System for intelligently checking and weighing water stored in vehicle water tank

Abstract

The invention relates to the technical field of intelligent metering, in particular to a system for intelligently checking and weighing water stored in a vehicle water tank, and the system comprises a multi-mode data collection module which is used for collecting video data and weighing data of a vehicle water tank area; the deep learning action recognition module is used for recognizing the action of opening the faucet by the driver; the target detection module is used for analyzing the water flow state through an optical flow method; the intelligent control logic module is used for judging whether weighing is started or not according to action recognition and the water flow state; the man-machine interaction module is used for guiding a driver to operate; automatic detection is achieved through the artificial intelligence video analysis technology, so that a traditional visual inspection mode of workers is completely replaced, the omission ratio is reduced, the problem of weighing errors caused by the fact that the water tank is not emptied is effectively solved, meanwhile, full-process automatic management is achieved through the intelligent control logic module, and the working efficiency is improved. The whole detection process does not need manual intervention, and the labor cost and the management cost of an enterprise are remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent metering technology, and in particular to a system for intelligently checking and weighing the water content in a vehicle water tank. Background Art

[0002] Traditionally, vehicle water tank water level verification relies primarily on visual inspection. This requires on-site observation of the tank valve status and empirical judgment of water drainage. This approach has significant limitations, particularly in vehicle-intensive environments like industrial and mining enterprises, where efficiency and quality are difficult to guarantee.

[0003] Existing technologies have two major shortcomings: First, the missed inspection rate during staff inspections is high, with statistics showing a rate of approximately 15%-20%. Due to factors such as fatigue, obstructed vision, and ambient light, inspectors may not accurately determine valve status, resulting in weighing errors before the tank is empty. Second, this method requires dedicated inspectors, which significantly increases labor costs in high-traffic scenarios. Furthermore, manual inspections are difficult to maintain 24 / 7, further reducing inspection quality at night or in inclement weather.

[0004] Therefore, in response to the above-mentioned problems of high missed detection rate and high labor cost, the present invention proposes a system for intelligently checking and weighing the water level in vehicle water tanks. By automatically identifying faucet operation actions and water flow status, unmanned detection of water level in vehicle water tanks can be achieved, effectively solving the accuracy and efficiency problems existing in manual inspections. Summary of the Invention

[0005] In order to overcome the problems of high missed detection rate and high labor cost in traditional vehicle water tank water level verification, the present invention proposes a system for intelligently checking and weighing the water level in vehicle water tanks.

[0006] The technical solution of the present invention is: a system for intelligently checking and weighing the water content in a vehicle water tank, comprising: Multimodal data acquisition module, used to collect video data and weighing data of the vehicle water tank area; A deep learning action recognition module is used to identify the driver's action of turning on the faucet; The target detection module is used to analyze the water flow state through the optical flow method; Intelligent control logic module, used to determine whether to start weighing based on motion recognition and water flow status; The human-computer interaction module is used to guide the driver's operations and provide feedback on test results.

[0007] Preferably, the multimodal data acquisition module includes an industrial-grade anti-shake camera and a weighing sensor, the camera has a low-light compensation function, and the timestamp synchronization calibration error between the weighing sensor and the video acquisition device is controlled within 50 milliseconds.

[0008] Preferably, the deep learning action recognition module is constructed based on an improved YOLOv7 spatiotemporal attention network model, which can accurately capture the characteristics of human hand movements through training and learning, especially recognize the key frame sequence of faucet rotation movements, and the model has been lightweight.

[0009] Preferably, the target detection module uses optical flow analysis technology to detect the water flow state by calculating the motion vectors of pixels in continuous video frames. It can distinguish between still water bodies and flowing water bodies, and comprehensively judge whether the water tank has completed drainage in combination with the output results of the deep learning model.

[0010] Preferably, the intelligent control logic module uses a dual-threshold judgment mechanism. When the system detects that the driver's action of turning on the faucet is more than 85% complete and the water flow duration is less than 5 seconds, the weighing process is automatically triggered. If it is detected that the driver fails to perform the specified action within the specified time, the alarm program is activated.

[0011] Preferably, the human-computer interaction module includes a high-brightness LED display and a voice prompt device. The display screen displays operation guidance information in real time to guide the driver to complete the water tank draining operation, and at the same time informs the operation status through visual feedback. When an abnormal situation is detected, the driver and management personnel are reminded through sound and light alarms.

