An intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects

Abstract

The present invention relates to the field of traffic facilities, and specifically to an intelligent early warning system for preventing accidental intrusion in highway reconstruction and expansion projects, including a sensor module for collecting environmental data around the construction area, a data processing module for fusing and processing the data collected by the sensor module, extracting target information, an intelligent analysis module for performing behavioral analysis and prediction on the target information, and judging whether the target has the risk of accidentally intruding into the construction area, and an early warning information release module for releasing early warning information to construction personnel and passing vehicles according to the early warning signal of the intelligent analysis module. The present invention overcomes the limitations of a single sensor through the collaborative work of a laser radar, a millimeter-wave radar, and a high-definition camera, and adopts an advanced data layer fusion algorithm, so that the system can grasp the motion state and position information of the target object more accurately, and provides a reliable data basis for subsequent intelligent analysis and early warning.

Description

Technical Field

[0001] The present invention relates to the field of traffic facilities, and in particular to an intelligent early warning system for preventing accidental entry in a highway reconstruction and expansion project. Background Art

[0002] During the construction of highway reconstruction and expansion projects, the surrounding environment of the construction area is complex, the traffic conditions of vehicles and pedestrians are changeable, and there is a high risk of accidental entry, which poses a serious threat to the life safety of construction workers and the safety of passing vehicles and pedestrians. Traditional safety warning measures, such as setting up static warning signs and roadblocks, often fail to promptly and effectively remind vehicles and pedestrians to pay attention to the construction area, especially in severe weather conditions with low visibility. The warning effect of these traditional methods is greatly reduced.

[0003] With the continuous development of science and technology, some simple monitoring systems based on a single sensor have gradually been applied, but these systems have obvious limitations. For example, although relying solely on lidar can accurately obtain the contour and distance information of the target object, its performance will be affected in bad weather, and the speed detection accuracy of the target object is limited; although millimeter-wave radar can penetrate bad weather to detect the speed and position of the target object, its ability to recognize the detailed features of the target object is weak; high-definition cameras can clearly obtain the appearance features of the target object, but when the light is dim or the target object moves quickly, the image may appear blurred, resulting in inaccurate extraction of target information, and its direct measurement capability of distance and speed is insufficient.

[0004] These single sensor systems are unable to fully and accurately grasp the dynamic situation around the construction area, and are unable to meet the high requirements of highway reconstruction and expansion projects for safety warnings. There is an urgent need for an intelligent system that can integrate the advantages of multiple sensors, accurately monitor and effectively warn. Summary of the invention

[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides an intelligent early warning system for preventing accidental intrusion in highway reconstruction and expansion projects, which can effectively solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention provides an intelligent early warning system for preventing accidental intrusion in highway reconstruction and expansion projects, comprising a sensor module, a data processing module, an intelligent analysis module and an early warning information release module; The sensor module includes a laser radar, a millimeter wave radar and a high-definition camera, which are used to collect environmental data around the construction area; The data processing module is connected to the sensor module and is used to perform fusion processing on the data collected by the sensor module and extract target information; The intelligent analysis module is connected to the data processing module and is used to perform behavior analysis and prediction on the target information to determine whether the target has a risk of accidentally entering the construction area; The early warning information issuing module is connected to the intelligent analysis module and is used to issue early warning information to construction personnel and passing vehicles according to the early warning signal of the intelligent analysis module.

[0007] Furthermore, the laser radar is used to scan the three-dimensional contours and distance information of objects around the construction area, the millimeter wave radar is used to detect the speed and position of the moving target, and the high-definition camera is used to obtain the appearance feature information of the target.

