Intelligent safety helmet capable of identifying building cracks

Abstract

The application relates to an intelligent safety helmet capable of identifying building cracks, and belongs to the technical field of video image processing, and solves the problem that the safety helmet cannot identify and record building cracks in real time in the prior art. The intelligent safety helmet comprises a camera module, a small integrated module, a micro projection module and goggles. The camera module comprises a camera unit and a processing unit. The camera unit is used for acquiring building images and sending the images to the processing unit. The processing unit identifies crack position information in the images based on an internal neural network, generates images with detection frames, and outputs the images to the small integrated module. The small integrated module stores the images with the detection frames, generates transparent images with the detection frames based on the crack position information, and delivers the transparent images to the micro projection module. The micro projection module projects the images to the goggles. The application realizes an intelligent safety helmet capable of identifying building cracks, storing and uploading specific positions, crack lengths and detection frames.

Description

Technical Field

[0001] This invention relates to the field of video image processing technology, and in particular to a smart safety helmet capable of identifying cracks in buildings. Background Technology

[0002] In terms of engineering quality inspection and acceptance of houses or structures and quality control at the construction site, cracking, especially in concrete, is not allowed.

[0003] Existing solutions primarily rely on manual inspections and traditional monitoring methods. Manual inspections involve specialized supervisors or quality management personnel from the construction unit periodically patrolling the construction site to check for cracks in beams, columns, slabs, or other components, recording and organizing the findings. This method is not only susceptible to subjective human factors, but also makes it difficult to comprehensively assess the overall construction quality through random sampling. Furthermore, the complex environment of construction sites and the sheer volume of crack inspection work required on-site are significant. Traditional monitoring methods use measuring tools such as rangefinders, levels, and plumb lines to measure key parameters like crack length, width, and depth. They may also use magnifying glasses or photographic equipment to capture crack details and record them for later analysis. This method often requires substantial equipment and specialized skills, resulting in high costs and time investment.

[0004] Existing smart safety helmets are primarily data-driven protective gear. Compared to traditional data collection methods, smart safety helmets eliminate the need for separate equipment, making them more convenient and practical. They often integrate multiple sensors to collect various data about workers during their work, such as posture, fatigue levels, and environmental factors like temperature, humidity, and lighting. Currently, most smart safety helmets on the market only offer protection, alarms, and video monitoring functions; a fully intelligent helmet capable of real-time detection and recording of building cracks is lacking. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a smart safety helmet that can identify building cracks, in order to solve the problem that existing smart safety helmets only have functions such as protection, alarm, video monitoring, and wireless communication, but lack the function of real-time capture and recording of building cracks.

[0006] On one hand, embodiments of the present invention provide a method for identifying building cracks using an intelligent safety helmet, the safety helmet comprising: a camera module, a small integrated module, a micro projection module, and goggles;

[0007] The camera module includes a camera unit and a processing unit;

[0008] The camera unit is used to acquire images of the building and send the images to the processing unit. The processing unit identifies the location information of cracks in the image based on the internally running neural network. The processing unit generates an image with a detection frame based on the crack location information and outputs it to the small integrated module. The small integrated module stores the image with the detection frame and generates a transparent image with the detection frame based on the crack location information and transmits it to the micro projection module.

[0009] The miniature projection module projects images onto the goggles.

[0010] Furthermore, the camera unit includes: a binocular depth camera and a high-definition RGB camera;

[0011] The high-definition RGB camera is used to capture high-definition building images and send the images to the processing unit;

[0012] The binocular depth camera is a black and white camera, located on the left and right sides of the high-definition RGB camera above the brim of the safety helmet. The binocular depth camera sends the captured black and white images to the processing unit.

[0013] The processing unit is used to identify cracks in high-definition building images, and based on the black and white images, it measures the distance information between the wearer and the crack using a binocular ranging method and outputs it to a small integrated module.

