering on the segmented traffic event clusters, and the system presets a minimum number of valid reads. and minimum passage duration .

[0060] When a certain pass event cluster contains less than Or the time span of its first and last tuples is less than At that time, the vehicle-mounted fusion processor identifies it as "passing noise" or "edge reflection signal" and removes it.

[0061] When the time span of a certain passage event cluster exceeds the preset lingering alarm threshold At a time (e.g., 10 seconds), the onboard fusion processor marks it as a "doorway loitering event," but still retains the data. A status notification indicating a standby condition is also generated in the temporary list.

[0062] The in-vehicle fusion processor counts all unique entries that pass the validity filter. The number of in-vehicle fusion processors will ultimately be retained. The collection is constructed into a list of radio frequency passengers (denoted as ). ), and use the base of the set as the number of people counted by radio frequency. The output is then sent to the subsequent fusion verification module.

[0063] Furthermore, the in-vehicle fusion processor performs spatiotemporal synchronization and alignment of the data stream through the following steps: The in-vehicle fusion processor maintains a unified system time base using an internally integrated high-precision real-time clock module. When the underlying driver receives raw video frames transmitted from the panoramic camera, the in-vehicle fusion processor captures the hardware interrupt moment and writes it as a timestamp to the video frame header, forming a video frame sequence. Simultaneously, when the vehicle-mounted fusion processor responds to a card reading interruption by the directional RFID card reader, it records the interruption time and associates it with the RFID card reading event stream data to form a discrete event sequence. .

[0064] The in-vehicle fusion processor performs time alignment operations based on buffer queues, due to the video frame sequence. The generation process suffers from algorithm inference latency. The onboard fusion processor opens two circular buffers, one for temporarily storing the processed visually estimated number of people and the other for storing the real-time updated radio frequency passenger list data.

[0065] Vehicle-mounted fusion processor sets verification trigger cycle (For example, 200 milliseconds) At each verification trigger moment, the onboard fusion processor reads the timestamp of the latest processed visual data frame. Subsequently, the vehicle-mounted fusion processor performs a backtracking search in the buffer of the radio frequency passenger list data, looking for timestamps less than or equal to... And closest RF list status, The vehicle-mounted fusion processor uses a nearest neighbor matching strategy or a zero-order hold strategy to lock the visual detection result at that moment and the corresponding RF list state into a data pair at the same physical moment.

[0066] The in-vehicle fusion processor performs logical associations in the spatial dimension. The perception area of ​​the panoramic camera corresponds to the current number of people inside the vehicle, and the perception area of ​​the directional RFID card reader corresponds to the cumulative number of people who have entered through the vehicle door.

[0067] At any alignment point during vehicle operation, the onboard fusion processor constructs a fusion verification feature vector. The feature vector contains: at the same time Visual instantaneous estimation of the number of people And the number of unique card numbers included in the radio frequency passenger list up to that time. , fuse verification feature vector This serves as input data for subsequent consistency decision logic.

[0068] Furthermore, the in-vehicle fusion processor uses a high-precision real-time clock module to achieve spatiotemporal synchronization of video stream data and RFID card reader event stream data, and constructs the current time... fusion verification feature vector Afterwards, the vehicle-mounted fusion processor enters the consistency verification stage, which is used to determine whether the visually estimated number of people data and the radio frequency passenger list data are consistent in terms of quantity statistics.

[0069] The in-vehicle fusion processor obtains fusion verification feature vectors Parse the time-aligned visually estimated number of people data (denoted as ). ) and the number of unique electronic tags contained in the radio frequency passenger list data (denoted as ) The absolute value of the difference between the number of people estimated by the vehicle-mounted fusion processor's computer vision and the number of unique electronic tags. (Right now ).

[0070] The in-vehicle fusion processor reads the tolerance threshold from the system configuration file on its non-volatile storage medium. Tolerance threshold It is a non-negative integer set based on the detection accuracy error of the panoramic camera and the rated passenger capacity of the carriage (e.g., setting...). or Tolerance threshold It is used to filter visual count fluctuations caused by overlapping or obstruction of people in the carriage or instantaneous changes in posture.

