Automatic identification and statistics method, system and equipment for personnel and machinery in single-machine construction area and storage medium

Abstract

The invention provides a single-machine construction area personnel and machinery automatic identification and statistics method, system and device, and a storage medium. The method comprises the following steps: S1, video data are captured and stored in a storage unit; s2, data preprocessing; s3, carrying out deep learning model identification on the preprocessed data; s4, storing the identified data to a storage unit, and carrying out real-time statistics and updating; and S5, monitoring an identification result through an intelligent algorithm, and carrying out early warning and response. Image acquisition, processing and analysis are realized through a high-definition camera device, an embedded computing unit and a data storage module, and more comprehensive information is provided for construction management; the system uses a deep learning algorithm to carry out real-time identification and classified statistics on construction site personnel and machinery, and carries out early warning and data management, so that the accuracy and efficiency of target detection and identification are improved; the system has the functions of all-weather monitoring, real-time updating and historical data retrieval, the efficiency and safety of construction management are improved, and the cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of building information technology, and in particular to a method, system, equipment and storage medium for automatic identification and statistics of personnel and machinery in a single-machine construction area. Background Art

[0002] With the acceleration of urbanization, large-scale construction projects are increasing, and the complexity and safety issues of construction site management are becoming increasingly prominent. Traditional construction management relies on manual inspections and records, which is inefficient and error-prone in large-scale, highly dynamic construction environments, and it is difficult to monitor the distribution and status of construction personnel and mechanical equipment in real time. In recent years, the development of technologies such as the Internet of Things, artificial intelligence, and image recognition has provided new possibilities for intelligent management of construction sites. However, most of the existing solutions on the market rely on complex system integration, cloud processing, or expensive hardware facilities, which limits their application in small and medium-sized construction sites or remote areas. In addition, existing technologies often have deficiencies in target recognition accuracy, real-time performance, and user-friendly interface, making it difficult to meet the diverse needs of construction sites. Summary of the invention

[0003] The first object of the present invention is to provide a method for automatically identifying and counting personnel and machinery in a single-machine construction area in response to the above-mentioned problems.

[0004] To achieve the above object, the present invention adopts the following technical solution:

[0005] A method for automatically identifying and counting personnel and machinery in a single-machine construction area comprises the following steps:

[0006] S1: Video data is captured and stored in a storage unit;

[0007] S2: data preprocessing;

[0008] S3: Perform deep learning model recognition on the preprocessed data;

[0009] S4: storing the identified data into a storage unit and performing statistics and updates in real time;

[0010] S5: Monitor identification results through intelligent algorithms and provide early warning and response.

[0011] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0012] As a preferred technical solution of the present invention: the video data capture in step S1 is achieved by a high-definition camera unit.

[0013] As a preferred technical solution of the present invention: the data preprocessing in step S2 is implemented by an embedded processing unit, including denoising and enhancement.

[0014] As a preferred technical solution of the present invention: the step S3 further includes the following sub-steps:

[0015] S31: training data set optimization;

[0016] S32: Data enhancement;

[0017] S33: Model pruning;

[0018] S34: Knowledge distillation;

[0019] S35: Quantization and compilation optimization.

[0020] The second object of the present invention is to provide a system for automatically identifying and counting personnel and machinery in a single-machine construction area.

[0021] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0022] A video data capture and storage module, wherein the video data capture and storage module is used to acquire and save video data;

[0023] A data preprocessing module, which is used to preliminarily process the collected data, improve image quality, make key features more prominent, and facilitate subsequent recognition;

[0024] A deep learning model recognition module, which is used to analyze video data and perform feature extraction and labeling;

[0025] A real-time statistics and update module, which is used to summarize the recognition results in real time and update them to a storage unit inside the system;

[0026] The early warning and response module is used to immediately trigger the early warning mechanism when illegal behavior or potential safety hazards are discovered.

[0027] A third object of the present invention is to provide an electronic device.

[0028] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0029] An electronic device includes a memory and a processor, wherein the memory stores an executable program, and the processor is configured to run the executable program to execute the above-mentioned method for automatically identifying and counting personnel and machinery in a single-machine construction area.

[0030] A fourth object of the present invention is to provide a non-volatile storage medium.

[0031] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0032] A non-volatile storage medium stores an executable program, and when the executable program is executed by a processor, the above-mentioned method for automatically identifying and counting personnel and machinery in a single-machine construction area is implemented.