[0008] Furthermore, the target information extracted includes the speed of the target object. , acceleration , Driving direction And the position coordinates of the target object on the two-dimensional plane The acquisition of the target information adopts a data layer fusion algorithm, and the specific steps include: S100, pre-processes and synchronizes data collected by laser radar, millimeter wave radar and high-definition camera; Perform timestamp calibration on the data collected by the laser radar, millimeter-wave radar and high-definition camera. The laser radar, millimeter-wave radar and high-definition camera are equipped with high-precision clocks. Time synchronization calibration is performed when the system starts. The network time protocol is used for clock synchronization. During the data collection process, the clock deviation of each sensor is monitored. If the deviation exceeds the preset threshold, the timestamp is corrected. The data collected by LiDAR, millimeter-wave radar and high-definition camera are screened. For LiDAR data, the signal strength of each scanning point is checked, and the scanning point data with signal strength below the preset strength threshold is eliminated; for millimeter-wave radar data, the data with speeds several times greater than the road speed limit or distance value jumps are eliminated; the image data of the high-definition camera is preprocessed, including image denoising, image enhancement, and detection and segmentation of target objects; S200, obtain the speed of the target object ; At the same timestamp, obtain the velocity component calculated by the lidar based on the position change of the vehicle contour point , , ..., , the velocity estimate of the lidar is obtained by vector synthesis , the calculation formula is: Where p is the number of selected contour points; At the same time, obtain the velocity value of the millimeter wave radar based on the Doppler effect and the speed value of the HD camera through frame-to-frame analysis ; right , and Perform weight allocation and set initial weight values ​​according to the nominal accuracy and reliability of LiDAR, millimeter-wave radar, and high-definition camera. The weight of LiDAR is , the weight of millimeter wave radar is , the HD camera weight is ,and ; The adjusted weights are used to calculate the fused speed value ; S300, obtaining the acceleration of the target object ; The speed value after fusion Get the fusion velocity sequence , , ..., , the fused speed sequence is smoothed, and the moving average filtering method is used to calculate the moving average value of the window size k as the smoothed speed value. The calculation formula is: For the smoothed velocity series , , ..., , using the central difference method, calculate the acceleration , the calculation formula is: in, is the time step, i.e., the interval between adjacent timestamps; S400: Obtaining the driving direction of the target object ; Standardize and evaluate the reliability of the direction vector. The direction vector obtained by the lidar After being calculated based on the change in the position of the contour points, it is standardized so that its modulus is unit length. The calculation formula is: The direction vector obtained by the millimeter wave radar according to the angle change of the target object ,Combining the angle measurement accuracy and the distance information of the target object, the reliability is evaluated; The HD camera analyzes the displacement direction of the target object from the image to obtain the direction vector ,Reliability is evaluated based on the location and size of the target object in the image and the resolution of the image; According to the reliability evaluation results of each direction vector, the weighting coefficient is determined. The weighting coefficient of the laser radar direction vector is: , the weighting coefficient of the millimeter wave radar direction vector is , the weighting coefficient of the HD camera direction vector is ,and ; Calculate the fused driving direction vector , the calculation format is: Then calculate The fused driving direction is obtained by the direction angle, and the calculation formula is: in, and yes The x and y components of S500, obtaining the position coordinates of the target object on the two-dimensional plane ; Design a hierarchical storage structure, use timestamp as index, and create a two-dimensional array T. Store timestamp as Data records, Store the distance and angle information of the laser radar, using differential coding compression storage. Store the speed, distance and angle data of the millimeter wave radar, using quantized coding and compressed storage. Store the coordinate information of the target position of the high-definition camera image; At each timestamp, the distance and angle information of the laser radar is converted into coordinate values ​​in the rectangular coordinate system, and fused with the target position coordinates of the high-definition camera image to obtain the fused comprehensive position coordinates , the calculation formula is: in, is the coordinate of the laser radar contour point, is the target position coordinate of the high-definition camera, and m is the number of contour points selected by the lidar.