[0014] Furthermore, the processing unit is also used to calculate the length of cracks in the high-definition building image using a pixel-filling method, and output the length information to the small integrated module; the pixel-filling method includes:

[0015] The high-definition building images captured by the high-definition RGB camera are converted into grayscale images;

[0016] Binarize the grayscale image;

[0017] The location of the crack is determined based on the detection frame;

[0018] Set the pixel values ​​outside the detection box to black;

[0019] Use Gaussian filtering to filter the image and remove noise;

[0020] The crack neighborhood is expanded using a morphological dilation algorithm.

[0021] The pixel length of the crack can be obtained by determining the pixel position occupied by the white pixels. The length of the crack can be calculated based on the ratio mapping between the image and reality.

[0022] Furthermore, the processing unit embeds a blob-formatted neural network based on YOLOv8 for intelligent crack identification; the YOLOv8-based neural network includes a backbone structure and a neck structure; the backbone structure extracts sampled feature maps at 32x, 16x, 8x, and 4x magnification and inputs them into the neck structure; the neck structure processes the sampled feature maps at each magnification and outputs the crack location information and confidence level corresponding to each magnification sampled feature map; in the neck structure of the YOLOv8-based neural network, VoVGSCSP replaces C2f in the YOLOv8 neck structure.

[0023] Furthermore, the YOLOv8-based improved neural network employs the EIoU loss function, including:

[0024]

[0025]

[0026] v = (q2 - q1) 2 ,

[0027]

[0028]

[0029]

[0030] Where A is the actual crack detection frame, B is the predicted crack detection frame, S is the area of ​​the detection frame; w1, h1 are the width and height of detection frame A, w2, h2 are the width and height of detection frame B, d is the Euclidean distance between the center points of the two detection frames, and c is the distance between the diagonals of the smallest rectangular closed region that can simultaneously contain the two frames.

[0031] Furthermore, the safety helmet also includes a sensor integration module, which is located at the top inside the safety helmet;

[0032] The sensor integration module includes a three-axis gyroscope, a three-axis accelerometer, a three-axis magnetometer, a temperature sensor, and a barometric pressure sensor.

[0033] The three-axis gyroscope, three-axis accelerometer, and three-axis magnetometer are used to acquire the wearer's posture information;

[0034] The sensor integration module sends the measured data to the small integration module.

[0035] Furthermore, the small integrated module includes: an ARM processor, a TF memory, and a GPS positioning module;

[0036] The ARM processor is used to decode the images transmitted by the processing unit, the data acquired by the sensor integration module, and the positioning data acquired by the GPS positioning module, and send them to the TF memory for storage.

[0037] The ARM processor is also used to overlay the distance information between the wearer and the crack, the crack length information, the positioning data obtained by the GPS positioning module, and the temperature and air pressure data measured by the sensor integration module onto a transparent image with a detection frame, and transmit it to the micro projection module via the HDMI interface.

[0038] Furthermore, the small integrated module also includes: a wireless communication module;

[0039] The wireless communication module includes a Bluetooth module and a WIFI module;

[0040] The Bluetooth module is used to connect the safety helmet and the mobile terminal and transmit data information in the TF memory;

[0041] The WIFI module is used to upload data from the TF memory to the cloud, send the audio signal of the helmet wearer to the helmets of other wearers and / or the command center, receive audio signals sent by other wearers through their helmets and / or the audio signals sent by the command center, and demodulate the received audio signals and transmit them to the ARM processor.

[0042] Furthermore, the helmet also includes a chin strap, a microphone, and a speaker. The chin strap has a wire, and the microphone and speaker are mounted on the chin strap and electrically connected to a small integrated module via the wire.

[0043] The small integrated module also includes: an audio module;

[0044] The ARM processor unpacks the received audio signal and restores it to the original digital audio signal;

[0045] The audio module is used to perform digital-to-analog conversion on the digital audio signal, adjust the volume, and output it to the speaker for playback.

[0046] The audio module is also used to perform analog-to-digital conversion, noise reduction, filtering, and equalization processing on the audio signal received from the microphone from the wearer, and then transmit it to the ARM processor for packaging before passing it to the WIFI module.

