Wagon balance area cheating identification method based on sensing monitoring and visual analysis coupling
By combining sensing monitoring and visual analysis, AI algorithms and data fusion technology are used to identify cheating behaviors in the construction site floor scale area, solving the problems of false alarms and missed reports in the existing technology, and achieving efficient and accurate cheat identification and prevention.
Patent Information
- Application Number
- CN202410991855.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
AI Technical Summary
The prior art has problems of false alarms and missed reports in the identification of cheating in floor scale areas during construction site construction. Due to the complex environment and insufficient materials, the accuracy of the algorithm is difficult to guarantee, resulting in a lot of waste of manpower and time.
Adopt the floor area cheat recognition method based on the coupling of sensing monitoring and visual analysis. Through the combination of sensing monitoring module and high-definition video monitoring module, AI algorithms are used to conduct real-time analysis, identify and judge potential cheating signs, and improve through data fusion and machine learning. Identification accuracy.
Effectively reduce false positives and missed reports, improve the accuracy and efficiency of identification of cheating behaviors, reduce labor costs, enhance the system's ability to resist fraud, and provide strong evidence support.
Smart Images

Figure CN118942005A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of weighing scales, and in particular to a method for identifying cheating in a weighing scale area based on the coupling of sensor monitoring and visual analysis. Background Art
[0002] At present, in the process of construction projects on construction sites, because the scale is an important equipment for measuring the weight of bulk goods, some lawless elements may use various means to cheat in the actual operation process, interfere with the sensors and instruments of the scale, in order to increase or reduce the weight of the goods, make illegal profits, and cause huge economic losses to the company.
[0003] Therefore, a cheating identification algorithm for the weighbridge area based on the coupling of sensor monitoring and visual analysis can analyze whether there are suspicious persons cheating by capturing images of the area around the weighbridge through sensor alarms and machine vision. It can help staff identify whether there are people cheating on the current weighbridge, and has become an effective means to prevent cheating in advance for material safety of construction sites.
[0004] The existing solutions to the above problems are:
[0005] ①Capture video clips and snapshot images before and after the current moment by analyzing the monitoring data of the weighing sensor of the floor scale and the alarm data of the instrument.
[0006] ② Through the video surveillance of the weighbridge area, the regional intrusion visual analysis algorithm and the personnel recognition visual analysis algorithm are superimposed. When someone lingers in the weighbridge area and behaves suspiciously, an intelligent analysis alarm is generated, and the captured images and video clips are retained.
[0007] The shortcomings of the prior art are:
[0008] ① Only by using the weighing sensor monitoring data and the instrument alarm to capture the video clips and images before and after the current moment, due to the complex environment of the construction site, the sensors and instruments will report many abnormal situations within a reasonable range. Therefore, the large number of captured video clips and images need to be verified by the staff after playing them frame by frame to determine whether there is any cheating behavior in the video. This process consumes a lot of personnel energy and time costs.
[0009] ② Through the video surveillance of the weighbridge area, multiple visual analysis algorithms such as regional intrusion and personnel identification are superimposed. Since the weighbridge is deployed in an outdoor environment, it is difficult to avoid the movement of people, which will cause a large number of false alarms. At the same time, since real cheating often occurs in poor visibility at night, the accuracy of the machine vision analysis algorithm will be greatly reduced, which is easy to cause omissions. In addition, since there is less material for cheating around the weighbridge when training the algorithm, the accuracy of the algorithm will be difficult to guarantee.
[0010] For example, the Chinese invention patent "A method and system for monitoring abnormal weighing data of vehicles without sensing" (application number 202010110717.7) does not jointly analyze and use important data such as video information and data weighing information collected by this patent, thereby reducing the recognition probability and easily causing false alarms. Summary of the invention
[0011] In view of the shortcomings of the prior art, the present invention is based on a method for identifying cheating in the weighbridge area by coupling sensor monitoring and visual analysis. The present invention not only combines the monitoring data of the weighbridge weighing sensor and the instrument alarm, but also combines the visual analysis algorithm to effectively filter the video clips in which no personnel enter the monitoring screen under the real sensor alarm, and efficiently save the staff's verification time. At the same time, the alarm of video analysis can be matched and corroborated with the alarm of sensor monitoring, effectively reducing the situation of false alarms and missed alarms.