[0009] Furthermore, the intelligent analysis module uses a deep learning algorithm to perform behavior analysis and prediction on target information, and the specific steps include: Set the target object information output by the data processing module at time t as vector , set in the time period N time step data are collected to form an input data sequence , shape is (N, 5); Construct a neural network model consisting of an LSTM hidden layer and a fully connected output layer. The LSTM hidden layer has H units and the input data sequence Input into the LSTM unit in sequence. At each time step t, the calculation of the LSTM unit includes the forget gate, input gate, cell state update and output gate, and finally the current hidden state is obtained. , containing historical behavior information of the target object; The fully connected output layer has two neurons, which are used to output the probability P of the target object mistakenly entering the construction area and the remaining warning time. , the calculation formula of the false entry probability output P is: in, is the weight matrix connecting the LSTM hidden layer and the false entry probability output neuron, is the bias vector; Remaining warning time The calculation formula is: in, is the weight matrix connecting the LSTM hidden layer and the remaining warning time output neurons, is the bias vector.

[0010] Furthermore, the intelligent analysis module is provided with a false entry probability threshold and minimum warning time threshold , if the model outputs the probability of false entry And the remaining warning time , it is determined that the target object has the potential risk of accidentally entering the construction area and an early warning signal is immediately issued.

[0011] Furthermore, the warning information release module includes an audible and visual alarm, a wireless communication unit and a wearable smart terminal. The audible and visual alarm is used to provide audible and visual warnings to on-site personnel and nearby vehicles. The wireless communication unit is used to send warning information to the on-board navigation system of a passing vehicle or the driver's mobile phone. The wearable smart terminal is used to send warning information to construction personnel.

[0012] Furthermore, the wireless communication unit adopts 4G / 5G communication technology to classify and customize the warning information according to the distance between the vehicle and the construction area and the driving direction when sending the warning information to the vehicle navigation system of the passing vehicle or the driver's mobile phone; For vehicles that are close to the construction area and traveling toward the construction area, high-priority detailed warning information will be sent, including the specific location of the construction area, a description of the degree of danger, and a recommended detour route; for vehicles that are far away or traveling away from the construction area, low-priority brief warning information will be sent, which only indicates the approximate location of the construction area and safety tips.

[0013] Furthermore, during the training process, the neural network model adopts an adaptive learning rate adjustment strategy to dynamically adjust the learning rate according to the change of the loss function during the training process, and introduces an L2 regularization term to prevent overfitting. The coefficient of the L2 regularization term is , the loss function calculation formula after adding the regularization term is: in, is the new loss function, is the original loss function, is the weight parameter of the neural network model.

[0014] Compared with the known prior art, the technical solution provided by the present invention has the following beneficial effects: (1) The present invention uses the collaborative work of laser radar, millimeter-wave radar and high-definition camera and adopts advanced data layer fusion algorithm to accurately obtain information such as the speed, acceleration, driving direction and position coordinates of the target object on the two-dimensional plane. This multi-sensor fusion method overcomes the limitations of a single sensor. The high-precision distance measurement of laser radar is combined with the accurate speed detection of millimeter-wave radar, and the high-definition camera can clearly capture the target's appearance features. This makes the system more accurate in grasping the motion state and position information of the target object, providing a reliable data basis for subsequent intelligent analysis and early warning.

[0015] (2) The intelligent analysis module of the present invention adopts a neural network model including an LSTM hidden layer and a fully connected output layer, which can effectively analyze and predict the behavior of the target object. Based on the historical behavior information of the target object, the deep learning algorithm can accurately determine whether the target has the risk of mistakenly entering the construction area, and calculate the probability of mistaken entry and the remaining warning time. This intelligent analysis method can predict potential dangers in advance, greatly improving the timeliness and accuracy of warnings. Compared with the traditional simple rule-based judgment method, it has stronger adaptability and reliability, and can better cope with complex and changing traffic scenes and target behaviors.