[0047] Furthermore, the safety helmet also includes an integrated signal transceiver located on the top of the helmet's exterior. This integrated transceiver is connected to a Wi-Fi module and a Bluetooth module via wires. It receives data that the Wi-Fi module needs to send to the cloud, other wearers' helmets, and the command center, amplifies the data, and then transmits it. It also receives data sent by the Bluetooth module to a mobile terminal, amplifies the data, and then transmits it. Additionally, it receives data from other wearers' helmets and data sent to the helmet by the command center, amplifies the data, and then sends it to the Wi-Fi module.

[0048] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0049] 1. The processing unit of the camera module identifies building cracks in high-definition building images captured by the high-definition RGB camera in real time, and identifies the location information of cracks in the image through the neural network running inside the processing unit. The processing unit measures the distance information between the wearer and the crack from the black and white image captured by the binocular depth camera using the binocular ranging method, calculates the length of the crack in the high-definition building image using the pixel filling method, and superimposes the crack detection box, the distance information between the wearer and the crack, and the crack length information onto the high-definition RGB image and passes it to the TF memory of the small integrated module for storage.

[0050] 2. The neural network running inside the processing unit introduces a 4x downsampling detector on the basis of the original YOLOv8's three-scale detector, which retains more detailed small-pixel target information in the feature mapping of the large-scale detector with low-dimensional high-frequency information; on the basis of the original YOLOv8, the activation function SiLU is replaced with h-swish, which reduces the computational overhead; the neck structure of the neural network based on YOLOv8 includes the cross-local network module VoVGSCSP replacing the C2f layer in the original YOLOv8 neck architecture, which reduces the computational complexity of the model and maintains its accuracy.

[0051] 3. In the original YOLOv8, the robustness of v in CIoU is weak, which leads to large variations in the value of the loss function and affects its performance. Furthermore, the use of the tangent function in v cannot directly meet the requirements for loss function normalization. Coefficients need to be introduced to achieve normalization, which increases the computational difficulty. Therefore, CIoU is improved to obtain a more efficient loss function eIoU.

[0052] 4. The small integrated module generates a transparent image containing a crack detection frame, distance information between the wearer and the crack, crack length information, positioning data, temperature, and air pressure data. This image is transmitted to the micro-projection module via an HDMI interface and projected onto the goggles. The wearer can then observe the augmented reality image through the goggles.

[0053] 5. The safety helmet of the present invention has a WIFI wireless communication module, which can upload data in the TF memory to the cloud and send the audio signal of the helmet wearer to the helmets of other wearers and / or the command center, and receive audio signals sent by other wearers through the helmets and / or the command center; it also has a Bluetooth wireless communication module to connect the helmet and the mobile terminal and transmit data information in the TF memory.

[0054] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0055] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0056] Figure 1 A bottom view of a smart safety helmet that can detect cracks in buildings;

[0057] Figure 2 A side view of a smart safety helmet that can identify cracks in buildings;

[0058] Figure 3 A cross-sectional view of a smart safety helmet that can detect cracks in buildings;

[0059] Figure 4 A front view of a smart safety helmet that can identify cracks in buildings;

[0060] Figure 5 A structural block diagram of a smart safety helmet that can identify cracks in buildings;

[0061] Figure 6 A smart safety helmet that can identify cracks in buildings is based on a neural network improved by YOLOv8.

[0062] Figure label:

[0063] 1- Protective casing;

[0064] 2-Brim;

[0065] 3-Memory foam padding;

[0066] 4-Mandibular band;

[0067] 5- Sensor integration module;

[0068] 6-Integrated signal receiver;

[0069] 7-Camera module;

[0070] 8. Safety goggles;

[0071] 9-Miniature projection module;

[0072] 10-Small integrated modules;

[0073] 11-Battery pack;

[0074] 12-Magnetic wireless charging module;

[0075] 13-Microphone;

[0076] 14 speakers;

[0077] 15-Dual-lens depth camera;

[0078] 16-High Definition RGB Camera;

[0079] 17-Metal sheet. Detailed Implementation

[0080] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0081] One specific embodiment of the present invention discloses a smart safety helmet capable of identifying cracks in buildings, such as... Figures 1-4 As shown.