[0071] The absolute value of the difference between the in-vehicle fusion processors With tolerance threshold ,when At that time, the vehicle-mounted fusion processor determines that the multimodal data at the current moment meets the consistency requirements. The determination result shows that the number of entities observed by the panoramic camera and the number of tags recorded by the directional RFID card reader are within the allowable error range, that is, the actual number of people in the carriage matches the number of people boarding the vehicle recorded by the card reader system.

[0072] Under conditions that meet consistency requirements, the vehicle-mounted fusion processor performs data acceptance based on data dimension priority. Since the RFID passenger list data contains definite personal identification numbers (UIDs), its data dimension is higher than that of visually estimated passenger data that only contains statistical counts. Therefore, the vehicle-mounted fusion processor confirms the RFID passenger list data as the final valid passenger list. The vehicle-mounted fusion processor then counts the unique electronic tags... This serves as the current precise output of passenger numbers.

[0073] The vehicle-mounted fusion processor generates a consistency status identifier (Status_OK). The vehicle-mounted fusion processor encapsulates the accurate number of passengers, the final valid passenger list, the current timestamp information, and the consistency status identifier into a verification result data packet, and stores it in the sending queue to wait for uploading.

[0074] Furthermore, when the onboard fusion processor determines the absolute value of the difference between the visually estimated number of people data and the radio frequency passenger list data... Greater than the tolerance threshold At that time, the vehicle-mounted fusion processor triggers the anomaly attribution and handling procedure.

[0075] The in-vehicle fusion processor calculates the difference direction indicator. , Scenario 1: Handling anomalies due to someone not having a card ( .

[0076] when At this time, the vehicle fusion processor determines that the "visual count overflow" state is present, which indicates that there are physical entities in the vehicle that have not been identified by the radio frequency system.

[0077] The vehicle fusion processor executes an output strategy based on the maximum safety value. To prevent potential overloading risks, the vehicle fusion processor confirms the visually estimated number of people with larger values ​​as the current actual number of passengers.

[0078] Simultaneously, the vehicle-mounted fusion processor generates a "personnel missing location card" alarm. The vehicle-mounted fusion processor calls an image annotation algorithm to highlight all detected human target boxes in the video frame at the current moment with a color different from normal targets (such as red). The vehicle-mounted fusion processor packages the visually estimated number of people data, radio frequency passenger list data, and video frames with highlighted marks as the first type of abnormal data packet.

[0079] Scenario 2: Handling anomalies where card is present but no one is present ( .

[0080] when At this time, the vehicle fusion processor determines that the "RF count overflow" state is in effect, which indicates that the RF passenger list data contains redundant electronic tags that do not have a corresponding physical entity.

[0081] The vehicle-mounted fusion processor initiates redundancy screening logic based on signal characteristics. The vehicle-mounted fusion processor analyzes the time series variance of the Received Signal Strength Indication (RSSI) of each electronic tag in the radio frequency passenger list data. The vehicle-mounted fusion processor marks tags with RSSI variance less than a preset stationary threshold (indicating that the tag has not moved with the human body) or whose reading timestamp is earlier than the current boarding window as "suspected redundant tags".

[0082] The vehicle-mounted fusion processor executes a dual output strategy: in the people count field, it outputs visually estimated people data to reflect the physical occupancy situation; in the personnel list field, it retains all card numbers in the radio frequency passenger list data, but adds an "abnormal verification" suffix to suspected redundant tags. This strategy is used to prevent the omission of underground personnel attendance due to visual missed detection, and at the same time, it alerts to possible "card-carrying for others" behavior.

[0083] The vehicle-mounted fusion processor uploads the first type of abnormal data packets or the second type of abnormal data packets containing tag information to the safety monitoring platform with the highest priority through the data transmission layer. The safety monitoring platform parses the abnormality type and links the on-site broadcast system to play the corresponding voice prompts (such as "Please check the positioning card" or "It is strictly forbidden to carry the card for others").

[0084] During the operation of the accurate passenger count and identity recognition system in the underground coal mine passenger compartment, the on-board fusion processor executes the equipment status monitoring subroutine in parallel. The equipment status monitoring subroutine is used to monitor the physical connection status of the panoramic camera and the data validity of the video stream data in real time, and executes automatic degradation and fallback logic when a fault is detected.