[0016] (3) The warning information release module of the present invention is equipped with an audible and visual alarm, a wireless communication unit and a wearable smart terminal, which can achieve all-round warning coverage for different objects. The audible and visual alarm can provide intuitive and strong audible and visual warnings to nearby personnel and vehicles at the construction site to attract their immediate attention; the wireless communication unit can send warning information to the on-board navigation system of passing vehicles or the driver's mobile phone, so that the driver can understand the construction situation ahead in advance and make corresponding driving decisions; the wearable smart terminal ensures that construction personnel can receive warning information in time to ensure their own safety. This diversified warning method can adapt to the information transmission needs in different scenarios, improve the accessibility and effectiveness of warning information, and minimize the possibility of accidental intrusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0020] The present invention will be further described below in conjunction with the embodiments. Example

[0021] Reference Figure 1 , an intelligent early warning system for preventing accidental intrusion in highway reconstruction and expansion projects, including a sensor module, a data processing module, an intelligent analysis module and an early warning information release module; The sensor modules include LiDAR, millimeter-wave radar, and high-definition cameras to collect environmental data around the construction area; The data processing module is connected to the sensor module and is used to perform fusion processing on the data collected by the sensor module and extract target information; The intelligent analysis module is connected to the data processing module to analyze and predict the target information and determine whether the target has the risk of accidentally entering the construction area; The early warning information release module is connected to the intelligent analysis module and is used to release early warning information to construction personnel and passing vehicles according to the early warning signals of the intelligent analysis module.

[0022] In a specific embodiment, the laser radar is installed in a high position around the construction area, such as the top of a street light pole or a specially built bracket, to ensure that its scanning range can cover the construction area and surrounding areas where vehicles and pedestrians may appear, and the field of view is not overly obstructed. The installation angle needs to be adjusted according to the shape of the construction area and the direction of surrounding roads so that it can obtain the most comprehensive and accurate three-dimensional contour and distance information.

[0023] The millimeter-wave radar is installed in a position close to the laser radar to ensure that its detection range has a certain overlap area with the laser radar to facilitate better data fusion. The installation direction of the millimeter-wave radar needs to be precisely calibrated so that it can accurately detect the speed and position information of the moving target.

[0024] High-definition cameras are installed in locations where they can clearly capture the surrounding conditions of the construction area, such as on the exterior walls of nearby buildings, gantries, and other high locations. The camera's field of view must also cover key areas, such as construction entrances, road bends, and other places where accidental intrusions are prone to occur. The camera needs to be fixed and anti-shake processed to ensure that the captured image is stable and clear.

[0025] Furthermore, the laser radar is used to scan the three-dimensional contours and distance information of objects around the construction area, the millimeter wave radar is used to detect the speed and position of the moving target, and the high-definition camera is used to obtain the appearance feature information of the target.

[0026] Furthermore, the target information is extracted including the speed of the target object. , acceleration , Driving direction And the position coordinates of the target object on the two-dimensional plane ; The acquisition of target information adopts the data layer fusion algorithm, and the specific steps include: S100, pre-processes and synchronizes data collected by laser radar, millimeter wave radar and high-definition camera; Perform timestamp calibration on the data collected by the laser radar, millimeter-wave radar and high-definition camera. The laser radar, millimeter-wave radar and high-definition camera are equipped with high-precision clocks. Time synchronization calibration is performed when the system starts. The network time protocol is used for clock synchronization. During the data collection process, the clock deviation of each sensor is monitored. If the deviation exceeds the preset threshold, the timestamp is corrected. The data collected by LiDAR, millimeter-wave radar and high-definition camera are screened. For LiDAR data, the signal strength of each scanning point is checked, and the scanning point data with signal strength below the preset strength threshold is eliminated; for millimeter-wave radar data, the data with speeds several times greater than the road speed limit or distance value jumps are eliminated; the image data of the high-definition camera is preprocessed, including image denoising, image enhancement, and detection and segmentation of target objects; S200, obtain the speed of the target object ; At the same timestamp, obtain the velocity component calculated by the lidar based on the position change of the vehicle contour point , , ..., , the velocity estimate of the lidar is obtained by vector synthesis , the calculation formula is: Where p is the number of selected contour points; At the same time, obtain the velocity value of the millimeter wave radar based on the Doppler effect and the speed value of the HD camera through frame-to-frame analysis ; right , and Perform weight allocation and set initial weight values ​​according to the nominal accuracy and reliability of LiDAR, millimeter-wave radar, and high-definition camera. The weight of LiDAR is , the weight of millimeter wave radar is , the HD camera weight is ,and ; The adjusted weights are used to calculate the fused speed value ; S300, obtaining the acceleration of the target object ; The speed value after fusion Get the fusion velocity sequence , , ..., , the fused speed sequence is smoothed, and the moving average filtering method is used to calculate the moving average value of the window size k as the smoothed speed value. The calculation formula is: For the smoothed velocity series , , ..., , using the central difference method, calculate the acceleration , the calculation formula is: in, is the time step, i.e., the interval between adjacent timestamps; S400: Obtaining the driving direction of the target object ; Standardize and evaluate the reliability of the direction vector. The direction vector obtained by the lidar After being calculated based on the change in the position of the contour points, it is standardized so that its modulus is unit length. The calculation formula is: The direction vector obtained by the millimeter wave radar according to the angle change of the target object , combining the angle measurement accuracy and the distance information of the target object to evaluate the reliability. If the target object is close and the angle measurement accuracy is high, a higher reliability weight is given; if the target object is far away or is at the edge of the radar detection area, the angle measurement accuracy decreases, and its reliability weight is reduced; The HD camera analyzes the displacement direction of the target object from the image to obtain the direction vector , the reliability is evaluated according to the position, size and resolution of the target object in the image; for example, if the target object occupies a large area in the image and its position is clear and definite, and the displacement direction is judged more accurately, a higher reliability weight is given; conversely, if the target object is small in the image and is affected by occlusion and other conditions, and there is a large uncertainty in the displacement direction judgment, its reliability weight is reduced.