[0033] The present invention provides a method, system, equipment and storage medium for automatic identification and statistics of personnel and machinery in a single-machine construction area, which has the following beneficial effects: through high-definition camera equipment, embedded computing units and local data storage modules, image data collection, processing and analysis are realized, providing more comprehensive information for construction management; the system uses a deep learning algorithm to perform real-time identification, classification and statistics of personnel and machinery on the construction site, and performs early warning and data management, thereby improving the accuracy and efficiency of target detection and identification; in addition, the system has all-weather monitoring, real-time data updating and historical data retrieval functions, which significantly improves the efficiency and safety of construction management and reduces management costs, and is particularly suitable for various types of building construction and mining scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of the method for automatic identification and statistics of personnel and machinery in a single-machine construction area provided by the present invention.

[0035] Figure 2 It is the overall system architecture diagram of the present invention.

[0036] Figure 3 This is a specific flow chart of step S3. DETAILED DESCRIPTION

[0037] The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

[0038] like Figure 1-2 As shown, a method for automatically identifying and counting personnel and machinery in a single-machine construction area includes the following steps:

[0039] S1: Video data is captured and stored in a storage unit;

[0040] Before deploying the high-definition camera unit, the optimal installation location is determined through construction area simulation and risk assessment to ensure full coverage of key areas (such as entrances, exits, material storage areas, work surfaces, etc.). A 360° panoramic camera or a combination of multiple fixed-angle cameras is used to achieve monitoring without blind spots.

[0041] The camera unit integrates the HDR function, which can automatically adjust the exposure time under extreme lighting conditions (such as strong light at noon and dim light at night) and merge images with different exposure values ​​so that both bright and dark parts can be clearly presented, avoiding overexposure or underexposure and keeping the image rich in details.

[0042] Taking into account the instability of the construction environment, the camera unit has a built-in image stabilization system, such as electronic image stabilization (EIS) or optical image stabilization (OIS), to reduce image blur caused by equipment vibration or wind and improve image quality.

[0043] S2: data preprocessing;

[0044] After the video data is captured, the image is processed using a simple method carried by edge computing devices such as embedded processing units. Apply bilateral filtering algorithms or wavelet denoising techniques to remove random noise in the image while keeping edge information clear. Unsharp Masking (UM) technology is then used to slightly sharpen the image and enhance detail contrast. Apply white balance and color correction. Automatic white balance adjustment ensures that the color is not distorted under different light sources. Through color space conversion (such as RGB to HSV) and color correction algorithms, the image tone is unified to facilitate subsequent color feature extraction. Apply image scaling and cropping. According to the input size requirements of the target detection network, high-quality scaling algorithms such as bicubic interpolation are used to adjust the image size. If necessary, apply region of interest (ROI) cropping technology to focus on areas where people and machinery may appear to reduce invalid calculations.

[0045] Adopt efficient data transmission protocols, such as H.264 / H.265 video encoding technology, to achieve low-latency, high-compression ratio transmission of real-time video streams. Video data is transmitted from the camera unit to the embedded processing unit via a high-speed bus (such as USB3.0, Ethernet over CoaXPress) or wirelessly (such as Wi-Fi6, 5G), ensuring data continuity and integrity.

[0046] After the server receives the data, it enters the preprocessing stage, which mainly involves the following aspects to improve the image quality and facilitate subsequent feature extraction and recognition:

[0047] The two-dimensional Wiener Filter algorithm is used for image denoising. The formula is:

[0048]

[0049] Where I(x,y) is the original image, G(x,y) is the estimated inverse of the noise power spectrum, and σ 2 is the estimation of signal-to-noise ratio, and * indicates the convolution operation.

[0050] This algorithm can effectively suppress salt and pepper noise, stripe noise, etc., and keep the image smooth and natural.

[0051] Use Histogram Equalization (HE) or Contrast Limited Adaptive Histogram Equalization (CLAHE). CLAHE is particularly suitable for images with a wide dynamic range. It avoids over-enhancement by limiting the increase in contrast. The formula is:

[0052]

[0053] Where c is the contrast enhancement factor, T is the brightness offset, p(x,y) is the grayscale probability of pixel position x,y, and p min and p max are the minimum and maximum values ​​of the grayscale histogram, respectively.

[0054] Gamma correction is used to adjust the image brightness. The formula is:

[0055] I'(x,y)=I(x,y) γ

[0056] In the formula, γ is the adjustment factor. A value less than 1 brightens the image, while a value greater than 1 darkens it. Color balance adjustment uses color space transformation, such as adjusting saturation and hue in HSV space.

[0057] S3: Perform deep learning model recognition on the preprocessed data;

[0058] In response to the specific identification needs of the construction area, this paper adopts a customized design of a deep learning model, combines the advantages of convolutional neural networks (CNN) and recurrent neural networks (RNN), and proposes a hybrid model architecture "CNN-RNN" to adapt to the human and mechanical feature extraction and behavior analysis in complex scenarios. The CNN part, such as ResNet-50 or MobileNetV2, is used to extract the spatial features of the image; the RNN part, such as LSTM or GRU, is used to capture temporal information, such as the action path of a person or the operation sequence of a machine. The specific steps are as follows: Figure 3 shown.

