Safety helmet wearing behavior detection system

Abstract

The invention discloses a safety helmet wearing behavior detection system, and belongs to the technical field of safety. Comprising an image acquisition module, a processing module and an alarm module, and the image acquisition module, the processing module and the alarm module are in circuit connection in sequence. The processing module comprises the following modules: a convolutional layer operation information extraction module; an area proposal network operation information extraction module; an ROI Pooling operation information extraction module; a classification module and a regression module; and a post-processing and decision-making module. Real-time analysis and recognition of production and manufacturing in the coal mine industry can be achieved, the safety helmet wearing detection system can record personnel who do not wear safety helmets and give an alarm immediately, automatic supervision can be achieved, and meanwhile the detection accuracy is improved through multi-level feature extraction and accurate positioning.

Description

Technical Field

[0001] The invention provides a helmet wearing behavior detection system, belonging to the technical field of safety. Background Art

[0002] Safety management of construction sites is crucial in high-risk industries such as construction, mining, and chemicals. Every year, there are many unreported injuries and deaths caused by construction accidents around the world, and a large proportion of these accidents are related to workers not wearing hard hats. Hard hats are one of the most basic personal protective equipment on construction sites, which can effectively protect workers' heads from falling objects, collisions, etc. Therefore, ensuring that all workers wear hard hats correctly when entering the construction site is an important measure to prevent accidents.

[0003] However, the current safety system is highly dependent on manual labor, and due to the large working area, large number of personnel, and long working hours, the efficiency of manual supervision is not satisfactory. Therefore, there is an urgent need for an automated system to assist in supervision; the existing automated detection system only uses the computer's visual system for judgment, which has low accuracy. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a technical solution: a helmet wearing behavior detection system, comprising an image acquisition module, a processing module, and an alarm module, and the image acquisition module, the processing module, and the alarm module are circuit-connected in sequence; the processing module comprises the following modules: a convolutional layer operation information extraction module; a region proposal network operation information extraction module; a ROIPooling operation information extraction module; a classification module and a regression module; a post-processing and decision module.

[0005] As an improvement, the image acquisition module is connected to a computer vision acquisition system, and the image acquisition module includes an input image preprocessing module, specifically including image scaling and normalization operations.

[0006] As an improvement, in the convolution layer operation information extraction module, the features of the image are extracted through multiple convolution layers, and the operation information of each convolution layer is recorded and analyzed to optimize the feature extraction process; the operation information of the convolution layer includes activation value and weight; the operation steps are as follows: input the preprocessed image to the convolution layer; perform convolution operation to extract feature map; record the activation value and weight information of the convolution layer; adjust the network parameters according to the activation value to optimize feature extraction;

[0007] As an improvement, in the region proposal network operation information extraction module, the region proposal network (RPN) generates candidate regions, i.e., regions containing people; the operation information of the RPN includes the location and confidence of the proposal box; the operation steps are as follows: based on the feature map of the convolutional layer, the RPN generates multiple region proposals; for each proposal, the confidence of whether it contains a person is calculated; the location of the proposal box and the corresponding confidence score are recorded; the proposal box with high confidence is selected to enter the next step of processing;

[0008] As an improvement, in the ROIPooling operation information extraction module, a pooling operation is performed on the candidate regions generated in the region proposal network operation information extraction module to convert regions of different sizes into feature maps of fixed sizes; the operation information of ROIPooling, that is, the feature vector after pooling, is recorded; the operation steps are as follows: the proposal box selected by the RPN is cropped and scaled to generate a feature map of fixed size; the feature vector after ROIPooling is recorded; and the feature vector is passed to the classification and regression modules.

[0009] As an improvement, in the classification module and regression module, the specific operations are as follows: the classification module determines whether there is a person in the area and whether he is wearing a helmet; the regression module adjusts the position and size of the bounding box; records the classification results and the corresponding confidence, as well as the coordinates of the regressed bounding box; adjusts the parameters of the classification and regression models according to historical information;

[0010] As an improvement, for the post-processing and decision-making module, the results of classification and regression are combined, and redundant detection frames are removed through methods such as non-maximum suppression to finally determine whether the person is wearing a helmet.