[0027] According to the reliability evaluation results of each direction vector, the weighting coefficient is determined. The weighting coefficient of the laser radar direction vector is: , the weighting coefficient of the millimeter wave radar direction vector is , the weight coefficient of the HD camera direction vector is ,and ; Calculate the fused driving direction vector , the calculation format is: Then by calculating The fused driving direction is obtained by the direction angle, and the calculation formula is: in, and yes The x and y components of S500, obtaining the position coordinates of the target object on the two-dimensional plane ; Design a hierarchical storage structure, use timestamp as index, and create a two-dimensional array T. Store timestamp as Data records, Store the distance and angle information of the laser radar, using differential coding compression storage. Store the speed, distance and angle data of the millimeter wave radar, using quantized coding and compressed storage. Store the coordinate information of the target position of the high-definition camera image; At each timestamp, the distance and angle information of the laser radar is converted into coordinate values ​​in the rectangular coordinate system, and fused with the target position coordinates of the high-definition camera image to obtain the fused comprehensive position coordinates , the calculation formula is: in, is the coordinate of the laser radar contour point, is the target position coordinate of the high-definition camera, and m is the number of contour points selected by the lidar.

[0028] Furthermore, the intelligent analysis module uses a deep learning algorithm to perform behavioral analysis and prediction on target information. The specific steps include: Set the target object information output by the data processing module at time t as vector , set in the time period N time step data are collected to form an input data sequence , shape is (N, 5); Construct a neural network model consisting of an LSTM hidden layer and a fully connected output layer. The LSTM hidden layer has H units and the input data sequence Input into the LSTM unit in sequence. At each time step t, the calculation of the LSTM unit includes the forget gate, input gate, cell state update and output gate, and finally the current hidden state is obtained. , containing historical behavior information of the target object; The fully connected output layer has two neurons, which are used to output the probability P of the target object mistakenly entering the construction area and the remaining warning time. , the calculation formula of the false entry probability output P is: in, is the weight matrix connecting the LSTM hidden layer and the false entry probability output neuron, is the bias vector; Remaining warning time The calculation formula is: in, is the weight matrix connecting the LSTM hidden layer and the remaining warning time output neurons, is the bias vector.

[0029] Furthermore, the intelligent analysis module is set with a false entry probability threshold and minimum warning time threshold , if the model outputs the probability of false entry And the remaining warning time , it is determined that the target object has the potential risk of accidentally entering the construction area and an early warning signal is immediately issued.