[0012] Preferably, the system also includes a vehicle identification subsystem, which triggers a high-definition snapshot camera or RFID reader through a ground sensor coil to obtain vehicle identity information and automatically retrieve pre-stored vehicle file data, which includes standard load capacity and water tank position parameters.

[0013] Preferably, the training data of the deep learning action recognition module is constructed by manually annotating the key action sequences of the faucet operation, the training samples are expanded using data enhancement techniques such as random rotation, brightness adjustment and noise addition, and the model training process is optimized using an improved loss function.

[0014] Preferably, the system uses multi-camera video stream splicing technology to eliminate detection blind spots, and splices monitoring images from multiple angles into a complete view through feature point matching and image fusion algorithms, while combining deep learning models to achieve full-view real-time action recognition and water flow status analysis.

[0015] Preferably, the system is provided with a remote management interface. When the system detects that the driver fails to perform operations as required or an abnormal situation occurs, it automatically generates an alarm message and transmits it to the management platform via the network. At the same time, it saves relevant video evidence to support management personnel to remotely view the on-site situation and handle it.

[0016] Beneficial effects of the present invention: 1. This invention achieves automated detection through artificial intelligence video analysis technology, thus completely replacing the traditional visual inspection method of staff, reducing the missed detection rate from the original 15%-20% to below 5%, significantly improving detection accuracy and reliability, and effectively avoiding the weighing error problem caused by the water tank not being empty. At the same time, the intelligent control logic module is adopted to realize the full process automation management. When the compliance operation is detected, the weighing process is automatically triggered, and an alarm is immediately issued in the event of an abnormality. The entire detection process does not require human intervention, which significantly reduces the company's labor and management costs.

[0017] 2. This invention uses multimodal data fusion technology, combined with industrial-grade anti-shake camera and weighing sensor data, to achieve millisecond-level time synchronization, ensuring data consistency and timeliness during the detection process, and providing accurate data support for subsequent analysis.

[0018] 3. Based on an improved YOLOv7 spatiotemporal attention network model, the system can accurately identify the driver's action characteristics when turning on the faucet and analyze the water flow status in real time through optical flow. This dual verification mechanism significantly improves the confidence of the detection results and avoids the misjudgment that may be caused by a single detection method. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 What is shown is a schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0021] The present invention provides an embodiment: a system for intelligently checking and weighing the water level in a vehicle water tank, comprising: Multimodal data acquisition module, used to collect video data and weighing data of the vehicle water tank area; A deep learning action recognition module is used to identify the driver's action of turning on the faucet; The target detection module is used to analyze the water flow state through the optical flow method; Intelligent control logic module, used to determine whether to start weighing based on motion recognition and water flow status; The human-computer interaction module is used to guide the driver's operations and provide feedback on test results.

[0022] The multimodal data acquisition module includes an industrial-grade anti-shake camera and a weighing sensor. The camera has a low-light compensation function, and the timestamp synchronization calibration error between the weighing sensor and the video acquisition device is controlled within 50 milliseconds.

[0023] Furthermore, the industrial-grade anti-shake camera and weighing sensor of the multimodal data acquisition module work together to ensure the accuracy and synchronization of the detection data, providing a reliable input source for the system. At the same time, the deep learning action recognition module can accurately capture the driver's operating actions. Combined with the optical flow analysis technology of the target detection module, it realizes the double verification of the water tank drainage status, thereby greatly improving the detection accuracy. The intelligent control logic module automatically triggers the weighing process or alarm mechanism according to the preset conditions, realizing unmanned operation of the entire process. Finally, the human-computer interaction module simplifies the driver's operating process through intuitive guidance and feedback, thereby improving the system's usability and execution efficiency. Through the coordinated cooperation of various modules, the system realizes the intelligent, automated and high-precision detection of water storage in vehicle tanks, effectively solving the problems of high missed detection rate and high labor cost in traditional manual detection.