[0011] As an improvement, the alarm module is connected to the alarm system circuit of the external terminal.

[0012] Beneficial effects of the present invention:

[0013] 1. The technology of the coal mine safety helmet wearing detection system can realize real-time analysis and identification of the production and manufacturing of the coal mining industry. The coal mine safety helmet wearing detection system can record the personnel who are not wearing safety helmets and immediately issue an alarm, which can realize automatic supervision. At the same time, it can improve the accuracy of detection through multi-level feature extraction and precise positioning.

[0014] 2. The optimized convolutional layer and RPN design ensure that the system can process video streams in real time.

[0015] 3. The system can be easily expanded to accommodate more types of personal protective equipment testing.

[0016] 4. Can be integrated with the alarm system to achieve real-time warning and improve safety management efficiency BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of a helmet wearing behavior detection system of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described below in conjunction with the accompanying drawings.

[0019] according to Figure 1 As shown: The present invention provides a helmet wearing behavior detection system: including an image acquisition module, a processing module, and an alarm module, and the image acquisition module, the processing module, and the alarm module are circuit-connected in sequence; the processing module includes the following modules: a convolutional layer operation information extraction module; a region proposal network operation information extraction module; a ROIPooling operation information extraction module; a classification module and a regression module; a post-processing and decision module.

[0020] The image acquisition module is connected to the computer vision acquisition system. The image acquisition module includes an input image preprocessing module, which specifically includes image scaling and normalization operations. Different images may have different resolutions when they are acquired. Scaling all input images to the same size ensures that the input layer of the model can process all images consistently, avoiding the waste of computing resources due to different sizes.

[0021] In the convolutional layer operation information extraction module, the features of the image are extracted through multiple convolutional layers. The operation information of each convolutional layer is recorded and analyzed to optimize the feature extraction process; the operation information of the convolutional layer includes activation value and weight; the operation steps are as follows:

[0022] Input the preprocessed image to the convolution layer. The preprocessed image has a uniform standard size and value range, ensuring that the convolution layer can process all inputs consistently and avoid computational problems caused by input differences. This improves computational efficiency, reduces the complexity and amount of computation in the convolution operation, and accelerates the feature extraction process.

[0023] Perform convolution operations to extract feature maps; the convolution operation slides over the image through a filter (convolution kernel) to extract local features (such as edges, textures, etc.). The convolution layer reduces the number of model parameters and improves computational efficiency through a parameter sharing mechanism.

[0024] Recording the activation values ​​and weight information of the convolutional layer helps to understand the decision-making process of the model, improve the interpretability of the model, and facilitate debugging and optimization. Real-time monitoring of the activation values ​​and weight changes of the convolutional layer can help adjust the learning rate and optimizer parameters to ensure stable model training.

[0025] The network parameters are adjusted according to the activation value to optimize feature extraction; adaptive adjustment of network parameters can optimize the feature extraction process and improve the recognition accuracy and efficiency of the model.

[0026] In the region proposal network operation information extraction module, the region proposal network (RPN) generates candidate regions, that is, regions containing people; the operation information of RPN includes the location and confidence of the proposal box; the operation steps are as follows:

[0027] Based on the feature map of the convolutional layer, RPN generates multiple region proposals; compared with the traditional sliding window method, RPN reduces the amount of computation for generating candidate regions by sharing the calculation of the convolutional layer.

[0028] For each proposal, calculate the confidence of whether it contains a person; by calculating the confidence of each proposal box, it can be determined whether the area contains the target object.

[0029] Recording the location of the proposal box and the corresponding confidence score can be used for subsequent non-maximum suppression (NMS) and visualization of target detection results; it helps to accurately locate the target object and improve the accuracy of detection.

[0030] Select the proposal box with high confidence to proceed to the next step.

[0031] In the ROIPooling operation information extraction module, the candidate regions generated in the region proposal network operation information extraction module are pooled to convert regions of different sizes into feature maps of fixed sizes; the ROIPooling operation information, i.e., the pooled feature vector, is recorded; the operation steps are as follows:

[0032] Crop and scale the proposal box selected by RPN to generate a fixed-size feature map to ensure consistency in subsequent processing;

[0033] Record the feature vector after ROIPooling, which can be used for subsequent analysis and debugging to ensure the integrity of feature information;

[0034] Pass the feature vector to the classification and regression modules to ensure that these modules can receive key feature information and perform accurate classification and regression tasks.