[0012] The technical solution of the present invention is: a method for identifying cheating in a weighbridge area based on the coupling of sensor monitoring and visual analysis, comprising a sensor monitoring module, a high-definition video monitoring module, a visual analysis module, and a cheating behavior intelligent judgment module, characterized in that: the sensor monitoring module comprises a weighbridge sensor warning module and a warning information analysis module; the weighbridge sensor warning module collects original data information of pressure sensors and instruments deployed at key positions of the weighbridge and its surroundings, the warning information analysis module analyzes the original warning data information, and extracts data and corresponding time points that may contain cheating behavior; the warning information analysis module uploads the analyzed suspicious data to the data storage module; the high-definition video monitoring module comprises an image data acquisition and supervision alarm upload channel; the image data The data is collected by using cameras to cover the weighbridge area, and the collected image data is transmitted to the visual analysis module in real time through the supervision alarm upload channel; the visual analysis module includes AI wandering alarm monitoring module, AI regional intrusion analysis module, AI face recognition detection module and AI alarm event module. The original collected data is processed through the AI wandering alarm monitoring module, AI regional intrusion analysis module and AI face recognition detection module. The AI alarm event module uses deep learning algorithms to analyze the video stream in real time, identify suspicious behaviors of personnel and intrusion behaviors of personnel in the weighbridge area, control the effective time of the visual analysis algorithm, and discover potential signs of cheating. The AI alarm event module uploads the analyzed suspicious data to the data storage module;
[0013] The cheating behavior intelligent judgment module includes a data storage module and a behavior judgment module. The data storage module collects data output by the visual analysis module and the sensor monitoring module respectively. The behavior judgment module integrates the abnormal data uploaded by the sensor monitoring module with the abnormal data results uploaded by the visual analysis module, and comprehensively judges whether there is cheating behavior through the algorithm model; the behavior judgment module compares the data at the moment of the ground scale sensor warning to see if there is an AI alarm event at the corresponding time point in the AI alarm event module. If the error between the sensor alarm time of the ground scale sensor warning and the AI alarm time is within 10s, it is considered that suspected behavior has occurred at that moment.
[0014] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized by: also including an alarm information uploading module, a weighing scale guard database and a cheating behavior result database, the abnormal data and processing results identified by the cheating behavior intelligent judgment module are uploaded to the alarm information uploading module, the alarm information uploading module performs alarm processing, and the weighing scale guard database and the cheating behavior result database store historical data.
[0015] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized in that: the sensor monitoring module first performs data collection, performs data collection on different instruments, and obtains data from various instruments and meters for data collection through the TF0 protocol and the TF3 protocol; then performs data processing, and finally sets a unified data transmission frequency at the unified Internet of Things gateway to transmit the processed data.
[0016] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized in that: data processing includes data cleaning, format conversion and data storage. Data cleaning will not process data with incorrect weighing value collection format or incomplete transmission data, and data format conversion will be performed between abnormal warning data and normal data; the data after format conversion is transmitted and stored through the Internet of Things gateway.
[0017] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized in that the video monitoring module transmits video data through a unified Internet of Things gateway.
[0018] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized in that: the early warning information analysis module analyzes the monitoring data of sensors and instruments, the early warning information analysis module forms a data fluctuation curve by collecting data every 10ms, analyzes and compares the conventional weighing change curve, and forms a smooth and stable curve by connecting the data at each time point. When the value of a certain point forms a sudden change on the coordinate axis, or the change does not conform to the conventional change, it is judged that abnormal data at the abnormal moment occurs.
[0019] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized in that: the AI alarm event module is based on target detection, behavior recognition algorithm, and regional intrusion algorithm based on deep learning, and is adapted to two data sources, one is to take real-time video stream for real-time intelligent analysis and judgment, and the other is to input the captured and saved video clips into the algorithm for intelligent analysis and judgment. The AI alarm event module centrally displays the recognition results of the three recognition algorithms, and the detection result is judged as the generation of an alarm.
[0020] According to the method for identifying cheating in the weighing scale area based on the coupling of sensor monitoring and visual analysis as described above, it is characterized by: introducing a machine learning mechanism in the intelligent judgment module for cheating behavior, pre-training a deep convolutional neural network on a large data set of video images collected around the weighing scale for three types of AI detection algorithms, and performing sample learning on the rich features of people and actions captured by the pre-trained model; freezing some layers of the training results, and fine-tuning the last few layers or the parts that can be completely self-learned, so as to improve the accuracy of the algorithm.
