An all-round intelligent anti-collision system and method for underground trackless equipment

Abstract

The present invention provides an all-round intelligent anti-collision system and method for underground trackless equipment, which relates to the field of anti-collision technology for underground trackless equipment. The system includes: an edge computing processing unit, a signal input and output unit, a handheld positioning card, a ranging radar unit, an infrared camera fill light unit and a 360° omnidirectional positioning base station unit; the edge computing processing unit is respectively connected to the signal input and output unit, the ranging radar unit, the infrared camera fill light unit and the 360° omnidirectional positioning base station unit, the handheld positioning card transmits a narrow pulse signal to the 360° omnidirectional positioning base station unit, and the handheld positioning card and the 360° omnidirectional positioning base station unit transmit communication signals to each other. The technical solution of the present invention overcomes the problem that the prior art cannot accurately perceive approaching personnel or fixed lanes, which easily causes misjudgment and misjudgment.

Description

Technical Field

[0001] The present invention relates to the technical field of anti-collision of underground trackless equipment, and in particular to an all-round intelligent anti-collision system and method for underground trackless equipment. Background Art

[0002] In underground operations, underground trackless transportation equipment plays a vital role as an important means of transportation. However, the underground working environment is complex and changeable, and there are many safety hazards, among which vehicle collision accidents are one of the more common safety problems. Although the underground trackless vehicle transportation system has the advantages of improving work efficiency and enhancing safety, there are great safety hazards in actual application, which can easily cause various safety accidents. The underground working space is narrow, the tunnels are crisscrossed, and the line of sight is seriously obstructed. Trackless vehicles can easily collide with tunnel walls, other vehicles, staff, etc. during driving, which will not only cause equipment damage and property loss, but also may cause casualties, bringing huge economic losses and negative impacts to mining companies. In addition, traditional anti-collision measures mainly rely on the driver's experience and attention, but they have great limitations and cannot effectively cope with the complex and changeable underground environment.

[0003] At present, the anti-collision method for underground trackless transportation equipment uses laser radar to sense the surrounding environment of the vehicle, and then sends alarm signals to locomotives and other transportation equipment. This method has difficulty in sensing approaching personnel and fixed lanes, and is prone to misjudgment and miscalculation.

[0004] Therefore, there is a need for an all-round intelligent anti-collision system and method for underground trackless equipment that can accurately sense approaching personnel or fixed lanes. Summary of the invention

[0005] The main purpose of the present invention is to provide an all-round intelligent anti-collision system and method for underground trackless equipment to solve the problem that the prior art cannot accurately sense approaching personnel or fixed lanes, which easily leads to misjudgment and misjudgment.

[0006] To achieve the above-mentioned purpose, the present invention provides an all-round intelligent collision avoidance system for underground trackless equipment, comprising: an edge computing processing unit, a signal input and output unit, a handheld positioning card, a ranging radar unit, an infrared camera fill light unit and a 360° omnidirectional positioning base station unit; the edge computing processing unit is respectively connected to the signal input and output unit, the ranging radar unit, the infrared camera fill light unit and the 360° omnidirectional positioning base station unit, the handheld positioning card transmits a narrow pulse signal to the 360° omnidirectional positioning base station unit, and the handheld positioning card and the 360° omnidirectional positioning base station unit transmit communication signals to each other.

[0007] The present invention also provides an all-round intelligent anti-collision method for underground trackless equipment, which specifically includes the following steps:

[0008] S1, the system is powered on and acquires millimeter wave radar signals, visual signals and UWB signals.

[0009] S2, the edge computing processing unit continuously receives millimeter-wave radar signals and converts them into 485 signals for distance conversion; when the distance is less than the set value, the vehicle environment detection algorithm is used to detect the front and rear of the vehicle, and the system issues an alarm or controls the vehicle to make an emergency stop based on the detection results.

[0010] S3, using the UWB signal to calculate the position information between the base station and the handheld positioning card, and sending an alarm message according to the position information.

[0011] S4, when the edge computing processing unit reads the visual signal, it identifies the driver's illegal operation through the driver behavior detection algorithm, and issues an alarm or stops the vehicle for the illegal operation.

[0012] Furthermore, the vehicle environment detection algorithm in step S2 includes the following steps:

[0013] S2.1, the system obtains a video stream as input data and preprocesses the video stream data, including: adjusting image size, pixel normalization and format conversion.

[0014] S2.2, the video stream obtained after preprocessing is input into the improved convolutional neural network (CNN) module for feature extraction, multi-scale feature map fusion and target detection.

[0015] S2.3, determine whether pedestrians and turn signs are detected.

[0016] S2.4, if a pedestrian is detected, the real-time detection data of the millimeter-wave radar is immediately read for auxiliary judgment. When the radar data is less than 25 meters, the system alarm is activated; if a turn sign is detected, it means that the vehicle is about to enter the turning state. At this time, the parking function for obstacles less than 0.5 meters determined by the millimeter-wave radar detection distance is temporarily cancelled to ensure that the vehicle can pass through narrow corners smoothly.

[0017] Furthermore, the driver behavior detection algorithm in step S4 includes the following steps:

[0018] S4.1, the system obtains a video stream as input data, and preprocesses the video stream data, including: adjusting image size, pixel normalization, and format conversion.

[0019] S4.2, the video stream obtained after preprocessing is input into the improved convolutional neural network CNN for target detection.

[0020] S4.3, classify the detected targets and distinguish between normal behavior and abnormal behavior.

[0021] S4.4, for abnormal behavior, activate system alarm or brake to stop.

[0022] Furthermore, the improved convolutional neural network CNN in step S2.2 and step S4.2 includes: a backbone network, a connection part and a task head.

[0023] The backbone network includes: an input layer, a first convolutional layer, a second convolutional layer, a first general inverted bottleneck module, a third convolutional layer, a second general inverted bottleneck module, a fourth convolutional layer, a third general inverted bottleneck module, a fifth convolutional layer, a fourth general inverted bottleneck module and a spatial pyramid pooling layer, which are connected in sequence.

[0024] The connection part includes: a sixth convolutional layer, a first adaptive multi-scale sampling layer, a first connection layer, a fifth universal inverted bottleneck module, a seventh convolutional layer, a second adaptive multi-scale sampling layer, a second connection layer, a sixth universal inverted bottleneck module, an eighth convolutional layer, a third connection layer, a seventh universal inverted bottleneck module, a ninth convolutional layer, a fourth connection layer and an eighth universal inverted bottleneck module, which are connected in sequence.