[0030] Furthermore, the warning information release module includes an audible and visual alarm, a wireless communication unit and a wearable smart terminal. The audible and visual alarm is used to provide audible and visual warnings to on-site personnel and nearby vehicles. The wireless communication unit is used to send warning information to the on-board navigation system of a passing vehicle or the driver's mobile phone. The wearable smart terminal is used to send warning information to construction personnel.

[0031] In a specific embodiment, the sound and light alarm is installed at key locations such as the boundary and entrance of the construction area to ensure that its sound and light can cover the construction site and personnel and vehicles within a certain range around it. The sound and light alarm is connected to the early warning information release module through a wired or wireless connection, receives the early warning signal and promptly issues a strong sound and light warning to remind on-site personnel and vehicles to pay attention to safety and avoid accidentally entering the construction area.

[0032] The wireless communication unit adopts a 4G / 5G communication module and is installed near the data processing module or the warning information release module to establish a stable communication connection with them. The wireless communication unit establishes a communication link with the on-board navigation system of a passing vehicle or the driver's mobile phone through the operator's network infrastructure to achieve remote sending of warning information. At the same time, in order to ensure the stability and reliability of communication, the wireless communication unit is equipped with a high-gain antenna and supports functions such as signal strength monitoring and automatic switching of backup frequency bands to cope with different network environments and signal interference conditions.

[0033] Wearable smart terminals are distributed to construction workers, such as smart bracelets or special devices built into helmets. These terminals are connected to the early warning information release module via Bluetooth or Wi-Fi to receive and display early warning information, such as dangerous conditions in the construction area, potential risk warnings, etc., to ensure that construction workers can keep abreast of the dynamics around the construction site and take corresponding safety measures.

[0034] Furthermore, the wireless communication unit uses 4G / 5G communication technology to classify and customize the warning information according to the distance between the vehicle and the construction area and the direction of travel when sending warning information to the on-board navigation system of the passing vehicle or the driver's mobile phone; For vehicles that are close to the construction area and traveling toward the construction area, high-priority detailed warning information will be sent, including the specific location of the construction area, a description of the degree of danger, and a recommended detour route; for vehicles that are far away or traveling away from the construction area, low-priority brief warning information will be sent, which only indicates the approximate location of the construction area and safety tips.

[0035] In a specific embodiment, the warning information release program runs in the warning information release module, and monitors the warning signal sent by the intelligent analysis module in real time. When the warning signal is received, the receiving object and sending method of the warning information are first determined based on the target object information and the geographical information of the construction area in the warning signal. For the sound and light alarm on site, a trigger signal is directly sent to make it emit a strong sound and light warning to remind on-site personnel and nearby vehicles to pay attention to safety; for passing vehicles, the warning information release program establishes a communication connection with the vehicle's on-board navigation system or the driver's mobile phone through a wireless communication unit, and classifies and customizes the warning information according to the distance between the vehicle and the construction area and the direction of travel; for vehicles that are close to the construction area, less than 100 meters away and are traveling towards the construction area, a high-speed alarm is sent. Detailed warning information of priority level includes the specific location of the construction area (such as through map coordinates or text description), description of the degree of danger (such as "large-scale machinery operation is in progress, the degree of danger is high") and recommended detour routes (such as "turn right 50 meters ahead and detour to other roads"); for vehicles that are far away, greater than 500 meters, or driving in the opposite direction from the construction area, a low-priority brief warning message is sent, which only prompts the approximate location of the construction area (such as "there is road construction 1 kilometer ahead, please drive carefully") and safety tips (such as "be careful to avoid construction vehicles and personnel"); for the wearable smart terminals of construction workers, the warning information release program sends the warning information to the terminals in the form of text, vibration or voice, to ensure that construction workers can understand potential dangerous situations in a timely manner and take corresponding protective measures.