[0059] S31: Training data set optimization: Collect and label a large amount of representative construction site images and video data, covering different lighting conditions, weather, personnel clothing and machine models, to ensure the comprehensiveness of model training;

[0060] S32: Data enhancement uses data enhancement techniques such as rotation, flipping, scaling, and color jitter to artificially expand the training set and enhance the generalization ability of the model; in terms of formula, for image rotation operations, the rotation transformation matrix can be used

[0061]

[0062] In the formula, θ is the rotation angle, which can improve the diversity and robustness of image features;

[0063] S33: Model pruning: Use sparse training and weight pruning techniques to remove weights that contribute less to model output and reduce the number of model parameters. The formula is as follows: L = L original +λ||W||1, where L original is the original loss function, λ is the regularization coefficient, and ||W||1 represents the L1 norm of the model parameters, which promotes weight sparsity;

[0064] S34: Knowledge distillation: Distill lightweight models from large pre-trained models to maintain high recognition performance. Knowledge is transferred through soft labels. Formula L distillation =T 2 KL(p T ||q S ), where p T is the probability distribution of the teacher model output, q S is the probability distribution of the student model output, KL is the Kullback-Leibler divergence, and T is the temperature parameter, which enhances the generalization ability of the model while reducing the amount of calculation;

[0065] S35: Quantization and compilation optimization: Model quantization technology converts floating-point operations into low-bit integer operations, such as 8 bits or lower, which greatly reduces memory usage and computing resource requirements. At the same time, a dedicated compiler is used to optimize the code and improve computing efficiency.

[0066] Through the above-mentioned model customization and optimization strategy, as well as the application of high-performance computing technology, the present invention achieves efficient operation on a single device while ensuring recognition accuracy, significantly reduces hardware costs and power consumption, adapts to the complex environment of the construction site, and improves the practicality and feasibility of automatic recognition and statistics of personnel and machinery in the construction area.

[0067] S4: storing the identified data into a storage unit and performing statistics and updates in real time;

[0068] Use lightweight database management systems (such as SQLite or LevelDB) to optimize data storage structures and reduce the overhead of write operations. Design efficient indexing strategies to ensure fast writing of real-time data and efficient retrieval of historical data. Ensure data consistency and integrity through transaction processing mechanisms.

[0069] Online aggregation algorithms (OLA), such as Count-Min Sketch or Lossy Counting, are used to perform real-time statistics on recognition results. These algorithms sacrifice a certain degree of accuracy in exchange for extremely low memory usage and computing costs, and are very suitable for single-machine environments with limited resources. Taking Count-Min Sketch as an example, its basic idea is to use multiple hash functions to map data into a matrix and count by accumulating matrix elements. The formula is as follows:

[0070] C[i,j]=min(C[i,j],C[i′,j′])h k (iitem)=i,g k (item)

[0071] In the formula, C is a two-dimensional matrix, h k and g k is the kth hash function, mapping elements to rows and columns respectively, and item is a count item in the recognition result.

[0072] The system speed is improved through a series of energy-efficiency-optimized update strategies, such as event-driven updates (data updates are triggered only when there are new recognition results or status changes, avoiding unnecessary periodic write operations and reducing energy consumption) and batch writes (collecting recognition results within a certain time window and writing them to the database in batches, reducing the number of I / O operations and improving energy efficiency).

[0073] S5: Monitor identification results through intelligent algorithms and provide early warning and response.

[0074] Deep learning-based abnormal behavior detection models, such as autoencoders or methods based on 3D Convolutional Neural Networks (3D CNNs), are used to perform real-time analysis of human behavior and mechanical operations.

[0075] The autoencoder is trained to learn normal behavior patterns. The input is a normal behavior sequence, and the output attempts to reconstruct the input. For test data, if the reconstruction error exceeds the preset threshold, it is considered abnormal. The formula is where x i is the input behavior sequence, is the reconstructed output and N is the sequence length.

[0076] 3D CNNs spatiotemporal convolutional networks are used to capture the spatiotemporal features in video sequences, and to identify abnormal behavior patterns by learning the spatiotemporal feature distribution of normal behaviors. The model outputs anomaly score S = f(3DConv(x)). If S>T (threshold T), an early warning is triggered, where f is the activation function and x is the spatiotemporal feature representation of the input video sequence.

[0077] Combined with the real-time location information of the machine and personnel, geometric algorithms or machine learning models are used to determine whether the safe operating procedures are violated. For example, if the distance between the machine and the personnel is lower than the safety threshold d safe .