[0085] The vehicle-mounted fusion processor performs dual fault detection for video stream data. The first detection is hardware link integrity detection. The vehicle-mounted fusion processor monitors for interruptions in the field synchronization signal (VSYNC) or line synchronization signal (HSYNC) of the video input interface. If no new hardware interruption signal is captured within a preset timeout period (e.g., 500 milliseconds), the vehicle-mounted fusion processor determines that a video signal loss fault has occurred. The second detection is image content logic detection. The vehicle-mounted fusion processor calculates the grayscale histogram and inter-frame difference coefficients of the current video frame. When the grayscale histogram shows that the pixel values ​​are concentrated in the extremely dark or extremely bright range (indicating that the lens is blocked or the exposure is abnormal), or when the inter-frame difference coefficients between multiple consecutive frames are lower than the preset noise floor threshold (indicating that the sensor memory is frozen), the vehicle-mounted fusion processor determines that a sensor logic crash fault has occurred.

[0086] In response to the detection of any of the above fault states, the vehicle fusion processor first attempts to send a power reset command to the panoramic camera. If the fault still exists after the reset, the vehicle fusion processor switches the system operating mode to the radio frequency-dominated fallback mode.

[0087] In the RF-dominated fallback mode, the vehicle fusion processor suspends the operation of the visual analysis algorithm module, stops the execution of consistency verification based on the absolute value of the difference, and establishes a single-channel data link containing only the directional RF identification card reader.

[0088] The vehicle-mounted fusion processor executes a compensatory output strategy based on single-mode data. The vehicle-mounted fusion processor reads the current radio frequency passenger list data and directly uses the number of unique electronic tags contained in the radio frequency passenger list data as the estimated number of passengers. At the same time, the vehicle-mounted fusion processor generates a visual blocking fault code, which contains a specific fault type identifier (signal loss, obstruction, or system crash).

[0089] The vehicle-mounted fusion processor encapsulates the estimated number of passengers, radio frequency passenger list data, and visual blocking fault codes into a degraded status data packet. The vehicle-mounted fusion processor uploads the degraded status data packet through the data transmission layer. After receiving the data, the safety monitoring platform blocks the corresponding video window on the large screen display system and overlays a graphic warning of "visual data unavailable" to remind dispatchers that the current statistical results are based only on radio frequency identification technology.

[0090] In the ground application layer of the accurate passenger count and identity recognition system for underground coal mine passenger carriages, the safety monitoring platform, as the core of data aggregation and interaction, is deployed on a ground server cluster. The safety monitoring platform receives verification result data packets, abnormal verification data packets, or degraded status data packets uploaded from various vehicle terminals in real time through the data transmission layer.

[0091] The safety monitoring platform has a pre-installed vehicle configuration database, which stores the unique number of each car, the rated passenger capacity threshold, and the fleet formation relationship. In order to achieve visual positioning, the safety monitoring platform establishes communication with the mine personnel / vehicle positioning system server through a standard data interface. The safety monitoring platform uses the unique number of each car as an index key to obtain the absolute coordinate data of each car in the underground roadway in real time, and performs spatiotemporal fusion association between the absolute coordinate data and the verification result data packet.

[0092] The safety monitoring platform drives the large screen display system to display the comprehensive monitoring interface. The comprehensive monitoring interface uses the mine's electronic map as the base map and dynamically refreshes the position icons of each car based on absolute coordinate data. The safety monitoring platform executes a hierarchical visualization strategy based on the status codes in the data packets: when the status is "consistent (Status_OK)" and the number of passengers is not exceeded, the safety monitoring platform displays the position icon in green normal state and overlays the "actual load / capacity" value label next to the icon.

[0093] When the number of passengers detected exceeds the rated passenger capacity threshold, the safety monitoring platform triggers a Level 1 overload alarm. The platform controls the corresponding carriage's location icon to flash red at a high frequency and pops up a forced confirmation window in the center of the interface. The forced confirmation window displays the number of passengers exceeding the limit and the current list of radio-frequency personnel. At the same time, the safety monitoring platform sends a voice synthesis command to the underground broadcasting system, driving the loudspeakers in the area where the carriage is located to play an overload warning sound.