[0036] Furthermore, during the training process of the neural network model, an adaptive learning rate adjustment strategy is adopted to dynamically adjust the learning rate according to the change of the loss function during the training process, and an L2 regularization term is introduced to prevent overfitting. The coefficient of the L2 regularization term is , the loss function calculation formula after adding the regularization term is: in, is the new loss function, is the original loss function, is the weight parameter of the neural network model.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects, characterized in that: It includes sensor module, data processing module, intelligent analysis module and early warning information release module; The sensor module includes a laser radar, a millimeter wave radar and a high-definition camera, which are used to collect environmental data around the construction area; The data processing module is connected to the sensor module and is used to perform fusion processing on the data collected by the sensor module and extract target information; The intelligent analysis module is connected to the data processing module and is used to perform behavior analysis and prediction on the target information to determine whether the target has a risk of accidentally entering the construction area; The early warning information issuing module is connected to the intelligent analysis module and is used to issue early warning information to construction personnel and passing vehicles according to the early warning signal of the intelligent analysis module.

2. According to claim 1, the intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects is characterized in that: The laser radar is used to scan the three-dimensional contours and distance information of objects around the construction area, the millimeter wave radar is used to detect the speed and position of the moving target, and the high-definition camera is used to obtain the appearance feature information of the target.

3. According to claim 1, the intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects is characterized in that: The target information extracted includes the speed of the target object , acceleration , Driving direction And the position coordinates of the target object on the two-dimensional plane ; The acquisition of the target information adopts a data layer fusion algorithm, and the specific steps include: S100, pre-processes and synchronizes data collected by laser radar, millimeter wave radar and high-definition camera; Perform timestamp calibration on the data collected by the laser radar, millimeter-wave radar and high-definition camera. The laser radar, millimeter-wave radar and high-definition camera are equipped with high-precision clocks. Time synchronization calibration is performed when the system starts. The network time protocol is used for clock synchronization. During the data collection process, the clock deviation of each sensor is monitored. If the deviation exceeds the preset threshold, the timestamp is corrected. The data collected by LiDAR, millimeter-wave radar and high-definition camera are screened. For LiDAR data, the signal strength of each scanning point is checked, and the scanning point data with signal strength below the preset strength threshold is eliminated; for millimeter-wave radar data, the data with speeds several times greater than the road speed limit or distance value jumps are eliminated; the image data of the high-definition camera is preprocessed, including image denoising, image enhancement, and detection and segmentation of target objects; S200, obtain the speed of the target object ; At the same timestamp, obtain the velocity component calculated by the lidar based on the position change of the vehicle contour point , , ..., , the velocity estimate of the lidar is obtained by vector synthesis , the calculation formula is: Where p is the number of selected contour points; At the same time, obtain the velocity value of the millimeter wave radar based on the Doppler effect and the speed value of the HD camera through frame-to-frame analysis ; right , and Perform weight allocation and set initial weight values ​​according to the nominal accuracy and reliability of LiDAR, millimeter-wave radar, and high-definition camera. The weight of LiDAR is , the weight of millimeter wave radar is , the HD camera weight is ,and ; The adjusted weights are used to calculate the fused speed value ; S300, obtaining the acceleration of the target object ; The speed value after fusion Get the fusion velocity sequence , , ..., , the fused speed sequence is smoothed, and the moving average filtering method is used to calculate the moving average value of the window size k as the smoothed speed value. The calculation formula is: For the smoothed velocity series , , ..., , using the central difference method, calculate the acceleration , the calculation formula is: in, is the time step, i.e., the interval between adjacent timestamps; S400: Obtaining the driving direction of the target object ; Standardize and evaluate the reliability of the direction vector. The direction vector obtained by the lidar After being calculated based on the change in the position of the contour points, it is standardized so that its modulus is unit length. The calculation formula is: The direction vector obtained by the millimeter wave radar according to the angle change of the target object ,Combining the angle measurement accuracy and the distance information of the target object, the reliability is evaluated; The HD camera analyzes the displacement direction of the target object from the image to obtain the direction vector ,Reliability is evaluated based on the location and size of the target object in the image and the resolution of the image; According to the reliability evaluation results of each direction vector, the weighting coefficient is determined. The weighting coefficient of the laser radar direction vector is: , the weighting coefficient of the millimeter wave radar direction vector is , the weighting coefficient of the HD camera direction vector is ,and ; Calculate the fused driving direction vector , the calculation format is: Then calculate The fused driving direction is obtained by the direction angle, and the calculation formula is: in, and yes The x and y components of S500, obtaining the position coordinates of the target object on the two-dimensional plane ; Design a hierarchical storage structure, use timestamp as index, and create a two-dimensional array T. Store timestamp as Data records, Store the distance and angle information of the laser radar, using differential coding compression storage. Store the speed, distance and angle data of the millimeter wave radar, using quantized coding and compressed storage. Store the coordinate information of the target position of the high-definition camera image; At each timestamp, the distance and angle information of the laser radar is converted into coordinate values ​​in the rectangular coordinate system, and fused with the target position coordinates of the high-definition camera image to obtain the fused comprehensive position coordinates , the calculation formula is: in, is the coordinate of the laser radar contour point, is the target position coordinate of the high-definition camera, and m is the number of contour points selected by the lidar.