[0025] The task header includes: detection layer.

[0026] Among them, the second universal inverted bottleneck module is also connected to the second connection layer, the third universal inverted bottleneck module is also connected to the first connection layer, the spatial pyramid pooling layer is also connected to the sixth convolutional layer, and the sixth convolutional layer is also connected to the eighth universal inverted bottleneck module, the seventh convolutional layer is also connected to the third convolutional layer, and the sixth universal inverted bottleneck module, the seventh universal inverted bottleneck module and the eighth universal inverted bottleneck module are respectively connected to the detection layer.

[0027] Furthermore, the general inverted bottleneck module in the improved convolutional neural network CNN includes two paths, the first path is a standard convolutional layer, and the second path introduces a deep convolutional layer based on the standard convolutional layer:

[0028] The first standard convolution is recorded as:

[0029] ;

[0030] in, is the feature obtained after the first convolution. is an input feature of size H×W×C; H, W, C represent the height, width, and number of channels respectively. is a convolution kernel of size k×k×C, The height and width of the convolution kernel are both k; is the bias term; is the ReLU activation function; Represents a standard convolution operation.

[0031] The second path introduces a deep convolutional layer based on the standard convolutional layer, which is denoted as:

[0032] ;

[0033] in, is the feature obtained after the second convolution. is the dilated convolution, ; Is the size of The convolution kernel of is the bias vector, Represents channel-wise convolution.

[0034] Concatenate the features of the first and second paths to obtain the concatenated features. :

[0035] ;

[0036] in, Represents a join operation.

[0037] Finally, the concatenated features are convolved to obtain the final output :

[0038] ;

[0039] in, The convolution kernel for convolution processing of the splicing features, To help improve the convolutional neural network model, adjust the bias term of the channel output.

[0040] Furthermore, the adaptive multi-scale sampling layer in the improved convolutional neural network CNN first downsamples the input features by different multiples to generate a multi-scale feature map:

[0041] ;

[0042] ;

[0043] ;

[0044] in is the input feature; Indicates that the features are proportional Zoom out.

[0045] The adaptive multi-scale sampling layer then calculates the complexity weight for each downsampled feature:

[0046] ;

[0047] in, is the complexity weight, is the complexity calculation function, include .

[0048] According to the complexity weight, features of different scales are adaptively upsampled:

[0049] ;

[0050] in, Upsampling the features, To adaptively adjust the upsampling ratio according to the weight, is the feature after adaptive upsampling.

[0051] Finally, the upsampled features of different scales are weighted and fused to obtain the final output :

[0052] ;

[0053] in, Take 1, 2, 3.

[0054] Furthermore, step S2 specifically includes the following steps:

[0055] S2.1, the edge computing processing unit continuously receives millimeter-wave radar signals, converts the millimeter-wave radar signals into 485 signals for distance conversion, and analyzes the converted distance in real time.

[0056] S2.2, when the distance is greater than 1 meter, it is within the safe range; when the distance is less than 1 meter, the system continuously monitors and uses the vehicle environment detection algorithm to detect the front and rear of the vehicle. If a turning sign is detected and the vehicle is in a turning state, and the instant distance is less than one meter, the radar voice alarm is triggered to remind the driver that the distance is less than the safe distance; if no turning sign is detected, when the distance is 0.5 meters < distance < 1 meter, the system will issue an alarm; when the distance is < 0.5 meters, the control motherboard of the edge computing processing unit will send a stop signal to the relay, causing the relay to control the vehicle's emergency stop button for emergency parking.

[0057] Furthermore, step S3 specifically includes the following steps:

[0058] S3.1, when processing UWB signals, the control motherboard of the edge computing processing unit will first detect whether the UWB signal has been received. If the UWB signal is detected, the distance, direction and angle between the base station and the handheld positioning card will be immediately analyzed.

[0059] If no UWB signal is detected, the vehicle environment detection algorithm is used to detect people in the front and rear of the vehicle. If pedestrians are detected in the front and rear video streams, the edge computing processing unit will analyze the 485 signal converted from the millimeter-wave radar signal, and make auxiliary judgments based on the distance data analyzed by the 485 signal. When the distance of an object is less than 25 meters, a voice alarm is triggered to remind the driver that there are pedestrians nearby who do not have a handheld positioning card with a UWB signal. That is, if there are pedestrians in the rear but the device has not received the UWB signal, and the radar data shows that there are objects within 25 meters, a warning will be sent to the driver, but the vehicle will not stop.

[0060] S3.2, if the distance between the base station and the handheld positioning card is not in the range of 25-30 meters, it means that it is within 25 meters, then first determine whether the handheld positioning card is in the front, back, left, or right of the vehicle.

[0061] S3.3, when the person carrying the positioning card is in front of or behind the vehicle, if the distance is between the second limit and the third limit, the first level alarm of UWB will be triggered; if the distance is between the first limit and the second limit, the second level alarm of UWB will be triggered; if it is less than the first limit, stop the vehicle immediately; when the person carrying the positioning card is on the left or right rear of the vehicle, the first level alarm and the second level alarm are the same, if it is less than half of the first level limit, stop the vehicle immediately.

[0062] Furthermore, the distances of the first-level limit, the second-level limit and the third-level limit in step S3.3 are 10 meters, 15 meters and 20 meters respectively.

[0063] The present invention has the following beneficial effects:

[0064] The present invention utilizes UWB technology, visual detection technology and millimeter wave radar technology to realize intelligent perception of people, vehicles and objects and illegal operations of drivers, effectively realizes all-round intelligent collision avoidance of trackless equipment, fills the domestic and foreign gaps in intelligent collision avoidance of underground trackless equipment, and has great market application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the specific implementation of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the specific implementation or the prior art description. Obviously, the drawings described below are some implementations of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0066] Figure 1 The present invention shows a structural diagram of an all-round intelligent anti-collision system for underground trackless equipment.

[0067] Figure 2A flow chart of the driver behavior detection algorithm in step S4 of the present invention is shown.

[0068] Figure 3 The improved convolutional neural network CNN structure diagram in step S2.2 and step S4.2 of the present invention is shown.