[0078] For each machine-personnel pair, use its coordinate position (x m ,y m ) and (x p ,y p ), calculate the Euclidean distance If d <d safe , it is judged as a potential safety hazard.

[0079] Once the intelligent algorithm identifies a violation or potential safety risk, the early warning signal is immediately triggered through the logic judgment module. The sound and light alarm device deployed on site is immediately activated according to the early warning signal, and emits obvious sound and light warnings to remind on-site personnel to pay attention to safety. At the same time, through the wireless communication module (such as based on the MQTT protocol), the alarm information (including location, type, timestamp, etc.) is encrypted and pushed to the mobile terminal of the security manager or the central monitoring center.

[0080] An automatic identification and statistics system for personnel and machinery in a single-machine construction area. The identification and statistics system includes the following modules:

[0081] Video data capture and storage module, the video data capture and storage module is used to obtain and save video data;

[0082] Data preprocessing module, which is used to preliminarily process the collected data, improve image quality, make key features more prominent, and facilitate subsequent recognition;

[0083] Deep learning model recognition module, which is used to analyze video data and perform feature extraction and labeling;

[0084] Real-time statistics and update module, which is used to summarize the recognition results in real time and update them to the storage unit inside the system;

[0085] The early warning and response module is used to immediately trigger the early warning mechanism when violations or potential safety hazards are discovered.

[0086] The present invention also provides an electronic device, comprising a processor and a memory for storing executable instructions of the processor, wherein the processor is configured to execute the executable instructions to implement the above-mentioned method for automatically identifying and counting personnel and machinery in a single-machine construction area.

[0087] The present invention also provides a non-volatile storage medium, in which an executable program is stored. When the executable program is executed by a processor, the method for automatically identifying and counting personnel and machinery in a single-machine construction area as described above is implemented.

[0088] Specifically, the above-mentioned method for automatic identification and statistics of personnel and machinery in the single-machine construction area is implemented in the following way:

[0089] After the system is started, the high-definition camera unit continuously captures the video stream, and the data is processed by the AI ​​algorithm in the embedded processing unit. First, the algorithm denoises and enhances the image, and then uses the deep learning model to identify the people and machines in the image, including but not limited to the facial features of the people, the color of the helmet, the type of machine, etc.; the recognition results are counted in real time and updated to the storage unit; when abnormal behavior or safety hazards are detected, the system immediately triggers a warning signal, through sound and light alarm or sends an alarm message through the wireless communication module.

[0090] The above-mentioned specific implementation methods are used to explain the present invention and are only preferred embodiments of the present invention, rather than limiting the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for automatically identifying and counting personnel and machinery in a single-machine construction area, characterized in that: The steps include: S1: Video data is captured and stored in a storage unit; S2: data preprocessing; S3: Perform deep learning model recognition on the preprocessed data; S4: storing the identified data into a storage unit and performing statistics and updates in real time; S5: Monitor identification results through intelligent algorithms and provide early warning and response.

2. The method for automatic identification and statistics of personnel and machinery in a single-machine construction area according to claim 1 is characterized by: The video data capture in step S1 is achieved by a high-definition camera unit.

3. The method for automatic identification and statistics of personnel and machinery in a single-machine construction area according to claim 1 is characterized by: The data preprocessing in step S2 is implemented by an embedded processing unit, including denoising and enhancement.

4. The method for automatic identification and statistics of personnel and machinery in a single-machine construction area according to claim 1 is characterized by: The step S3 further comprises the following sub-steps: S31: training data set optimization; S32: Data enhancement; S33: Model pruning; S34: Knowledge distillation; S35: Quantization and compilation optimization.

5. A system for automatically identifying and counting personnel and machinery in a single-machine construction area, characterized by: The identification and statistics system includes the following modules: A video data capture and storage module, wherein the video data capture and storage module is used to acquire and save video data; A data preprocessing module, which is used to preliminarily process the collected data, improve image quality, make key features more prominent, and facilitate subsequent recognition; A deep learning model recognition module, which is used to analyze video data and perform feature extraction and labeling; A real-time statistics and update module, which is used to summarize the recognition results in real time and update them to a storage unit inside the system; The early warning and response module is used to immediately trigger the early warning mechanism when illegal behavior or potential safety hazards are discovered.

6. An electronic device, comprising a memory and a processor, characterized in that: The memory stores an executable program, and the processor is configured to run the executable program to execute the steps of the method for automatic identification and statistics of personnel and machinery in a single-machine construction area according to any one of claims 1-4.

7. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores an executable program, and when the executable program is executed by the processor, the steps of the method for automatic identification and statistics of personnel and machinery in a single-machine construction area are implemented as described in any one of claims 1-4.

Images