[0021] The beneficial effect of the present invention is that by combining high-precision sensor monitoring with advanced visual analysis technology, efficient and accurate identification and early warning of various cheating behaviors in the weighing process of the scale can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a system structure diagram of the present invention.
[0023] Figure 2 It is a data collection and transmission flow chart of the present invention. DETAILED DESCRIPTION
[0024] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.
[0025] like Figure 1 As shown, a method for identifying cheating in a weighbridge area based on the coupling of sensor monitoring and visual analysis includes a sensor monitoring module, a high-definition video monitoring module, a visual analysis module, and a cheating behavior intelligent judgment module.
[0026] (1) Sensor monitoring module: including the scale sensor warning module and the warning information analysis module. High-precision pressure sensors and instruments are deployed at key locations of the scale and its surroundings to monitor the weight changes of the scale in real time, and identify abnormal conditions of sensors and instruments to provide basic monitoring data for cheating identification. The scale sensor warning module collects these raw data information, and the warning information analysis module analyzes the warning raw data information to extract data that may contain cheating behavior and the corresponding time points. The warning information analysis module uploads the analyzed suspicious data to the data storage module.
[0027] (2) High-definition video monitoring module: including image data acquisition and supervision alarm upload channel. Image data acquisition uses high-definition cameras to fully cover the weighbridge area, and uses starlight-level video monitoring equipment, combined with night vision and wide dynamic functions to ensure that personnel activities can be clearly recorded under various lighting conditions. The collected image data is then transmitted to the visual analysis module in real time through the supervision alarm upload channel.
[0028] (3) The visual analysis module includes an AI loitering alarm monitoring module, an AI area intrusion analysis module, an AI face recognition detection module, and an AI alarm event module. The AI loitering alarm monitoring module, the AI area intrusion analysis module, and the AI face recognition detection module process the original collected data. The AI alarm event module uses a deep learning algorithm to perform real-time analysis of the video stream to identify suspicious human behavior and human intrusion behavior in the weighing area, control the effective time of the visual analysis algorithm, detect potential signs of cheating, and eliminate a large number of normal daytime activity scenes with false alarms. The AI alarm event module uploads the analyzed suspicious data to the data storage module.
[0029] (4) The cheating behavior intelligent judgment module includes a data storage module and a behavior judgment module. The data storage module collects data output by the visual analysis module and the sensor monitoring module respectively. The behavior judgment module integrates the abnormal data uploaded by the sensor monitoring module with the abnormal data results uploaded by the visual analysis module, and comprehensively judges whether there is cheating behavior through the algorithm model. That is, the behavior judgment module compares the AI alarm event module with the AI alarm event module at the same time through the data of the moment of the scale sensor warning. If the error between the sensor alarm time of the scale sensor warning and the AI alarm time is within 10s, it is considered that suspected cheating behavior has occurred at that moment, and a warning message is issued in time.
[0030] like Figure 1As shown, the present invention may also include an alarm information uploading module, a weighbridge guard database, and a cheating behavior result database, etc. The abnormal data identified by the cheating behavior intelligent judgment module and the processing results are uploaded to the alarm information uploading module, and the alarm information uploading module performs alarm processing. The weighbridge guard database and the cheating behavior result database store historical data, and useful data can be transmitted to the local database as deep learning data, thereby continuously improving the system recognition efficiency and recognition accuracy.