[0069] The reference numerals in the above drawings are:

[0070] 101. Switching input module; 102. Wireless communication module; 103. Data storage module; 104. Switching output module; 105. Bus chip; 106. Control mainboard; 107. Cooling fan; 108. Power supply processing module; 109. Deep learning computing card; 110. Image acquisition processing module; 111. Ethernet module; 112. 485 module; 113. CAN to 485 chip; 201. Cancel parking button; 202. Current limiting resistor R1; 203. Reset button; 204. Current limiting resistor R2; 205, parking relay; 206, current limiting resistor R3; 207, warning light; 208, current limiting resistor R4; 209, alarm speaker; 210, power amplifier unit; 211 touch screen; 301, power indicator module; 302, positioning card power supply module; 303, sound and light alarm module; 304, vibration alarm module; 305, positioning card power management module; 306, pulse transmitter; 307, positioning card main chip; 308, power amplifier; 309, sleep wake-up chip; 310, positioning card handshake communication chip; 402, CAN interface; 403, radar power supply; 404, radar conversion circuit; 405 radar main chip; 406 PMIC module; 407 receiving module; 408 RF front-end components; 409 transmitting module; 410 receiving antenna; 411 transmitting antenna composition; 502, camera network interface; 503, DSP processing chip; 504, CCD chip; 505, infrared LED fill light; 506, video filter; 507, video storage card; 508, high-definition camera; 601, base station network interface; 602, first antenna; 603, first pulse receiver; 604, second antenna; 605, second pulse receiver; 606, third antenna; 607, third pulse receiver; 608, base station handshake communication chip; 609, positioning main control chip; 610, base station controller; 612, positioning base station power management module; 613, power detection module; 614, synchronizer. DETAILED DESCRIPTION

[0071] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1

[0072] like Figure 1 An all-round intelligent anti-collision system for underground trackless equipment is shown, comprising: an edge computing processing unit, a signal input and output unit, a handheld positioning card, a ranging radar unit, an infrared camera fill light unit and a 360° omnidirectional positioning base station unit; the edge computing processing unit is respectively connected to the signal input and output unit, the ranging radar unit, the infrared camera fill light unit and the 360° omnidirectional positioning base station unit, the handheld positioning card transmits a narrow pulse signal to the 360° omnidirectional positioning base station unit, and the handheld positioning card and the 360° omnidirectional positioning base station unit transmit communication signals to each other.

[0073] The edge computing processing unit consists of a switch input module 101; a wireless communication module 102; a data storage module 103; a switch output module 104; a bus chip 105; a control mainboard 106; a cooling fan 107; a power supply processing module 108; a deep learning computing power card 109; an image acquisition processing module 110; an Ethernet module 111; a 485 module 112; and a CAN to 485 chip 113.

[0074] The main function of the edge computing processing unit is to work closely with the ranging radar unit, infrared fill light camera unit and 360° omnidirectional positioning base station unit. The CAN to 485 chip and 485 module will convert the CAN signal output by the ranging radar unit into a 485 signal, and upload it to the control mainboard through the bus chip to obtain the distance information between the vehicle and the obstacle. If a dangerous distance is detected, the edge computing processing unit will transmit the alarm / stop signal to the switch output module through the bus chip, and trigger the alarm or stop command through the current limiting resistor.

[0075] The deep learning computing power card is specifically used to accelerate deep learning tasks and provide support for the intelligent system detection function. It can provide computing resources for the system, accelerate model reasoning, and improve computing efficiency, greatly ensuring the stable operation of detection tasks.

[0076] The Ethernet module is the receiving end of the data transmission between the infrared fill light camera unit and the 360° omnidirectional positioning base station unit. It sends the driver's behavior information and the environment information in front and behind the vehicle to the image acquisition and processing module, and performs abnormal behavior detection and environmental analysis through the driver behavior detection algorithm and the vehicle environment detection algorithm. If the driver's illegal operation or abnormal behavior is detected, the system will automatically alarm or stop. The edge computing processing unit also works with the 360° omnidirectional positioning base station unit to determine the location of personnel and equipment through positioning data to ensure that the situation of entering the dangerous area can be warned and handled in time. In addition, all sensor data and system status will be stored in the data storage module and connected to the local area network through the wireless communication module, which is convenient for managers to view equipment operation log records and alarm information.

[0077] The power supply processing module of the edge computing processing unit provides stable power support for each module to ensure the continuous operation of the system.

[0078] When the temperature of the edge computing device reaches 50 degrees Celsius or above, the cooling fan can start in time to dissipate heat and ensure the stable operation of the entire device.

[0079] The signal input and output unit includes a cancel parking button 201 , a current limiting resistor R1 202 , a reset button 203 , a current limiting resistor R2 204 , a parking relay 205 , a current limiting resistor R3 206 , an alarm light 207 , a current limiting resistor R4 208 , an alarm horn 209 , a power amplifier unit 210 , and a touch display screen 211 .

[0080] The signal input and output unit is mainly responsible for receiving the user's control instructions and transmitting them to the control main board. It is also responsible for feeding back the information detected by the main board to the user.

[0081] Each current limiting resistor can complete the voltage division task required by the entire circuit, ensure the voltage safety and stability of the switch input and output module, and enable the equipment connected to the switch input and output module to function normally.

[0082] When the driver wants to cancel parking while working (the driver sometimes remotely controls the vehicle to work. At this time, there is no one in the cab, and the system will determine that the driver has left the seat to park, so the parking needs to be canceled at this time; or there are many obstacles around the working scene, and the radar will alarm to stop, which will also affect the work) the cancel parking button will transmit the signal to the switch input module through the current limiting resistor R1, and then transmit the command to the edge computing processing unit through the bus chip, and the control motherboard will parse and release the parking state. When the work is completed and the parking is to be resumed, the signal of the reset button is transmitted to the switch input module through the current limiting resistor R2, and then the bus chip transmits the command to the control motherboard to resume the parking state.

[0083] When the control main board detects an abnormal situation (the driver leaves the seat, the UWB level 3 alarm means parking) and the vehicle needs to be parked, the parking command is transmitted to the switch output module through the bus chip, and finally enters the parking relay through the current limiting resistor R3. The parking relay activates the vehicle's emergency stop button to stop the vehicle.

[0084] When the control main board detects an abnormal situation (the driver's cabin occupant sticks his head or stretches out his hand, the radar is too close, there are people in the front or rear, UWB level 1 or 2 alarms) and an alarm is needed, the control main board will send out an alarm signal, which will be transmitted to the switch output module through the bus chip, and finally enter the alarm light and alarm horn through the current limiting resistor R4 and the power amplifier unit, realizing a dual sound and light alarm.