[0082] A smart safety helmet capable of identifying cracks in buildings, the safety helmet comprising: a camera module, a small integrated module, a micro projection module, and goggles;

[0083] The camera module includes a camera unit and a processing unit;

[0084] The camera unit is used to acquire images of the building and send the images to the processing unit. The processing unit identifies the location information of cracks in the image based on the internally running neural network. The processing unit generates an image with a detection frame based on the crack location information and outputs it to the small integrated module. The small integrated module stores the image with the detection frame and generates a transparent image with the detection frame based on the crack location information and transmits it to the micro projection module.

[0085] The miniature projection module projects images onto the goggles using optical waveguide technology.

[0086] The camera unit includes: a binocular depth camera and a high-definition RGB camera;

[0087] The high-definition RGB camera is used to capture high-definition building images and send the images to the processing unit;

[0088] The binocular depth camera is a black and white camera, located on the left and right sides of the high-definition RGB camera above the brim of the safety helmet. The binocular depth camera sends the captured black and white images to the processing unit.

[0089] The processing unit is used to identify cracks in high-definition building images, and based on the black and white images, it measures the distance information between the wearer and the crack using a binocular ranging method and outputs it to a small integrated module.

[0090] The processing unit is also used to calculate the length of cracks in high-definition building images using a pixel-filling method, and output the length information to a small integrated module; the pixel-filling method includes:

[0091] The high-definition building images captured by the high-definition RGB camera are converted into grayscale images;

[0092] Binarize the grayscale image;

[0093] The location of the crack is determined based on the detection frame;

[0094] Set the pixel values ​​outside the detection box to black;

[0095] Use Gaussian filtering to filter the image and remove noise;

[0096] The crack neighborhood is expanded using a morphological dilation algorithm.

[0097] The pixel length of the crack can be obtained by determining the pixel position occupied by the white pixels. The length of the crack can be calculated based on the ratio mapping between the image and reality.

[0098] Specifically, the camera module's processing unit processes the photos in real time at a rate of 25 frames per second. Then, it inputs the distance information between the wearer and the crack and the crack length information in each frame into the small integrated module via a TYPEC to USB data cable. At the same time, it also inputs the processed image into the small integrated module.

[0099] The processing unit embeds a YOLOv8-based improved neural network in blob format for intelligent crack identification. The YOLOv8-based improved neural network includes a backbone structure and a neck structure. The backbone structure extracts sampled feature maps at 32x, 16x, 8x, and 4x magnification and inputs them into the neck structure. The neck structure processes the sampled feature maps at each magnification and outputs the crack location information and confidence level corresponding to each magnification sampled feature map. In the neck structure of the YOLOv8-based improved neural network, VoVGSCSP replaces C2f in the YOLOv8 neck structure.

[0100] like Figure 6 As shown, the backbone structure sequentially includes a first CBH, a second CBH, a first C2f, a third CBH, a second C2f, a fourth CBH, a third C2f, a fifth CBH, a fourth C2f, and an SPPF; the first to third C2f are used to output 4x, 8x, and 16x sampling feature maps, respectively, and the SPPF is used to output 32x sampling feature maps;

[0101] In the neck structure of the improved YOLOv8 neural network, VoVGSCSP replaces C2f in the YOLOv8 neck structure, and CBH replaces Conv in YOLOv8.

[0102] The improved neural network convolutional module CBH consists of Conv2d, BN, and h-swish. h-swish replaces the SiLU activation function in YOLOv8's Conv, and h-swish is an activation function layer with added non-linearity. Conv2d is a two-dimensional convolutional layer, BN is a batch sample normalization layer, C2f is a connection layer, SPPF is a spatial pyramid pooling layer for decomposing convolutions, Concat in the neck structure is a connection layer used to fuse two feature maps, Upsample is used to recover the image size reduction and information loss caused by the previous downsampling convolution and pooling operations through upsampling, and VoVGSCSP is a cross-local network module used to improve network inference speed.