[0031] 2. Key technical points:
[0032] (1) The data acquisition of the sensor monitoring module and the high-definition video monitoring module of the present invention requires multi-source data fusion technology:
[0033] The sensor monitoring module of the present invention adopts an efficient data fusion Internet of Things gateway by installing symmetrical sensors under the floor scale and connecting the sensor values to the weighing instrument, and can process heterogeneous data from different sensors. First, data collection is performed, and the data collection has been connected to dozens of instrument data on the market such as DS3, DS8, DS10, DS12, DS17, D12, D2008, so that different instruments can be processed. Then, the sensor monitoring module obtains various instrument data for data collection through tf0 protocol and tf3 protocol; that is, the sensor monitoring module obtains the sensor monitoring alarm data corresponding to the instrument on the basis of the weighing data of the instrument at the moment, and fuses the data with the measurement time, and can obtain the alarm types such as impedance matching error and sensor matching error of the connected instrument at the same time. Then, data processing is performed, such as data cleaning, format conversion and data storage. Data cleaning will not process the data with incorrect weighing value collection format or incomplete transmission data. If the weighing data undergoes a sudden change within a unit time, or the deviation rate is too large, it is judged as abnormal warning data, and the abnormal warning data and normal data are converted into data format. Since the weighing data results are asynchronous with the data transmission frequency of each sensor monitoring, the format-converted data is transmitted and stored through the IoT gateway, and a unified data transmission frequency is set at the unified IoT gateway. The unified transmission frequency is set as the subsequent alarm value time and video capture value time to ensure the data basis, thereby ensuring that the entire system has a unified time base, and ensuring that the cheating behavior intelligent judgment module can integrate the visual acquisition data and sensor acquisition data.
[0034] The high-definition video surveillance module of the present invention needs to process video stream data, connect to several video surveillance models, connect through different protocols, and transmit video data through a unified Internet of Things gateway. The unified Internet of Things gateway means that multiple video surveillance equipment models can come from one or more equipment manufacturers, connect through the same or different protocols, and output as a data source with the same signal format. This time, one equipment manufacturer, different video surveillance equipment models, and all video surveillance equipment are processed by the same protocol.
[0035] After the basic data collection steps are completed, the monitoring data of sensors and instruments are analyzed in the early warning information analysis module, and one part is connected to the alarm data of the instrument itself. After the other part obtains the monitoring data, a data fluctuation curve is formed by collecting data every 10ms, and the conventional weighing change curve is analyzed and compared. The data at each time point are connected to form a smooth and stable curve. When the value of a certain point forms a sudden change on the coordinate axis, or the change does not conform to the conventional change, it is judged that abnormal data at the abnormal moment occurs.
[0036] (2) The visual analysis module uses intelligent visual analysis algorithms:
[0037] The AI loitering alarm monitoring module is used to set the target detection area. That is, based on the coverage area of the video surveillance field installed on both sides and the top of the weighbridge, an algorithm is used to detect whether there are people in the target detection area and wandering around the weighbridge.
[0038] The input of the AI regional intrusion analysis module is an image or video frame. The target detection area is set (the target detection area is a closed polygonal area). The AI regional intrusion analysis module algorithm automatically calculates whether there is a person in the target detection area. The main judgment basis is whether the boundary range of the person's identification box is within the drawn target detection area. At the same time, the algorithm is applicable to the scene recognition of one person and multiple people. Since the scene of weighbridge cheating is mostly at night and single-person action, in order to avoid false alarms, the algorithm effective time can be customized. The configuration of this algorithm is mainly used in scenes with less traffic and at night.
[0039] The AI face recognition detection module performs face recognition on the persons identified by the AI regional intrusion analysis module.
[0040] The AI alarm event module is based on deep learning target detection, behavior recognition algorithms, and regional intrusion algorithms. It adapts to two data sources: one is to take real-time video streams for real-time intelligent analysis and judgment, and the other is to input captured and saved video clips into the algorithm for intelligent analysis and judgment. The AI alarm event module centrally displays the recognition results of the above three recognition algorithms, and the detection results are judged as alarm generation.
[0041] (3) Adaptive cheating pattern learning: Since the video clips of the alarm moments saved by sensors and instruments in the early stage are manually reviewed one by one, we have accumulated images of cheating scenes that far exceed those that can be learned on the market. The machine learning mechanism is introduced into the intelligent judgment module of cheating behavior. First, we pre-train deep convolutional neural networks (such as ResNet, MobileNet-V2, Inception-V3, etc.) on a large data set of video images collected around the weighing scale for the three types of AI detection algorithms required, and use these pre-trained models to capture the rich features of people and actions for sample learning. The training results are frozen in part of the layers, and the last few layers or the parts that can be completely self-learned are fine-tuned to improve the accuracy of the algorithm. Manual labeling is used in the early stage, and auxiliary training is gradually performed on unlabeled data in the later stage. The nearest neighbor correlation loss (NNCL) is used to optimize the model parameters, or an improved high-speed kernel function is used to accelerate the model learning performance and improve the recognition accuracy.