[0085] The touch screen is used to display system status, alarm information and execution feedback to the operator in real time. It communicates with the bus chip through the 485 module and presents the processing results of the edge computing processing unit to the user in a visual way.

[0086] The handheld positioning card is composed of a power indicator module 301, a positioning card power supply module 302, an audible and visual alarm module 303, a vibration alarm module 304, a positioning card power management module 305, a pulse transmitter 306, a positioning card main chip 307, a power amplifier 308, a sleep wake-up chip 309, and a positioning card handshake communication chip 310. The handheld positioning card is a mobile positioning module in the system, responsible for real-time tracking of the specific location of the card holder, and works in conjunction with the 360° omnidirectional positioning base station unit.

[0087] The main chip of the positioning card will continue to send out unique identification and related data, and the pulse transmitter will convert the digital signal of the main chip of the positioning card into a pulse signal suitable for wireless transmission. The signal strength is amplified by the power amplifier and transmitted to the positioning base station. The handshake communication chip of the positioning card exchanges information with the handshake communication chip of the base station to complete the information synchronization between the positioning card and the base station. When the base station detects that the positioning card is too close, it will transmit the alarm signal back to the main chip of the positioning card through the handshake communication chip of the positioning card, and the sleep wake-up chip will call the main chip of the positioning card to trigger the sound and light alarm module and the vibration alarm module to remind the wearer.

[0088] The positioning card power supply module provides stable power support for its internal chip and transmitter module. The positioning card power management module is responsible for intelligently monitoring the battery status and optimizing power consumption management to extend battery life. When the battery is low, the power indicator module will display a red light to remind the operator to charge the positioning card in time.

[0089] The infrared fill light camera unit includes a camera network interface 502, a DSP processing chip 503, a CCD chip 504, an infrared LED fill light 505, a video filter 506, a video storage card 507, and a high-definition camera 508. The infrared fill light camera unit provides an infrared light source for the camera, thereby capturing high-quality images.

[0090] In a dark environment with no visible light or dim light, the infrared LED fill light will actively project infrared light onto the object, so that the infrared light enters the lens for imaging after being reflected by the object. After the high-definition camera captures the image, it filters out infrared light and trims the incident light through a video filter, and transmits the filtered image to the CCD chip, which converts the light into an electronic signal. The DSP processing chip is responsible for processing the electronic signals collected by the image sensor and converting them into real images that can be seen. The generated image will be stored in the video memory card and transmitted to the edge computing processing unit through the network interface to retrieve the camera's video stream in real time for video display and image processing. The network interface is connected to the Ethernet module of the edge computing processing unit, transmitting image data at a rate of 100 megabits to provide support for subsequent analysis functions.

[0091] The ranging radar unit is composed of a CAN interface 402 , a radar power supply 403 , a radar conversion circuit 404 , a radar main chip 405 , a PMIC module 406 , a pulse receiving module 407 , a RF front-end component 408 , a transmitting module 409 , a receiving antenna 410 , and a transmitting antenna 411 .

[0092] The ranging radar unit is mainly responsible for real-time monitoring and distance measurement of the surrounding environment, ensuring that the system can accurately perceive the relative distance of surrounding obstacles during operation. The radar power supply adopts a snap-on design, which can firmly connect to the radar conversion circuit inlet and provide stable power supply for the entire radar unit. The PMIC module has a voltage stabilization function and can reasonably control the power supply to ensure stable power supply of the entire radar unit. When the voltage is abnormal, it can protect the entire radar circuit in time. High-frequency electromagnetic waves are emitted through the RF front-end component. When the electromagnetic waves encounter the surface of an object, they will be reflected, and the reflected signal will be transmitted back to the radar main chip via the receiving antenna. The transmitting antenna adopts a uniform spacing design, which can improve the performance indicators of the radar system such as gain and beam width. The transmitting antenna ensures the stability of the radar unit signal reception and provides guarantee for subsequent signal analysis. The radar main chip digitizes the echo signal and calculates the relative distance between the object and the system. After the calculation, the data will be transmitted to the edge computing processing unit through the CAN interface for further data analysis and decision-making.

[0093] The core components of the ranging radar unit are the radar main chip and the radar conversion circuit. The radar chip parses the collected echo signal into distance data and transmits it to the edge computing processing unit. In some special cases, such as when an obstacle is detected or the car is facing a wall when reversing, the ranging radar unit detects that the target is too close, and sends an alarm signal through the control motherboard. The power amplifier unit drives the alarm light and the alarm horn to alert the operator of potential dangers around. If necessary, the radar main chip will immediately transmit the data to the signal input and output unit through the CAN interface, triggering the parking relay to perform an emergency stop to ensure the safety of the equipment and operators.

[0094] The ranging radar unit also collaborates with other perception modules (such as infrared fill light camera units). In complex environments, the system can not only detect the presence of objects, but also classify targets based on the results of image recognition. The edge computing processing unit fuses radar data and image data captured by the camera to achieve more accurate target recognition and judgment. The radar signal is combined with image information to achieve timely alarm and emergency stop when necessary, and there will be no false alarms that hinder driving in special circumstances. Communication between the ranging radar unit and the 485 module is also particularly important. The data collected by the radar will be transmitted to other modules via the 485 module, giving the entire system multi-level information perception capabilities. In addition, the detection range of the ranging radar unit can be set through the touch display, which is convenient for the driver to make corresponding adjustments according to different industrial and mining environments.

[0095] The 360° omnidirectional positioning base station unit is composed of a base station network interface 601; a first antenna 602; a first pulse receiver 603; a second antenna 604; a second pulse receiver 605; a third antenna 606; a third pulse receiver 607; a base station handshake communication chip 608; a positioning main control chip 609; a base station controller 610; a positioning base station power management module 612; a power detection module 613; and a synchronizer 614. The 360° omnidirectional positioning base station unit is mainly responsible for the signal interaction with the card holding positioning unit to achieve accurate two-way positioning of personnel and equipment.

[0096] The first antenna, the second antenna, and the third antenna will simultaneously capture the pulse signal emitted by the handheld positioning unit and jointly obtain the distance, direction, and angle information. The first pulse receiver, the second pulse receiver, and the third pulse receiver will transmit the signal to the positioning main control chip for analysis. The analysis data is uploaded to the edge computing processing unit through the base station network interface to obtain the precise location of the cardholder.