[0103] The neural network based on the improved YOLOv8 uses the EIoU loss function, including:

[0104]

[0105]

[0106] v = (q2 - q1) 2 ,

[0107]

[0108]

[0109]

[0110] Where A is the actual crack detection frame, B is the predicted crack detection frame, S is the area of ​​the detection frame; w1, h1 are the width and height of detection frame A, w2, h2 are the width and height of detection frame B, d is the Euclidean distance between the center points of the two detection frames, and c is the distance between the diagonals of the smallest rectangular closed region that can simultaneously contain the two frames.

[0111] like Figure 6 As shown, a neural network based on YOLOv8 generates four detection bounding boxes. Each detection bounding box includes the (x,y) coordinates of the four corners of the detection box and the confidence information calculated by the EIoU loss function. The detection bounding box with the highest confidence is adopted.

[0112] Specifically, the neural network needs to be trained before use. First, a training dataset of 504 images is collected and prepared, including images with cracks and manually labeled information. Then, the images are input into the neural network through forward propagation. The network generates candidates and calculates the loss function between the candidate and the ground truth bounding box to evaluate the accuracy of object detection. Through backpropagation, the network parameters are optimized based on the gradient of the loss function, enabling the network to better predict the location and category of the target. The training process requires 300 epochs of iteration until the predetermined training objective is achieved.

[0113] The helmet also includes a sensor integration module, which is located at the top inside the helmet;

[0114] The sensor integration module includes a three-axis gyroscope, a three-axis accelerometer, a three-axis magnetometer, a temperature sensor, and a barometric pressure sensor.

[0115] The three-axis gyroscope, three-axis accelerometer, and three-axis magnetometer are used to acquire the wearer's posture information;

[0116] The sensor integration module sends the measured data to the small integration module.

[0117] Specifically, the sensor integration module is a ten-axis sensor, and the raw data needs to be compensated before use.

[0118] The raw accelerometer data, a 16-bit signed integer, is read and zero-biased. The device is held stationary, and the accelerometer output is recorded for a period of time. The average value is calculated and used as the zero-bias compensation. The raw accelerometer data is then compensated using the calibration parameters to obtain the acceleration value.

[0119] The raw gyroscope data, a 16-bit signed integer, is read and zero-biased. The device is kept stationary, and the gyroscope output is recorded for a period of time. The average value is calculated and used as the zero-bias compensation. The raw gyroscope data is then compensated using the calibration parameters to obtain the angular velocity value.

[0120] The raw magnetometer data, in this specific embodiment a 16-bit signed integer, is read and calibrated for its magnetic field. This typically requires rotating the device in different orientations and positions to record a series of magnetometer output data. The raw magnetometer data is then calibrated using calibration parameters to eliminate nonlinearities and biases in the magnetic field.

[0121] Read the raw temperature data, which is a 20-bit signed integer. Calculate the temperature calibration parameters, including the calibration coefficient and calibration constant, based on the calibration data read directly from the sensor. Use the calibration parameters to convert the raw temperature data into a temperature value.

[0122] Read the raw air pressure data, which is a 20-bit unsigned integer. Calculate the calibration parameters based on the calibration data, including the calibration coefficients and calibration constants. Use these calibration parameters to convert the raw air pressure data into air pressure values.

[0123] Specifically, the small integrated module includes: an ARM processor, a TF memory, and a GPS positioning module;

[0124] The ARM processor is used to decode the images transmitted by the processing unit, the data acquired by the sensor integration module, and the positioning data acquired by the GPS positioning module, and send them to the TF memory for storage.

[0125] The ARM processor is also used to overlay the distance information between the wearer and the crack, the crack length information, the positioning data obtained by the GPS positioning module, and the temperature and air pressure data measured by the sensor integration module onto a transparent image with a detection frame, and transmit it to the micro projection module via the HDMI interface.