[0024] Furthermore, the installation position and number of cameras are described, specifically: The camera is installed on a column or gantry on one side of the weighing area to ensure that its lens can clearly capture the water filling valve of the truck water tank and the driver's operating area with a radius of at least 1.5 meters around it. The installation height is preferably 2.5 meters to 4 meters, and a slightly downward angle is adopted (for example, an angle of 10-30 degrees with the horizontal plane) to reduce interference from ground reflections and maximize coverage of the entire operation, while avoiding occlusion of key areas due to differences in driver height or vehicle models. At the same time, the system deploys at least one high-definition industrial camera with infrared night vision and high dynamic range (HDR) functions to ensure clear imaging all day long; to enhance the robustness of the system, an auxiliary camera can be added at a symmetrical position or at another position that can form a certain viewing angle difference (for example, 30-60 degrees) to solve the occlusion problem that may occur in a single perspective and provide richer visual information for redundant judgment.

[0025] The deep learning action recognition module is built based on the improved YOLOv7 spatiotemporal attention network model. Through training and learning, the model can accurately capture the characteristics of human hand movements, especially the key frame sequence of faucet rotation movements, and the model has been lightweight.

[0026] Furthermore, the improvements of YOLOv7 are described, specifically: Specific structural adjustments of the spatiotemporal attention network: This paper innovatively integrates a lightweight temporal feature extraction and attention fusion module (STAFM) in parallel with the backbone network (ELAN module) of YOLOv7. The STAFM module mainly includes: Temporal convolution branch: Uses a 3D convolution kernel (e.g., 3x3x3) to process the feature maps of T consecutive frames (T is usually 3-7 frames) to directly capture the motion patterns of the hand and faucet in the temporal dimension.

[0027] Spatial attention enhancement branch: For the feature map of each frame, an improved channel attention and spatial attention concatenation mechanism is used (for example, first generating channel weights through global average pooling and fully connected layers, and then generating spatial attention maps through 1x1 convolution) to enhance the salient area features related to the tap operation.

[0028] Cross-temporal and spatial feature fusion: The dynamic features extracted by the temporal convolution branch are weightedly fused with the static features processed by the spatial attention enhancement branch. The weights are dynamically learned by a small self-attention network, which determines the contribution of dynamic or static features to the current action recognition based on the contextual information of the current frame sequence.

[0029] Through this STAFM module, the model can more effectively learn the temporal interaction between the hand and the faucet, such as the start, duration, and end of rotation. Compared with the original YOLOv7, tested on a specially constructed faucet operation video dataset, this improved model improved the mAP (mean Average Precision) indicator for faucet opening / closing action recognition by at least 8-15%. In particular, the recognition accuracy and robustness were significantly enhanced under rapid operation, slight occlusion, and complex backgrounds.

[0030] The fusion logic of optical flow method and action recognition: This method integrates optical flow pre-screening and keypoint precision positioning. First, the system uses an efficient dense optical flow algorithm (based on optical flow estimation using the RAFT model) to analyze the pixel-level motion of the faucet and candidate hand regions in the video stream. If continuous optical flow (lasting longer than N_flow frames, for example, 6-8 frames) is detected within the designated region, with an amplitude exceeding a preset threshold V_flow (for example, pixel displacement greater than 5 pixels / frame) and a motion direction consistent with a typical rotation pattern (for example, a near-circular arc trajectory in the hand region or a rotating vector field in the faucet body), it is identified as a potential valid action. At this point, the system activates the improved YOLOv7 model to accurately detect and locate human keypoints (especially hands) and faucet components (such as the valve handle and water outlet) in the image frames within this time window. If the optical flow analysis detects no significant motion or the motion pattern does not match, the subsequent complex model calculation is skipped, effectively filtering out static, irrelevant shaking, or non-operational movements. This significantly reduces computing resource consumption and improves system response speed and overall operational efficiency.

[0031] The target detection module uses optical flow analysis technology to detect the water flow state by calculating the motion vectors of pixel points in continuous video frames. It can distinguish between still water bodies and flowing water bodies, and combine the output results of the deep learning model to comprehensively judge whether the water tank has completed drainage. This technology effectively avoids the defect of traditional visual detection being affected by changes in lighting.

[0032] The intelligent control logic module adopts a dual-threshold judgment mechanism. When the system detects that the driver's action of turning on the faucet is more than 85% complete and the water flow duration is less than 5 seconds, it automatically triggers the weighing process. If it is detected that the driver does not perform the specified action within the specified time, the alarm program is activated.

[0033] The human-computer interaction module includes a high-brightness LED display and a voice prompt device. The display shows operation guidance information in real time to guide the driver to complete the water tank draining operation, and at the same time informs the operation status through visual feedback. When an abnormal situation is detected, the driver and management personnel are reminded through sound and light alarms.