[0097] The base station controller is responsible for managing the received signal strength and quality, ensuring that the signals between antennas do not conflict, and monitoring the operating status of the entire base station to ensure that the system continues to work stably.

[0098] The synchronizer is mainly responsible for time synchronization. It uses the time-of-flight (TOF) and time-difference-of-arrival (TDOA) positioning algorithms to enable the pulse receiving module, antenna, and base station controller to process signals at the same time node, avoiding positioning errors caused by time differences and reducing signal delays.

[0099] The positioning base station power management module is responsible for reasonable power scheduling, and the positioning base station power detection module is responsible for detecting the power supply and power status. Together, they ensure that the base station can work normally under uninterrupted power supply. The base station handshake communication chip exchanges signals with the positioning card handshake communication chip to establish two-way communication.

[0100] Embodiment 2

[0101] The present invention also provides an all-round intelligent anti-collision method for underground trackless equipment, which specifically includes the following steps:

[0102] S1, the system is powered on and acquires millimeter wave radar signals, visual signals and UWB signals.

[0103] S2, the edge computing processing unit continuously receives millimeter-wave radar signals and converts them into 485 signals for distance conversion; when the distance is less than the set value, the vehicle environment detection algorithm is used to detect the front and rear of the vehicle, and the system issues an alarm or controls the vehicle to make an emergency stop based on the detection results.

[0104] S3, using the UWB signal to calculate the position information between the base station and the handheld positioning card, and sending an alarm message according to the position information.

[0105] S4, when the edge computing processing unit reads the visual signal, it identifies the driver's illegal operation through the driver behavior detection algorithm, and issues an alarm or stops the vehicle for the illegal operation.

[0106] Specifically, the vehicle environment detection algorithm in step S2 includes the following steps:

[0107] S2.1, the system obtains the video stream as input data and preprocesses the video stream data, including: adjusting image size, pixel normalization and format conversion.

[0108] S2.2, the video stream obtained after preprocessing is input into the improved convolutional neural network (CNN) module for feature extraction, multi-scale feature map fusion and target detection.

[0109] S2.3, determine whether pedestrians and turn signs are detected.

[0110] S2.4, if a pedestrian is detected, the real-time detection data of the millimeter-wave radar is immediately read for auxiliary judgment. When the radar data is less than 25 meters, the system alarm is activated; if a turn sign is detected, it means that the vehicle is about to enter the turning state. At this time, the parking function for obstacles less than 0.5 meters determined by the millimeter-wave radar detection distance is temporarily cancelled to ensure that the vehicle can pass through narrow corners smoothly.

[0111] Specifically, Figure 2 As shown, the driver behavior detection algorithm in step S4 includes data preprocessing of the video stream, feature extraction using an improved convolutional neural network (CNN) module, and then object detection and classification of the video stream, including the following steps:

[0112] S4.1, the system obtains the video stream as input data and preprocesses the video stream data, including: adjusting the image size, pixel normalization and format conversion. The image size adjustment is to scale to 640x640, and pixel normalization is to map the pixel value to [0,1] to ensure that the data matches the input requirements of the model.

[0113] S4.2, the video stream obtained after preprocessing is input into the improved convolutional neural network (CNN) module for feature extraction, multi-scale feature map fusion and target detection; the preprocessed image frame is passed to the improved convolutional neural network (CNN). In CNN, the network extracts feature information at different levels through multi-layer convolution operations, and gradually aggregates low-level information such as edges and textures to high-level semantic features. The network uses multi-scale convolution, which can process targets of different sizes at the same time, ensuring good detection effects for both small and large targets. After reasoning, the network outputs information such as target category and confidence.

[0114] S4.3, classify the detected targets and distinguish between normal and abnormal behaviors. After the detection is completed, the model classifies the identified targets and distinguishes between normal and abnormal behaviors as the basis for subsequent decision-making.

[0115] S4.4, for abnormal behavior, start the system alarm or brake to stop. The system executes corresponding instructions based on the detection and classification results. For example, if the driver probes into the cab and reaches out, it will trigger an abnormal behavior alarm, and if the driver leaves the seat, it will trigger a brake to stop. This workflow realizes end-to-end intelligent target detection, ensuring real-time and accuracy.

[0116] The all-round intelligent anti-collision method for underground trackless equipment provided by the present invention is mainly realized by the driver behavior detection algorithm and the vehicle environment detection algorithm. The two adopt the same improved CNN convolutional neural network structure, and the difference lies in the different data sets used for the training model. The driver behavior detection algorithm uses the data set training model made of abnormal behavior pictures of the driver in the cab, such as probing, reaching out, leaving the seat, etc., and is mainly used in the infrared camera in the cab; the vehicle environment detection algorithm uses the distribution pictures of people in front and behind the vehicle and the pictures of the turn sign at the corner of the lane to train the model, and is mainly used in the infrared cameras in front and behind the vehicle.

[0117] Specifically, Figure 3 As shown, the improved convolutional neural network CNN in step S2.2 and step S4.2 includes: a backbone network, a connection part and a task head.

[0118] The backbone network includes: an input layer Input, a first convolutional layer Conv, a second convolutional layer Conv, a first general inverted bottleneck module C3_UIB, a third convolutional layer Conv, a second general inverted bottleneck module C3_UIB, a fourth convolutional layer Conv, a third general inverted bottleneck module C3_UIB, a fifth convolutional layer Conv, a fourth general inverted bottleneck module C3_UIB and a spatial pyramid pooling layer SPPF, which are connected in sequence.

[0119] The connection part includes: the sixth convolution layer Conv, the first adaptive multi-scale sampling layer AMS, the first connection layer Concat, the fifth universal inverted bottleneck module C3_UIB, the seventh convolution layer Conv, the second adaptive multi-scale sampling layer AMS, the second connection layer Concat, the sixth universal inverted bottleneck module C3_UIB, the eighth convolution layer Conv, the third connection layer Concat, the seventh universal inverted bottleneck module C3_UIB, the ninth convolution layer Conv, the fourth connection layer Concat and the eighth universal inverted bottleneck module C3_UIB, which are connected in sequence.

[0120] The task header includes: detection layer Detect.