[0042] Images and video clips that have been manually reviewed and determined to be cheating behaviors can be continuously optimized and trained for algorithms. A self-learning and updated cheating pattern library can be formed based on historical cheating image data, and different behaviors such as the status, location, and action of cheaters can be trained to improve the accuracy of algorithm recognition.
[0043] (4) Immediate warning and countermeasures: Once cheating behavior is identified, the system immediately triggers an alarm and sends a notification to the management personnel through the network. At the same time, it can link the on-site equipment to take countermeasures, such as locking the weighing scale, recording on-site evidence, and using on-site sound column alarms.
[0044] 3. Implementation steps of the present invention:
[0045] (1) Installation and deployment: Arrange sensors on the scale in a reasonable manner, connect all sensor monitoring data to the instrument, and deploy starlight-level surveillance cameras and top panoramic cameras around the scale to ensure full coverage without blind spots.
[0046] (2) System debugging and training: Implement the training of the fusion algorithm, connect the video surveillance clips captured by the sensor monitoring and the alarm generated by the instrument to the visual analysis algorithm, match the video stream captured by the visual analysis alone as the alarm data and the time of the sensor alarm, thereby filtering out some false alarm scenes. Implement parameter tuning and training of the visual analysis model to gradually adapt it to the specific complex and nighttime environment of the construction site.
[0047] (3) Operation and monitoring: By deploying a method for identifying cheating in the weighbridge area based on the coupling of sensor monitoring and visual analysis at the project site, both the weighbridge monitoring and video surveillance are put into real-time monitoring status, data is continuously collected and analyzed, and abnormal events are automatically identified.
[0048] (4) Feedback and optimization: Based on the actual operation situation, feedback information is collected and the optimization algorithm is continuously adjusted to improve system performance.
[0049] This technical solution achieves accurate identification and efficient prevention of cheating behaviors in the weighing scale area through the dual protection of integrated sensor monitoring and visual analysis, providing strong technical support for the security of corporate assets.
[0050] 1. Improve recognition accuracy: By combining the accuracy of sensor monitoring with the intuitiveness of visual analysis, the algorithm can more comprehensively capture the various characteristics of cheating behavior. The sensor monitoring of the scale can accurately measure weight changes, while visual analysis can directly observe intrusions in the area around the scale and abnormal behavior of personnel. The two complement each other, reduce false positives and missed positives, and improve the accuracy of cheating identification.
[0051] 2. Enhanced anti-fraud capabilities: Cheaters may circumvent a single monitoring method through various means (such as installing equipment to change weight sensing, etc.). Combining visual monitoring can effectively identify these physical interventions or manipulation behaviors, making it more difficult for cheating behaviors to escape detection, and enhancing the system's anti-fraud capabilities.
[0052] 3. Adaptability to complex scenarios: Different cargo types, loading methods and environmental conditions may interfere with a single monitoring method. Combining two monitoring methods can provide more stable and reliable monitoring results in different scenarios, such as greater adaptability in situations where light changes greatly and cargo blocks sensors.
[0053] 4. Provide more evidence support: When suspected cheating behavior is identified, it can provide not only abnormal records of weight data, but also intuitive video evidence, which provides strong support for subsequent verification and evidence, helps to quickly resolve disputes and improve law enforcement or management efficiency.
[0054] 5. Promote intelligent management: This comprehensive analysis method is conducive to the introduction of more advanced artificial intelligence algorithms, such as deep learning models, to automatically analyze complex cheating patterns, realize a more intelligent and automated monitoring system, reduce labor costs, and improve management efficiency.