4. According to claim 1, the intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects is characterized in that: The intelligent analysis module uses a deep learning algorithm to perform behavior analysis and prediction on target information. The specific steps include: Set the target object information output by the data processing module at time t as vector , set in the time period N time step data are collected to form an input data sequence , shape is (N, 5); Construct a neural network model consisting of an LSTM hidden layer and a fully connected output layer. The LSTM hidden layer has H units and the input data sequence Input into the LSTM unit in sequence. At each time step t, the calculation of the LSTM unit includes the forget gate, input gate, cell state update and output gate, and finally the current hidden state is obtained. , containing historical behavior information of the target object; The fully connected output layer has two neurons, which are used to output the probability P of the target object mistakenly entering the construction area and the remaining warning time. , the calculation formula of the false entry probability output P is: in, is the weight matrix connecting the LSTM hidden layer and the false entry probability output neuron, is the bias vector; Remaining warning time The calculation formula is: in, is the weight matrix connecting the LSTM hidden layer and the remaining warning time output neurons, is the bias vector.

5. According to claim 1, the intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects is characterized in that: The intelligent analysis module is provided with a false entry probability threshold and minimum warning time threshold , if the model outputs the probability of false entry And the remaining warning time , it is determined that the target object has the potential risk of accidentally entering the construction area and an early warning signal is immediately issued.

6. The intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects according to claim 1 is characterized in that: The early warning information release module includes an audible and visual alarm, a wireless communication unit and a wearable smart terminal. The audible and visual alarm is used to provide audible and visual warnings to on-site personnel and nearby vehicles. The wireless communication unit is used to send early warning information to the on-board navigation system of a passing vehicle or the driver's mobile phone. The wearable smart terminal is used to send early warning information to construction personnel.

7. The intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects according to claim 6 is characterized in that: The wireless communication unit uses 4G / 5G communication technology to classify and customize the warning information according to the distance between the vehicle and the construction area and the driving direction when sending warning information to the vehicle navigation system of the passing vehicle or the driver's mobile phone; For vehicles that are close to the construction area and traveling toward the construction area, high-priority detailed warning information will be sent, including the specific location of the construction area, a description of the degree of danger, and a recommended detour route; for vehicles that are far away or traveling away from the construction area, low-priority brief warning information will be sent, which only indicates the approximate location of the construction area and safety tips.

8. The intelligent early warning system for preventing accidental entry in highway reconstruction and expansion projects according to claim 4 is characterized in that: During the training process, the neural network model adopts an adaptive learning rate adjustment strategy to dynamically adjust the learning rate according to the change of the loss function during the training process, and introduces the L2 regularization term to prevent overfitting. The coefficient of the L2 regularization term is , the loss function calculation formula after adding the regularization term is: in, is the new loss function, is the original loss function, is the weight parameter of the neural network model.

Images