[0121] Among them, the second universal inverted bottleneck module is also connected to the second connection layer, the third universal inverted bottleneck module is also connected to the first connection layer, the spatial pyramid pooling layer is also connected to the sixth convolutional layer, and the sixth convolutional layer is also connected to the eighth universal inverted bottleneck module, the seventh convolutional layer is also connected to the third convolutional layer, and the sixth universal inverted bottleneck module, the seventh universal inverted bottleneck module and the eighth universal inverted bottleneck module are respectively connected to the detection layer.

[0122] The improved convolutional neural network model (referred to as model in this invention) in the driver behavior detection algorithm and the vehicle environment detection algorithm integrates the universal inverted bottleneck (UIB) block and the adaptive multi-scale sampling (AMS) module, aiming to improve the feature extraction capability and computational efficiency of the model, especially for the deployment of embedded devices and real-time task applications.

[0123] The others are the basic structures of convolutional neural networks: Conv is the convolution layer, which is responsible for feature extraction and reduces the size of the feature map by step size. Convolutions of different sizes are used in this network structure to extract features at different scales. The role of the SPPF layer is to aggregate the features of multi-scale convolutions to improve the robustness of detection. Concat represents the connection operation, which splices the feature maps of different layers together to form richer features. Detect is the final detection layer, which performs detection at different scales to ensure good detection performance for targets of different sizes.

[0124] Based on the CNN convolutional neural network, the traditional C3 module is replaced by the C3_UIB universal inverted bottleneck module. UIB has the advantage of an inverted bottleneck structure. One path continues to maintain the standard convolution unchanged, and the other path introduces a deep convolution layer, making the network more flexible when expanding features and mixing channels. The deep convolution is added before the expansion layer and between the expansion and projection layers, which enhances the expressive power of feature extraction and optimizes the interaction of spatial and channel features. After the introduction of the UIB module, the model can adaptively process features of different complexities and flexibly adjust the allocation of computing resources between different network layers.

[0125] Specifically, the universal inverted bottleneck module in the improved convolutional neural network CNN includes two paths. The first path is a standard convolutional layer, and the second path introduces a deep convolutional layer based on the standard convolutional layer:

[0126] The first standard convolution is recorded as:

[0127] ;

[0128] in, is the feature obtained after the first convolution. is an input feature of size H×W×C; H, W, C represent the height, width, and number of channels respectively. is a convolution kernel of size k×k×C, The height and width of the convolution kernel are both k; is the bias term; is the ReLU activation function; Represents a standard convolution operation.

[0129] The second path introduces a deep convolutional layer based on the standard convolutional layer, which is denoted as:

[0130] ;

[0131] in, is the feature obtained after the second convolution. is the dilated convolution, ; Is the size of The convolution kernel of is the bias vector, Represents channel-wise convolution.

[0132] Concatenate the features of the first and second paths to obtain the concatenated features. :

[0133] ;

[0134] in, Represents a join operation.

[0135] Finally, the concatenated features are convolved to obtain the final output :

[0136] ;

[0137] in, The convolution kernel for convolution processing of the splicing features, To help improve the convolutional neural network model, adjust the bias term of the channel output.

[0138] Specifically, the adaptive multi-scale sampling layer in the improved convolutional neural network CNN first downsamples the input features by different multiples to generate a multi-scale feature map:

[0139] ;

[0140] ;

[0141] ;

[0142] in is the input feature; Indicates that the features are proportional Zoom out.

[0143] The adaptive multi-scale sampling layer then calculates the complexity weight for each downsampled feature:

[0144] ;

[0145] in, is the complexity weight, is the complexity calculation function, include .

[0146] According to the complexity weight, features of different scales are adaptively upsampled:

[0147] ;

[0148] in, Upsampling the features, To adaptively adjust the upsampling ratio according to the weight, is the feature after adaptive upsampling.

[0149] Finally, the upsampled features of different scales are weighted and fused to obtain the final output :

[0150] ;

[0151] in, Take 1, 2, 3.

[0152] In the upsampling stage, the adaptive multi-scale sampling (AMS) module is introduced to replace the traditional fixed upsampling method. The AMS module can achieve efficient processing of complex areas by adaptively adjusting the upsampling ratio of different regions according to the complexity of the input feature map. Specifically, the AMS module can dynamically adjust the upsampling scale according to the local complexity of the input feature map to adapt to the feature changes in different regions. This adaptive upsampling strategy can effectively reduce the amount of calculation in simple areas, while improving the performance of small target detection and edge feature processing. By adaptively upsampling different areas, the driver behavior detection algorithm and the vehicle environment detection algorithm can better capture the detailed information in the image, thereby improving the accuracy of target detection. At the same time, since the AMS module can dynamically adjust the upsampling scale according to the complexity of the input feature map, it can achieve fast inference speed while maintaining high detection accuracy. This makes the model more suitable for real-time target detection tasks.

[0153] Specifically, step S2 includes the following steps:

[0154] S2.1, the edge computing processing unit continuously receives millimeter-wave radar signals, converts the millimeter-wave radar signals into 485 signals for distance conversion, and analyzes the converted distance in real time.

[0155] S2.2, when the distance is greater than 1 meter, it is within the safe range; when the distance is less than 1 meter, the system continuously monitors and uses the vehicle environment detection algorithm to detect the front and rear of the vehicle. If a turning sign is detected and the vehicle is in a turning state, and the instant distance is less than one meter, the radar voice alarm is triggered to remind the driver that the distance is less than the safe distance; if no turning sign is detected, when the distance is 0.5 meters < distance < 1 meter, the system will issue an alarm; when the distance is < 0.5 meters, the control motherboard of the edge computing processing unit will send a stop signal to the relay, causing the relay to control the vehicle's emergency stop button for emergency parking.

[0156] The edge computing processing unit will continuously receive radar CAN signals, convert CAN signals into 485 signals for distance conversion, and analyze the converted distance in real time. When the distance is greater than 1 meter, it is within the safe range; when the distance is less than 1 meter, the system will continue to monitor and use the vehicle environment detection algorithm to detect the front and rear of the vehicle. If a turning sign is detected (a special turning sign will be posted at the place where the turn is about to take place, and the system will detect this sign), it means that it is in a turning state. Even if the distance is less than one meter, the vehicle will not stop, but will trigger the radar voice alarm to remind the driver that it is less than the safe distance; if the turning sign is not detected, then when the distance is 0.5 meters < distance < 1 meter, the system will issue an alarm, and when the distance is < 0.5 meters, the control motherboard of the edge computing processing unit will send a stop signal to the relay, so that the relay controls the emergency stop button of the vehicle for emergency parking.