Claims
1. A method for identifying cheating in a weighbridge area based on the coupling of sensor monitoring and visual analysis, comprising a sensor monitoring module, a high-definition video monitoring module, a visual analysis module, and a cheating behavior intelligent judgment module, characterized in that: Sensor monitoring module: including the scale sensor warning module and the warning information analysis module; the scale sensor warning module collects the original data information of the pressure sensors and instruments deployed at the scale and the key positions around it, and the warning information analysis module analyzes the original warning data information to extract the data and corresponding time points that may contain cheating behavior; The early warning information analysis module uploads the analyzed suspicious data to the data storage module; High-definition video monitoring module: including image data acquisition and supervision alarm upload channel; image data acquisition uses cameras to cover the weighbridge area, and the collected image data is transmitted to the visual analysis module in real time through the supervision alarm upload channel; the visual analysis module includes AI wandering alarm monitoring module, AI regional intrusion analysis module, AI face recognition detection module and AI alarm event module. The original collected data is processed through the AI wandering alarm monitoring module, AI regional intrusion analysis module and AI face recognition detection module. The AI alarm event module uses deep learning algorithms to analyze the video stream in real time, identify suspicious behaviors of personnel and intrusion behaviors of personnel in the weighbridge area, control the effective time of the visual analysis algorithm, and discover potential signs of cheating. The AI alarm event module uploads the analyzed suspicious data to the data storage module; the cheating behavior intelligent judgment module includes a data storage module and a behavior judgment module. The data storage module collects data output by the visual analysis module and the sensor monitoring module respectively. The behavior judgment module integrates the abnormal data uploaded by the sensor monitoring module with the abnormal data results uploaded by the visual analysis module, and comprehensively judges whether there is cheating through the algorithm model; The behavior judgment module compares the data at the time of the scale sensor warning to see if there is an AI alarm event at the corresponding time point in the AI alarm event module. If the error between the sensor alarm time of the scale sensor warning and the AI alarm time is within 10s, it is considered that suspected cheating behavior has occurred at that moment.
2. The method for identifying cheating in a weighbridge area based on coupling sensor monitoring and visual analysis according to claim 1 is characterized in that: It also includes an alarm information upload module, a weighbridge guard database and a cheating behavior result database. The abnormal data and processing results identified by the cheating behavior intelligent judgment module are uploaded to the alarm information upload module, and the alarm information upload module performs alarm processing. The weighbridge guard database and the cheating behavior result database store historical data.
3. The method for identifying cheating in a weighbridge area based on coupling sensor monitoring and visual analysis according to claim 1 is characterized in that: The sensor monitoring module first collects data from different instruments and obtains data from various instruments through the TF0 protocol and TF3 protocol; then it processes the data and finally sets a unified data transmission frequency at the unified IoT gateway to transmit the processed data.
4. The method for identifying cheating in a weighbridge area based on coupling sensor monitoring and visual analysis according to claim 3 is characterized in that: Data processing includes data cleaning, format conversion and data storage. Data cleaning will not process data with incorrect weighing value collection format or incomplete transmission data, and data format conversion will be performed between abnormal warning data and normal data. The data after format conversion is transmitted and stored through the Internet of Things gateway.
5. The method for identifying cheating in a weighbridge area based on coupling sensor monitoring and visual analysis according to claim 1 is characterized in that: The video surveillance module transmits video data through a unified IoT gateway.
6. The method for identifying cheating in weighbridge areas based on sensor monitoring and visual analysis coupling according to claim 1 is characterized in that: The early warning information analysis module analyzes the monitoring data of sensors and instruments. The early warning information analysis module forms a data fluctuation curve by collecting data every 10ms, analyzes and compares the conventional weighing change curve, and connects the data at each time point to form a smooth and stable curve. When the value of a certain point forms a sudden change on the coordinate axis, or the change does not conform to the conventional change, it is judged that abnormal data at the abnormal moment occurs.
7. The method for identifying cheating in a weighbridge area based on the coupling of sensor monitoring and visual analysis according to claim 1 is characterized in that: The AI alarm event module is based on deep learning target detection, behavior recognition algorithm, and regional intrusion algorithm. It adapts to two data sources. One is to take real-time video stream for real-time intelligent analysis and judgment, and the other is to input the captured and saved video clips into the algorithm for intelligent analysis and judgment. The AI alarm event module centrally displays the recognition results of the three recognition algorithms, and the detection results are judged as alarm generation.
8. The method for identifying cheating in weighbridge areas based on sensor monitoring and visual analysis coupling according to claim 7 is characterized in that: A machine learning mechanism is introduced into the cheating behavior intelligent judgment module. The deep convolutional neural network is pre-trained on a large data set of video images collected around the weighing scale for three types of AI detection algorithms. The pre-trained model is able to capture the rich features of people and actions for sample learning. The training results are frozen in some layers, and the last few layers or the parts that can be completely self-learned are fine-tuned to improve the accuracy of the algorithm.
Citation Information
Patent Citations
Non-perception wagon balance vehicle weighing data abnormity monitoring method and system
CN113295253A