[0094] When an abnormal verification data packet is received, the safety monitoring platform triggers a level-two abnormal linkage. For the "personnel missing positioning card" abnormality, the safety monitoring platform automatically retrieves the associated video stream of the carriage and displays the video frames with highlighted marks in the sidebar. For the "multiple cards or cross-reading" abnormality, the safety monitoring platform marks the ID number of the suspected redundant tag in the personnel list with a special color (such as yellow). The safety monitoring platform provides a manual confirmation interface, and the corresponding alarm flashing state is cleared only after the dispatcher performs the "aware" or "handled" operation.

[0095] When a downgraded status data packet is received, the safety monitoring platform triggers a Level 3 equipment fault alert. The platform updates the location icon of the corresponding carriage to a gray or yellow fault state and overlays a "visual loss" graphic label on top of the icon. The platform cuts off the real-time video preview stream of the carriage to save bandwidth, only displays the number of people estimated based on radio frequency identification, and notes "low confidence data" in the data panel. All alarm events and handling records are written to the operation log database for safety auditing.

[0096] In the ground application layer of the accurate passenger count and identification system for underground coal mine passenger carriages, the safety monitoring platform executes historical data tracing and emergency assistance logic based on various data packets uploaded from the data transmission layer. This logic is used to establish a traceable passenger record database and provide real-time personnel information query services when an emergency situation is triggered.

[0097] The security monitoring platform is equipped with a high-capacity historical database server. The platform performs structured parsing on the received verification result data packets, abnormal verification data packets, and degradation status data packets, and stores them in the historical database server according to the timestamp sequence. Each stored record contains a unique carriage number, accurate passenger number, personnel identity list (UID), video keyframes, and verification status identifier. The historical database server constructs a multi-dimensional index structure including UID index, carriage number index, and time index to support high-concurrency data retrieval requests.

[0098] The safety monitoring platform provides a reverse lookup function for passenger travel routes. In response to the query command of a specific person's electronic tag ID input by the dispatch and command center, the safety monitoring platform retrieves all historical passenger travel records associated with that ID from the historical database server. The safety monitoring platform sorts the search results in ascending order by timestamp and generates a passenger travel route report. The passenger travel route report includes the carriage number, boarding and alighting times, and corresponding video screenshots of the passenger at different time periods, which are used for work attendance verification or determination of illegal passenger travel behavior.

[0099] The safety monitoring platform provides a playback function of the transport history for specific carriages. In response to the input carriage number and time period query command, the safety monitoring platform retrieves the continuous monitoring data of the carriage within the specified time period. The safety monitoring platform plays back the internal video recording of the carriage and the dynamic personnel list at the corresponding time on the large screen display system in a time-axis synchronized manner.

[0100] The safety monitoring platform integrates emergency rescue auxiliary logic. It continuously monitors the status data of third-party underground environmental monitoring systems or vehicle operation monitoring systems through standard industrial data interfaces (such as OPCUA or WebAPI). When it receives an accident trigger signal (such as a vehicle derailment alarm or gas over-limit alarm) or an emergency rescue instruction issued by the dispatch and command center, the safety monitoring platform immediately executes the accident data protection procedure.

[0101] The safety monitoring platform performs the last known status extraction operation. It retrieves the last verification result data packet uploaded by the accident-related carriage before the communication interruption or the accident occurred from the historical database server. The safety monitoring platform performs a write protection operation on the data packet to prevent it from being overwritten by the loop recording mechanism. The safety monitoring platform parses the personnel identity list in the data packet and generates an emergency rescue list containing the names, employee numbers, work groups, and last known locations of the personnel in distress.