[0034] The system also includes a vehicle identification subsystem, which triggers a high-definition snapshot camera or RFID reader through a ground sensor coil to obtain vehicle identity information and automatically retrieve pre-stored vehicle file data, which includes standard load capacity and water tank position parameters.

[0035] The training data for the deep learning action recognition module is constructed by manually annotating key action sequences of faucet operations. The training samples are expanded using data augmentation techniques such as random rotation, brightness adjustment, and noise addition. The model training process is optimized using an improved loss function to ensure that the model can adapt to action recognition needs under different environmental conditions.

[0036] The system uses multi-camera video stream stitching technology to eliminate detection blind spots, and stitches monitoring images from multiple angles into a complete view through feature point matching and image fusion algorithms. At the same time, it combines deep learning models to achieve full-view real-time action recognition and water flow status analysis, thereby significantly improving the coverage and accuracy of the detection system.

[0037] The system is equipped with a remote management interface. When the system detects that the driver fails to perform operations as required or an abnormal situation occurs, it automatically generates an alarm message and transmits it to the management platform through the network. At the same time, it saves relevant video evidence, supports management personnel to remotely view the on-site situation and handle it, and realizes closed-loop management of the detection process.

[0038] See also Figure 1 , further, the workflow of the present invention is described, specifically: When a vehicle enters the weighing area, the ground sensor coil automatically triggers the license plate recognition system, which obtains vehicle information through a snapshot camera or RFID technology. The system then retrieves pre-stored vehicle file data (including standard load capacity and water tank location parameters) and simultaneously activates the multi-angle, industrial-grade anti-shake camera installed on the side of the vehicle to begin collecting video data of the water tank area.

[0039] After the vehicle stops, the LED display screen shows clear operating instructions, prompting the driver to get out of the vehicle and open the water tank faucet. At the same time, the system provides auxiliary guidance through voice prompts to ensure that the driver clearly understands the operating requirements and create standardized operating conditions for subsequent AI detection.

[0040] The deep learning action recognition module analyzes the video stream captured by the camera in real time, identifies the driver's hand movements based on the improved YOLOv7 spatiotemporal attention network model, and determines whether the standard action of turning on the faucet is accurately performed. This process uses key frame sequence analysis technology to ensure the accuracy of action recognition.

[0041] The target detection module uses the optical flow method to calculate the motion vectors of pixels in continuous video frames, accurately analyzes the water flow state at the water tank drain outlet, and thus distinguishes between still and flowing water bodies. It also cross-validates the results with the motion recognition to ensure the reliability of the detection results.

[0042] The intelligent control logic module integrates the results of action recognition and water flow analysis. When it detects that the action completion rate exceeds 85% and the water flow duration is less than 5 seconds, it is determined to be a compliant operation and automatically prepares to start the weighing process. If no compliant operation is detected within the time limit, the alarm mechanism is triggered.

[0043] After confirming that the water tank is drained, the system guides the driver to leave the weighing area through the LED display screen, and automatically calibrates the weighing sensor to prepare for accurate weighing, ensuring that the weighing process is not interfered with by personnel.

[0044] The system automatically triggers the scale to start weighing. The weighing sensor collects the vehicle weight data and compares it with the vehicle tare weight data to calculate the net weight. The entire process is fully automated to avoid errors caused by human intervention.

[0045] When the system detects that the driver has not performed the prescribed action or the operation does not meet the requirements, it will immediately trigger an audible and visual alarm, and transmit the abnormal information to the management platform through the network, notifying relevant personnel to intervene and handle the matter, and at the same time save relevant video evidence for reference.

[0046] The system automatically stores complete inspection process data (including video records, motion recognition results, water flow analysis data and weighing results) in the database, generates an inspection report, and can output it to the enterprise management system through an interface to achieve data traceability.

[0047] After completing all inspection and weighing processes, the system displays the inspection results on the LED display and prompts the driver to leave. At the same time, it resets all inspection modules and prepares to receive the inspection task of the next vehicle, realizing continuous automated operation.