[0126] Specifically, the step of the ARM processor generating the transparent image to be passed to the micro-projection module includes S1-S5.

[0127] S1. Use the third-party library OpenCV to generate a completely white BGR image.

[0128] S2. Based on the (x,y) coordinate information of the four corners of the detection frame, the detection frame is superimposed on the all-white image generated in step S1. The crack length information, the distance information from the wearer to the crack, the positioning data obtained by the GPS positioning module, and the temperature and air pressure data measured by the sensor integration module are superimposed on the all-white image with the detection frame in red font.

[0129] S3. Convert the BGR format image to BGRA format image using the cv2.cvtColor function.

[0130] S4. Set the A channel pixel value of the white pixels to 0, that is, the white pixels become transparent, forming a transparent image that only contains red information.

[0131] Specifically, by using a for loop to search the pixels of the BGR image, all white pixels can be found.

[0132] S5. The miniature projection module receives the lens image generated by the small integrated module through the HDMI interface and projects the transparent image onto the goggles.

[0133] Specifically, when the micro-projection module does not receive image data from the ARM processor, it projects a transparent desktop that only displays the current time and day of the week onto the goggles; when the micro-projection module receives image data from the ARM processor, it projects the transparent image generated in step S4 onto the transparent desktop of the goggles that displays the time and day of the week, forming a transparent image that displays the time, day of the week, crack length information, distance from the person to the crack information, GPS positioning data, temperature and air pressure data, and a crack detection frame.

[0134] The small integrated module also includes: a wireless communication module;

[0135] The wireless communication module includes a Bluetooth module and a WIFI module;

[0136] The Bluetooth module is used to connect the safety helmet and the mobile terminal and transmit data information from the TF memory to the mobile terminal.

[0137] The WIFI module is used to upload data from the TF memory to the cloud, send the audio signal of the helmet wearer to the helmets of other wearers and / or the command center, receive audio signals sent by other wearers through their helmets and / or the audio signals sent by the command center, and demodulate the received audio signals and transmit them to the ARM processor.

[0138] The helmet also includes a chin strap, a microphone, and a speaker. The chin strap has a wire, and the microphone and speaker are mounted on the chin strap and electrically connected to a small integrated module via the wire.

[0139] The small integrated module also includes: an audio module;

[0140] The ARM processor unpacks the received audio signal and restores it to the original digital audio signal;

[0141] The audio module is used to perform digital-to-analog conversion on the digital audio signal, adjust the volume, and output it to the speaker for playback.

[0142] The audio module is also used to perform analog-to-digital conversion, noise reduction, filtering, and equalization processing on the audio signal received from the microphone from the wearer, and then transmit it to the ARM processor for packaging before passing it to the WIFI module.

[0143] The helmet also includes an integrated transceiver located on the top of the helmet's exterior. This transceiver connects to a Wi-Fi module and a Bluetooth module via wires. It receives data from the Wi-Fi module that needs to be sent to the cloud, other helmets worn by other users, and the command center, amplifies the data, and then transmits it. It also receives data from the Bluetooth module that is sent to a mobile terminal, amplifies the data, and then transmits it. Furthermore, it receives data from other helmets worn by other users and data from the command center sent to the helmet, amplifies the data, and then sends it to the Wi-Fi module.

[0144] Specifically, the chin strap also includes a buckle metal piece. When the buckle metal pieces at both ends of the chin strap are in contact, it is considered that the smart helmet has been worn correctly. Then, the power switch on the smart helmet is turned on to control the start and stop of the entire smart helmet.

[0145] Specifically, the helmet also includes a battery pack housed in battery compartments on both sides of the helmet. The battery pack powers the camera module and a small integrated module located inside the rear of the smart helmet. The camera module's processing unit is equipped with a Robotics Vision Core 2 processor comprised of a Myriad X VPU chip, designed for low-power, high-efficiency machine vision processing and neural network operation.

[0146] Specifically, the safety helmet also includes a magnetic wireless charging module, which is located on the rear outer side of the smart safety helmet; the magnetic wireless charging module is connected to a small integrated module and is used to power the small integrated module in an emergency.