[0102] The safety monitoring platform uses a private wireless network to automatically send the emergency rescue list and the last video screenshot of the accident compartment to the command screen in the dispatch and command center and the handheld terminals of the rescue team. The emergency rescue auxiliary logic uses the latest snapshot data stored on the ground server to provide accurate information on the number and identity of the people in the vehicle for rescue decision-making, eliminating blind spots in the rescue.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for accurate passenger counting and identification in underground coal mine passenger carriages, characterized in that, The method is applied to an in-vehicle fusion processor and specifically includes the following steps: Step S100: Start data synchronization acquisition for a single trip: When the carriage enters the operating cycle or the door is open, control the panoramic camera to acquire real-time video stream data, and control the directional RFID card reader to generate RFID card reading event stream data when it senses the RFID signal of the positioning card; receive the real-time video stream data and the RFID card reading event stream data; Step S200: Parallel feature extraction of multimodal data: Frame-by-frame human target detection is performed on the video stream data using a pre-set visual analysis model to output visually estimated number of people; simultaneously, the radio frequency card reading event stream data is processed using a time window clustering algorithm to remove redundant reading data and extract unique identification identifiers to generate radio frequency passenger list data. Step S300: Perform consistency verification calculation for multimodal data: obtain the numerical value of the visually estimated number of people data and the element count value of the radio frequency passenger list data, calculate the absolute value of the difference between the two, and determine whether the absolute value of the difference is within the preset allowable error range. Step S400: Generate a verification result data packet: When the absolute value of the difference is within the allowable error range, the statistical result is determined to be valid, and the radio frequency passenger list data is used as the current passenger list; when the absolute value of the difference exceeds the allowable error range, the anomaly type is determined according to the positive or negative sign of the difference and an anomaly classification identifier code is generated; the final determined number of passengers, the radio frequency passenger list data, and the anomaly classification identifier code are encapsulated into a verification result data packet. Step S500, Data Upload and Platform-Side Linkage Response: The verification result data packet is sent to the ground application layer through the data transmission layer; the ground application layer parses the verification result data packet and triggers alarm logic when it contains the abnormal classification identifier code or the number of passengers exceeds the rated capacity.

2. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: Step S200, which uses a preset visual analysis model to output visually estimated number of people, specifically includes: The original video frame image is extracted from the video stream data, and geometric distortion correction and contrast-limited adaptive histogram equalization are performed on the original video frame image to generate an enhanced grayscale image. After scaling and normalizing the enhanced grayscale image, it is input into a target detection network using a lightweight convolutional neural network model, and outputs an original detection result set containing the confidence scores of candidate targets; non-maximum suppression processing is performed on the original detection result set, and a list of valid targets is generated according to a preset confidence threshold and an overlap threshold; The number of targets in the list of valid targets at the current time t is counted as the number of visually detected people. The arithmetic mean of the number of visually detected people in the most recent k frames is calculated using a sliding time window smoothing filter of length k, and the smoothed estimated number of visually detected people data is obtained.

3. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: The step S200 of generating the radio frequency passenger list data specifically includes the following steps: During the time the door remains open and the subsequent preset delay period, the RF card reader event stream data is continuously cached, and the RF card reader event stream data consists of multiple discrete card reader tuples; The cached RFID card reader event stream data is grouped according to the unique identification code of the electronic tag to form multiple single card time series, and the time interval between two adjacent card reader tuples in the series is calculated. If there is a breakpoint in the single card time series where the time interval is greater than the preset maximum signal interruption threshold, the series is divided into multiple independent passage event clusters; Perform validity filtering on the segmented access event clusters: remove access event clusters containing fewer than the minimum number of valid reads or with a time span shorter than the minimum access duration; A cluster of passage events whose time span exceeds a preset lingering alarm threshold is designated as a lingering event at the entrance. The unique identification codes of all electronic tags that pass the validity filter are counted to construct the radio frequency passenger list data.

4. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: Prior to step S300, a step of performing spatiotemporal synchronization and alignment of the data stream is also included: A high-precision real-time clock module is used to maintain a unified system time base, and acquisition timestamps are applied to the video stream data and the RFID card reader event stream data, respectively. When the vehicle-mounted fusion processor receives the original video frame transmitted by the panoramic camera in the underlying driver, it captures the hardware interrupt moment and writes it as the acquisition timestamp into the video frame header to form a video frame sequence. Establish a visual data buffer for temporarily storing the visually estimated number of people data and an RF data buffer for temporarily storing the RF passenger list data; At each verification trigger moment, the acquisition timestamp of the latest processed video frame is read, and a backtracking search is performed in the radio frequency data buffer to find the radio frequency list state that is closest to the acquisition timestamp of the video frame that is less than or equal to the acquisition timestamp of the video frame. By employing a nearest neighbor matching strategy or a zero-order hold strategy, the visual detection result at that moment and the corresponding radio frequency list state are locked as a data pair at the same physical moment.

5. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: The processing in step S400 when the absolute value of the difference is within the allowable error range specifically includes: Read the tolerance threshold set based on the detection accuracy error of the panoramic camera and the rated passenger capacity of the carriage; If the absolute value of the difference is less than or equal to the tolerance threshold, it is determined that the multimodal data at the current moment meets the consistency requirements. Data acceptance is performed based on data dimension priority: the radio frequency passenger list data containing the identity of the person is confirmed as the final valid passenger list, and the number of elements in the radio frequency passenger list data is output as the current accurate number of passengers.

6. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: The processing in step S400 when the absolute value of the difference exceeds the allowable error range specifically includes: Calculate the direction indicator of the difference between the visually estimated number of people data and the number of elements in the radio frequency passenger list data; If the difference direction indicator is greater than the tolerance threshold, it is determined to be a visual count overflow state, and an output strategy based on the maximum safety value is executed: the visually estimated number of people with the larger value is confirmed as the current actual number of passengers, and a missing person location card alarm indicator is generated. If the difference direction indicator is less than the negative tolerance threshold, it is determined to be an RF count overflow state, and the redundancy screening logic based on signal characteristics is initiated: analyze the time series variance of the received signal strength indication of each electronic tag in the RF passenger list data, and mark tags with variance less than the preset static threshold or abnormal reading timestamps as suspected redundant tags. A dual output strategy is implemented: the visually estimated number of people is output to reflect the physical occupancy situation, while all card numbers in the radio frequency passenger list data are retained and an anomaly verification suffix is ​​added to the suspected redundant tags.

7. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: The method also includes equipment status monitoring and automatic degradation steps: The device status monitoring subroutine is executed in parallel to monitor the hardware interrupt signal of the video input interface and calculate the grayscale histogram and inter-frame difference coefficient of the current video frame. When a video signal loss fault or sensor logic crash is detected, an attempt is made to send a power reset command to the panoramic camera. If the fault persists after resetting, the system operating mode will be switched to RF-dominated fallback mode: the operation of the visual analysis algorithm module will be suspended, the consistency check will be stopped, the number of unique electronic tags contained in the RF passenger list data will be used as the estimated number of passengers, and a visual blocking fault code will be generated, encapsulated in a degraded state data packet, and uploaded.

8. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: The processing of the ground application layer in step S500 specifically includes: Communication is established with the mine personnel / vehicle positioning system server through a standard data interface, and the absolute coordinate data of each car in the underground roadway is obtained in real time using the unique number of the car as the index key. The absolute coordinate data and the verification result data packet are spatiotemporally fused and correlated to drive the large screen display system to dynamically refresh the carriage location icon on the mine electronic map; When the abnormal classification code is received or overcrowding is detected, the position icon of the corresponding carriage is turned to flash, and a voice synthesis command is sent to the underground broadcasting system.

9. The method for accurate passenger counting and identification in underground coal mine passenger carriages according to claim 1, characterized in that: Step S500 also includes emergency rescue auxiliary processing: The ground application layer continuously monitors the status data of the third-party downhole environmental monitoring system or vehicle operation monitoring system; when an accident trigger signal or emergency rescue instruction is received, the verification result data packet last uploaded by the accident-related carriage is retrieved from the historical database server; Write protection is applied to the verification result data packet to prevent it from being overwritten by the loop recording mechanism. The personnel identity list within the packet is then parsed to generate an emergency rescue list, which is then sent to the rescue team's handheld terminal via a private wireless network message push service.

10. A system for accurate passenger counting and identification in underground coal mine passenger carriages, characterized in that, The system includes: The underground sensing layer includes an on-board terminal installed in the underground passenger compartment, the on-board terminal including a panoramic camera, a directional radio frequency identification card reader and an on-board fusion processor; The panoramic camera is used to collect real-time video stream data inside the carriage; the directional RFID card reader is used to generate card reading event data containing card number and timestamp. The vehicle-mounted fusion processor is connected to the panoramic camera and the directional radio frequency identification card reader respectively, and is configured to perform the method for accurate number of passengers and identity recognition in underground coal mine passenger compartments as described in any one of claims 1 to 9; The data transmission layer, including a mining industrial ring network or a wireless private network, is connected to the vehicle-mounted fusion processor and is used to upload the verification result data packet; The ground application layer includes a security monitoring platform, a dispatch and command center, and a large-screen display system. The security monitoring platform is connected to the data transmission layer and is used to parse the verification result data packets and perform overload judgment and alarm logic.

Images