[0157] Specifically, step S3 includes the following steps:

[0158] S3.1, when processing UWB signals, the control motherboard of the edge computing processing unit will first detect whether the UWB signal has been received. If the UWB signal is detected, the distance, direction and angle between the base station and the handheld positioning card will be immediately analyzed.

[0159] If no UWB signal is detected, the vehicle environment detection algorithm is used to detect people in the front and rear of the vehicle. If pedestrians are detected in the front and rear video streams, the edge computing processing unit will analyze the 485 signal converted from the millimeter-wave radar signal, and make auxiliary judgments based on the distance data analyzed by the 485 signal. When the distance of an object is less than 25 meters, a voice alarm is triggered to remind the driver that there may be pedestrians without a UWB signal handheld positioning card nearby. That is, if there are pedestrians behind but the device has not received the UWB signal, and the radar data shows that there are objects within 25 meters, a warning will be sent to the driver, but the car will not stop.

[0160] S3.2, if the distance between the base station and the handheld positioning card is not in the range of 25-30 meters, it means that it is within 25 meters, then first determine whether the handheld positioning card is in the front, back, left, or right of the vehicle.

[0161] S3.3, when the person carrying the positioning card is in front of or behind the vehicle, if the distance is between the second limit and the third limit, the first level alarm of UWB will be triggered; if the distance is between the first limit and the second limit, the second level alarm of UWB will be triggered; if it is less than the first limit, stop the vehicle immediately; when the person carrying the positioning card is on the left or right rear of the vehicle, the first level alarm and the second level alarm are the same, if it is less than half of the first level limit, stop the vehicle immediately.

[0162] When processing UWB signals, the control motherboard will first detect whether the UWB signal has been received. If it is detected, it will immediately parse the distance, direction and angle between the base station and the positioning card, and display them on the touch screen. If the distance between the base station and the positioning card is greater than 25 meters and less than 30 meters, it is in a safe distance and will be displayed as a blue dot on the touch screen; when the distance is less than 25, it will be displayed as a red dot. If it is not in the 25-30 meter range, it means that it is within 25 meters (the maximum communication distance between the base station and the positioning card is only 30 meters. If it exceeds 30 meters, no signal can be received, so it must be within 25 meters). At this time, it is necessary to first determine whether the positioning card is in the front, back, left, and right of the vehicle (the parking distance in the front and back is large, and the parking distance in the left and right is small. If the parking distance is set to 7 meters, the front and back will stop when it is less than 7 meters, and the positioning cards on the left and right will only stop when it is less than half of the set threshold, that is, 3.5 meters. However, the distance between the first and second alarms does not need to be halved. If a 10-15 meter second alarm is set, then the alarm is 10-15 meters regardless of the front, back, left, and right. When the person carrying the positioning card is in the front and rear of the vehicle, if the distance is between the second level limit and the third level limit, the first level alarm of UWB will be triggered. If the distance is between the first level limit and the second level limit, the second level alarm of UWB will be triggered. If it is less than the first level limit, the vehicle will stop immediately (for example, the third level limit is 20, the second level is 15, and the first level is 10. If it is greater than 15 and less than 20, the first level alarm will be triggered. If it is greater than 10 and less than 15, the second level alarm will be triggered. If it is less than 10, the vehicle will stop immediately); when the person carrying the positioning card is in the left and right rear of the vehicle, the first level alarm and the second level alarm are the same. If it is less than half of the first level limit, the vehicle will stop immediately. If the system cannot detect the UWB signal all the time, in order to prevent people without positioning cards from appearing nearby, the vehicle environment detection algorithm will be used to detect people in the front and rear of the vehicle. If pedestrians are detected in the front and rear video streams, the radar data will be read and the radar data will be used for auxiliary judgment. When there are pedestrians in the rear and there are objects within 25 meters of the radar, a warning will be sent to the person, but the vehicle will not stop.

[0163] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. An all-round intelligent anti-collision method for underground trackless equipment, applied to an all-round intelligent anti-collision system for underground trackless equipment, comprising: Edge computing processing unit, signal input and output unit, handheld positioning card, ranging radar unit, infrared camera fill light unit and 360° omnidirectional positioning base station unit; the edge computing processing unit is respectively connected with the signal input and output unit, ranging radar unit, infrared camera fill light unit and 360° omnidirectional positioning base station unit, the handheld positioning card transmits the narrow pulse signal to the 360° omnidirectional positioning base station unit, and the handheld positioning card and the 360° omnidirectional positioning base station unit transmit communication signals to each other; It is characterized in that an all-round intelligent anti-collision method for underground trackless equipment specifically comprises the following steps: S1, the system is powered on and acquires millimeter wave radar signals, visual signals and UWB signals; S2, the edge computing processing unit continuously receives millimeter-wave radar signals and converts them into 485 signals for distance conversion; when the distance is less than the set value, the vehicle environment detection algorithm is used to detect the front and rear of the vehicle, and the system issues an alarm or controls the vehicle to make an emergency stop based on the detection results; S3, using the UWB signal to calculate the position information between the base station and the handheld positioning card, and sending an alarm message according to the position information; S4, when the edge computing processing unit reads the visual signal, it identifies the driver's illegal operation through the driver behavior detection algorithm, and issues an alarm or stops the vehicle for the illegal operation; The vehicle environment detection algorithm in step S2 includes the following steps: S2.1, the system obtains a video stream as input data and preprocesses the video stream data, including: adjusting image size, pixel normalization and format conversion; S2.2, the video stream obtained after preprocessing is input into the improved convolutional neural network (CNN) module for feature extraction, multi-scale feature map fusion and target detection; S2.3, determining whether pedestrians and turn signs are detected; S2.4: If a pedestrian is detected, the real-time detection data of the millimeter-wave radar is immediately read for auxiliary judgment. When the radar data is less than 25 meters, the system alarm is activated; if a turning sign is detected, it means that the vehicle is about to enter the turning state. At this time, the obstacle less than 0.5 meters determined by the millimeter-wave radar detection distance is temporarily cancelled to ensure that the vehicle can pass through narrow corners smoothly; The driver behavior detection algorithm in step S4 includes the following steps: S4.1, the system obtains a video stream as input data, and preprocesses the video stream data, including: adjusting image size, pixel normalization, and format conversion; S4.2, the video stream obtained after preprocessing is input into the improved convolutional neural network CNN for target detection; S4.3, classify the detected targets and distinguish between normal and abnormal behaviors; S4.4, for abnormal behavior, start the system alarm or brake to stop; The improved convolutional neural network CNN in step S2.2 and step S4.2 includes: a backbone network, a connection part and a task head; The backbone network includes: an input layer, a first convolutional layer, a second convolutional layer, a first general inverted bottleneck module, a third convolutional layer, a second general inverted bottleneck module, a fourth convolutional layer, a third general inverted bottleneck module, a fifth convolutional layer, a fourth general inverted bottleneck module and a spatial pyramid pooling layer connected in sequence; The connection part includes: a sixth convolution layer, a first adaptive multi-scale sampling layer, a first connection layer, a fifth universal inverted bottleneck module, a seventh convolution layer, a second adaptive multi-scale sampling layer, a second connection layer, a sixth universal inverted bottleneck module, an eighth convolution layer, a third connection layer, a seventh universal inverted bottleneck module, a ninth convolution layer, a fourth connection layer and an eighth universal inverted bottleneck module, which are connected in sequence; The task head includes: detection layer; Among them, the second universal inverted bottleneck module is also connected to the second connection layer, the third universal inverted bottleneck module is also connected to the first connection layer, the spatial pyramid pooling layer is also connected to the sixth convolutional layer, and the sixth convolutional layer is also connected to the eighth universal inverted bottleneck module, the seventh convolutional layer is also connected to the third convolutional layer, and the sixth universal inverted bottleneck module, the seventh universal inverted bottleneck module and the eighth universal inverted bottleneck module are respectively connected to the detection layer.