[0147] Compared with existing technologies, this embodiment provides a smart safety helmet that can identify building cracks. The processing unit of the camera module identifies building cracks in high-definition building images captured by a high-definition RGB camera in real time, and identifies the location information of cracks in the image through a neural network running inside the processing unit. The processing unit measures the distance information between the wearer and the crack from the black and white image captured by the binocular depth camera using the binocular ranging method, calculates the length of the crack in the high-definition building image using the pixel filling method, and superimposes the crack detection frame, the distance information between the wearer and the crack, and the crack length information onto the high-definition RGB image and transmits it to the TF memory of the small integrated module for storage.

[0148] The neural network running inside the processing unit introduces a 4x downsampling detector on top of the original YOLOv8's three-scale detector, preserving more detailed small-pixel target information in the feature mapping of the low-dimensional, high-frequency large-scale detector. Based on the original YOLOv8, the activation function SiLU is replaced with h-swish, reducing computational overhead. The neck structure of the improved YOLOv8-based neural network includes the cross-local network module VoVGSCSP replacing the C2f layer in the original YOLOv8 neck architecture, reducing the computational complexity of the model while maintaining its accuracy.

[0149] In the original YOLOv8, the robustness of v in CIoU is weak, which leads to large variations in the value of the loss function and affects its performance. Furthermore, the use of the tangent function in v cannot directly meet the requirements for loss function normalization, and coefficients need to be introduced to achieve normalization, which increases the computational difficulty. Therefore, CIoU is improved to obtain a more efficient loss function eIoU.

[0150] The safety helmet of the present invention has a WIFI wireless communication module, which can upload data in the TF memory to the cloud and send the audio signal of the helmet wearer to the helmets of other wearers and / or the command center, and receive audio signals sent by other wearers through the helmets and / or the command center; it also has a Bluetooth wireless communication module to connect the helmet and the mobile terminal and transmit data information in the TF memory.

[0151] The safety helmet of the present invention generates a transparent image with a crack detection frame, distance information between the wearer and the crack, crack length information, positioning data obtained by a GPS positioning module, and temperature and air pressure data measured by a sensor integration module through a small integrated module. The image is then transmitted to a micro-projection module via an HDMI interface and projected onto the goggles, allowing the wearer to observe the augmented reality image through the goggles.

[0152] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0153] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart safety helmet capable of identifying cracks in buildings, characterized in that, The safety helmet includes: a camera module, a small integrated module, a micro projection module, and goggles; The camera module includes a camera unit and a processing unit; The camera unit is used to acquire images of the building and send the images to the processing unit. The processing unit identifies the location information of cracks in the image based on the internally running neural network. The processing unit generates an image with a detection frame based on the crack location information and outputs it to the small integrated module. The small integrated module stores the image with the detection frame and generates a transparent image with the detection frame based on the crack location information and transmits it to the micro projection module. The processing unit embeds a YOLOv8-based improved neural network in blob format for intelligent crack identification. The YOLOv8-based improved neural network includes a backbone structure and a neck structure. The backbone structure extracts sampled feature maps at 32x, 16x, 8x, and 4x magnification and inputs them into the neck structure. The neck structure processes these sampled feature maps at each magnification and outputs the crack location information and confidence level corresponding to each magnification. In the neck structure of the YOLOv8-based improved neural network, VoVGSCSP replaces C2f in the YOLOv8 neck structure. The improved YOLOv8-based neural network uses EIoU Loss functions include: ， ， ， , ， ； Where A is the actual crack detection frame, B is the predicted crack detection frame, S is the area of ​​the detection frame; w1, h1 are the width and height of detection frame A, w2, h2 are the width and height of detection frame B, d is the Euclidean distance between the center points of the two detection frames, and c is the distance between the diagonals of the smallest rectangular closed region that can simultaneously contain the two frames. The miniature projection module projects images onto the goggles.