[0035] In the classification module and regression module, the specific operations are:

[0036] The classification module determines whether there is someone in the area and whether he is wearing a helmet, thus achieving dual detection of the target; and the judgment of the classification module provides reliable target information for the subsequent regression module, thus improving the accuracy of the overall detection.

[0037] The regression module adjusts the position and size of the bounding box to accurately locate the target object and improve the detection accuracy.

[0038] Recording the classification results and the corresponding confidence levels, as well as the regressed bounding box coordinates, can be used for subsequent analysis and debugging to ensure the integrity of the information.

[0039] Adjusting the parameters of the classification and regression models based on historical information can make the models more adaptable to actual application scenarios and improve detection accuracy.

[0040] For the post-processing and decision-making module, the results of classification and regression are combined, and redundant detection frames are removed through methods such as non-maximum suppression, which effectively reduces the phenomenon of false positives and false negatives, improves the accuracy of detection, and ultimately determines whether the personnel are wearing safety helmets. The alarm module is connected to the alarm system circuit of the external terminal, and ultimately the staff who are not wearing safety helmets are displayed and warned through the alarm system.

[0041] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A helmet wearing behavior detection system, characterized in that: It includes an image acquisition module, a processing module, and an alarm module, and the image acquisition module, the processing module, and the alarm module are connected in sequence; the processing module includes the following modules: a convolutional layer operation information extraction module; a region proposal network operation information extraction module; a ROIPooling operation information extraction module; Classification module and regression module; Post-processing and decision-making module.

2. A helmet wearing behavior detection system according to claim 1, characterized in that: The image acquisition module is connected to the computer vision acquisition system, and the image acquisition module includes an input image preprocessing module, which specifically includes image scaling and normalization operations.

3. A helmet wearing behavior detection system according to claim 1, characterized in that: In the convolutional layer operation information extraction module, the features of the image are extracted through multiple convolutional layers, and the operation information of each convolutional layer is recorded and analyzed to optimize the feature extraction process; the operation information of the convolutional layer includes activation value and weight; the operation steps are as follows: input the preprocessed image to the convolutional layer; perform convolution operation to extract feature map; record the activation value and weight information of the convolutional layer; adjust the network parameters according to the activation value to optimize feature extraction.

4. A helmet wearing behavior detection system according to claim 1, characterized in that: In the region proposal network operation information extraction module, the region proposal network (RPN) generates candidate regions, that is, regions containing people; the operation information of RPN includes the location and confidence of the proposal box; the operation steps are as follows: based on the feature map of the convolution layer, RPN generates multiple region proposals; for each proposal, calculate the confidence of whether it contains a person; record the location of the proposal box and the corresponding confidence score; Select the proposal box with high confidence to proceed to the next step.

5. A helmet wearing behavior detection system according to claim 1, characterized in that: In the ROIPooling operation information extraction module, the candidate regions generated in the region proposal network operation information extraction module are pooled to convert regions of different sizes into feature maps of fixed sizes; the ROIPooling operation information, i.e., the feature vector after pooling, is recorded; the operation steps are as follows: the proposal box selected by the RPN is cropped and scaled to generate a feature map of fixed size; the feature vector after ROIPooling is recorded; and the feature vector is passed to the classification and regression modules.

6. A helmet wearing behavior detection system according to claim 1, characterized in that: In the classification module and regression module, the specific operations are as follows: the classification module determines whether there is a person in the area and whether he is wearing a helmet; the regression module adjusts the position and size of the bounding box; records the classification results and the corresponding confidence levels, as well as the coordinates of the regressed bounding box; and adjusts the parameters of the classification and regression models based on historical information.

7. A helmet wearing behavior detection system according to claim 1, characterized in that: For the post-processing and decision-making module, the results of classification and regression are combined, and redundant detection frames are removed through methods such as non-maximum suppression to finally determine whether the person is wearing a helmet.

8. A helmet wearing behavior detection system according to claim 1, characterized in that: The alarm module is connected to the alarm system circuit of the external terminal.

Images