2. The all-round intelligent anti-collision method for underground trackless equipment according to claim 1, characterized in that: The general inverted bottleneck module in the improved convolutional neural network CNN consists of two paths. The first path is a standard convolutional layer, and the second path introduces a deep convolutional layer based on the standard convolutional layer: The first standard convolution is recorded as: ； in, is the feature obtained after the first convolution. is an input feature of size H×W×C; H, W, C represent the height, width, and number of channels respectively. is a convolution kernel of size k×k×C, The height and width of the convolution kernel are both k; is the bias term; is the ReLU activation function; represents a standard convolution operation; The second path introduces a deep convolutional layer based on the standard convolutional layer, which is denoted as: ； in, is the feature obtained after the second convolution. is the dilated convolution, ; Is the size of The convolution kernel of is the bias term vector; Represents channel-by-channel convolution; Concatenate the features of the first and second paths to obtain the concatenated features. : ； in, Represents a connection operation; Finally, the concatenated features are convolved to obtain the final output : ； in, The convolution kernel for convolution processing of the splicing features, To help improve the convolutional neural network model, adjust the bias term of the channel output.

3. The all-round intelligent anti-collision method for underground trackless equipment according to claim 2, characterized in that: The adaptive multi-scale sampling layer in the improved convolutional neural network CNN first downsamples the input features by different multiples to generate a multi-scale feature map: ； ； ； in is the input feature; Indicates that the features are proportional Zoom out; The adaptive multi-scale sampling layer then calculates the complexity weight for each downsampled feature: ； in, is the complexity weight, is the complexity calculation function, include ; According to the complexity weight, features of different scales are adaptively upsampled: ； in, For the characteristics of mining, To adaptively adjust the upsampling ratio according to the weight, is the feature after adaptive upsampling; Finally, the upsampled features of different scales are weighted and fused to obtain the final output : ； in, Take 1, 2, 3.

4. The all-round intelligent anti-collision method for underground trackless equipment according to claim 1, characterized in that: Step S2 specifically includes the following steps: S2.1, the edge computing processing unit continuously receives millimeter-wave radar signals, converts the millimeter-wave radar signals into 485 signals for distance conversion, and analyzes the converted distance in real time; S2.2, when the distance is greater than 1 meter, it is within the safe range; when the distance is less than 1 meter, the system continuously monitors and uses the vehicle environment detection algorithm to detect the front and rear of the vehicle. If a turning sign is detected and the vehicle is in a turning state, and the instant distance is less than one meter, the radar voice alarm is triggered to remind the driver that the distance is less than the safe distance; if no turning sign is detected, when the distance is 0.5 meters < distance < 1 meter, the system will issue an alarm; when the distance is < 0.5 meters, the control motherboard of the edge computing processing unit will send a stop signal to the relay, causing the relay to control the vehicle's emergency stop button for emergency parking.

5. The all-round intelligent anti-collision method for underground trackless equipment according to claim 1, characterized in that: Step S3 specifically includes the following steps: S3.1, when processing UWB signals, the control motherboard of the edge computing processing unit will first detect whether the UWB signal has been received. If the UWB signal is detected, the distance, direction and angle between the base station and the handheld positioning card will be immediately analyzed; If no UWB signal is detected, the vehicle environment detection algorithm is used to detect people in the front and rear of the vehicle. If pedestrians are detected in the front and rear video streams, the edge computing processing unit will analyze the 485 signal converted from the millimeter-wave radar signal, and make auxiliary judgments based on the distance data analyzed by the 485 signal. When the distance of an object is less than 25 meters, a voice alarm is triggered to remind the driver that there are pedestrians without a UWB signal handheld positioning card nearby. That is, if there are pedestrians behind the vehicle but the device has not received the UWB signal, and the radar data shows that there are objects within 25 meters, a warning will be sent to the driver, but the vehicle will not stop. S3.2, if the distance between the base station and the handheld positioning card is not within the range of 25-30 meters, it means that it is within 25 meters, then first determine whether the handheld positioning card is in the front, back, left, or right of the vehicle; S3.3, when the person carrying the positioning card is in front of or behind the vehicle, if the distance is between the second limit and the third limit, the first level alarm of UWB will be triggered; if the distance is between the first limit and the second limit, the second level alarm of UWB will be triggered; if it is less than the first limit, stop the vehicle immediately; when the person carrying the positioning card is on the left or right rear of the vehicle, the first level alarm and the second level alarm are the same, if it is less than half of the first level limit, stop the vehicle immediately.

6. The all-round intelligent anti-collision method for underground trackless equipment according to claim 5, characterized in that: The distances of the first-level limit, the second-level limit and the third-level limit in step S3.3 are 10 meters, 15 meters and 20 meters respectively.

Images