2. The smart safety helmet according to claim 1, characterized in that, The camera unit includes: a binocular depth camera and a high-definition RGB camera; The high-definition RGB camera is used to capture high-definition building images and send the images to the processing unit; The binocular depth camera is a black and white camera, located on the left and right sides of the high-definition RGB camera above the brim of the safety helmet. The binocular depth camera sends the captured black and white images to the processing unit. The processing unit is used to identify cracks in high-definition building images, and based on the black and white images, it measures the distance information between the wearer and the crack using a binocular ranging method and outputs it to a small integrated module.

3. The smart safety helmet according to claim 2, characterized in that, The processing unit is also used to calculate the length of cracks in high-definition building images using a pixel filling method, and output the length information to a small integrated module. The over-pixel filling method includes: The high-definition building images captured by the high-definition RGB camera are converted into grayscale images; Binarize the grayscale image; The location of the crack is determined based on the detection frame; Set the pixel values ​​outside the detection box to black; Use Gaussian filtering to filter the image and remove noise; The crack neighborhood is expanded using a morphological dilation algorithm. The pixel length of the crack can be obtained by determining the pixel position occupied by the white pixels. The length of the crack can be calculated based on the ratio mapping between the image and reality.

4. The smart safety helmet according to claim 1, characterized in that, The helmet also includes a sensor integration module, which is located at the top inside the helmet; The sensor integration module includes a three-axis gyroscope, a three-axis accelerometer, a three-axis magnetometer, a temperature sensor, and a barometric pressure sensor. The three-axis gyroscope, three-axis accelerometer, and three-axis magnetometer are used to acquire the wearer's posture information; The sensor integration module sends the measured data to the small integration module.

5. The smart safety helmet according to claim 1 or 4, characterized in that, The small integrated module includes: an ARM processor, a TF memory, and a GPS positioning module; The ARM processor is used to decode the images transmitted by the processing unit, the data acquired by the sensor integration module, and the positioning data acquired by the GPS positioning module, and send them to the TF memory for storage. The ARM processor is also used to overlay the distance information between the wearer and the crack, the crack length information, the positioning data obtained by the GPS positioning module, and the temperature and air pressure data measured by the sensor integration module onto a transparent image with a detection frame, and transmit it to the micro projection module via the HDMI interface.

6. The smart safety helmet according to claim 5, characterized in that, The small integrated module also includes: a wireless communication module; The wireless communication module includes a Bluetooth module and a WIFI module; The Bluetooth module is used to connect the safety helmet and the mobile terminal and transmit data information in the TF memory; The WIFI module is used to upload data from the TF memory to the cloud, send the audio signal of the helmet wearer to the helmets of other wearers and / or the command center, receive audio signals sent by other wearers through their helmets and / or the audio signals sent by the command center, and demodulate the received audio signals and transmit them to the ARM processor.

7. The smart safety helmet according to claim 6, characterized in that, The helmet also includes a chin strap, a microphone, and a speaker. A wire is provided in the chin strap, and the microphone and speaker are mounted on the chin strap and electrically connected to a small integrated module via the wire. The small integrated module also includes: an audio module; The ARM processor unpacks the received audio signal and restores it to the original digital audio signal; The audio module is used to perform digital-to-analog conversion on the digital audio signal, adjust the volume, and output it to the speaker for playback. The audio module is also used to perform analog-to-digital conversion, noise reduction, filtering, and equalization processing on the audio signal received from the microphone from the wearer, and then transmit it to the ARM processor for packaging before passing it to the WIFI module.

8. The smart safety helmet according to claim 7, characterized in that, The helmet also includes an integrated transceiver located on the top of the helmet. It is connected to a WIFI module and a Bluetooth module via wires. The transceiver is used to receive data that the WIFI module needs to send to the cloud, other helmets worn by other users, and the command center, and then transmits the data after amplification. It receives data sent to the mobile terminal by the Bluetooth module, enhances the data, and then transmits it; it also receives data from other wearers' helmets and data sent to the helmet by the command center, enhances the data, and then sends it to the WIFI module.

Images