[0048] Furthermore, the present invention provides an embodiment, in a standard working scenario: This system is deployed at the scale points where logistics vehicles enter industrial and mining enterprises. When a loaded truck enters the weighing area, the system automatically identifies the vehicle information and activates the multi-angle camera. The driver completes the water tank draining operation according to the instructions on the LED screen. The AI ​​system completes the action recognition and water flow detection within 10 seconds, automatically triggers the weighing process after confirming compliance, and uploads the detection data to the enterprise ERP system in real time, realizing unmanned operation of the entire process. Compared with the original manual detection method, the efficiency is improved by 300%, and the missed detection rate is reduced to less than 3%.

[0049] Furthermore, the present invention provides an embodiment for low-light scenes at night: In the nighttime operation environment of coal mines, the system is equipped with a low-light compensation camera, which can still clearly capture the driver's operating actions under insufficient lighting conditions. Combined with infrared auxiliary light sources, it ensures the quality of video acquisition. The deep learning model adapts to the characteristics of dark light environments through enhanced training, maintaining an action recognition accuracy rate of more than 95%, solving the problem of traditional manual nighttime detection difficulties.

[0050] Furthermore, the present invention provides an embodiment for adapting multiple vehicle models: At the mixed vehicle weighing center in the logistics park, the system automatically identifies the water tank position characteristics of different vehicles through the pre-stored vehicle model database, dynamically adjusts the camera angle and detection parameters for different vehicle models such as tank trucks and flatbed trucks, and uses transfer learning technology to enable the AI ​​model to quickly adapt to the detection requirements of new vehicle models, realizing accurate detection of multiple vehicle models by one system, thereby reducing the cost of repeated equipment investment.

[0051] Furthermore, the present invention provides an embodiment, in severe weather scenarios: In the windy and rainy environment of coastal ports, the system uses industrial cameras with IP67 protection level and special anti-shake algorithms, which can still stably collect video data under strong wind conditions. By adding a rain interference filtering module, the accuracy of optical flow analysis is improved, thereby ensuring that the detection success rate can still be maintained at more than 85% in typhoon weather, greatly improving operational continuity in extreme weather conditions.

[0052] Through the above steps, automated detection is achieved through artificial intelligence video analysis technology, which completely replaces the traditional visual inspection method of staff, thereby reducing the missed detection rate and effectively avoiding the weighing error problem caused by the water tank not being empty. At the same time, the intelligent control logic module is used to realize full-process automated management. When a compliant operation is detected, the weighing process is automatically triggered, and an alarm is immediately issued in an abnormal situation. The entire detection process does not require human intervention, which significantly reduces the company's labor and management costs, so as to solve the problems of high missed detection rate and high labor cost of traditional vehicle water tank water storage inspection.

Claims

1. A system for intelligently checking and weighing the water level in a vehicle water tank, characterized in that: Includes: Multimodal data acquisition module, used to collect video data and weighing data of the vehicle water tank area; A deep learning action recognition module is used to identify the driver's action of turning on the faucet; The target detection module is used to analyze the water flow state through the optical flow method; Intelligent control logic module, used to determine whether to start weighing based on motion recognition and water flow status; The human-computer interaction module is used to guide the driver's operations and provide feedback on test results.

2. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The multimodal data acquisition module includes an industrial-grade anti-shake camera and a weighing sensor. The camera has a low-light compensation function, and the timestamp synchronization calibration error between the weighing sensor and the video acquisition device is controlled within 50 milliseconds.

3. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The deep learning action recognition module is built based on the improved YOLOv7 spatiotemporal attention network model. Through training and learning, the model can accurately capture the characteristics of human hand movements, especially the key frame sequence of faucet rotation movements, and the model has been lightweight.

4. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The target detection module uses optical flow analysis technology to detect the water flow state by calculating the motion vectors of pixel points in continuous video frames. It can distinguish between still water bodies and flowing water bodies, and comprehensively judge whether the water tank has completed drainage based on the output results of the deep learning model.

5. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The intelligent control logic module uses a dual-threshold judgment mechanism. When the system detects that the driver's action of turning on the faucet is more than 85% complete and the water flow duration is less than 5 seconds, it automatically triggers the weighing process. If it is detected that the driver does not perform the specified action within the specified time, the alarm program is activated.

6. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The human-computer interaction module includes a high-brightness LED display and a voice prompt device. The display shows operation guidance information in real time to guide the driver to complete the water tank draining operation, and at the same time informs the operation status through visual feedback. When an abnormal situation is detected, the driver and management personnel are reminded through sound and light alarms.

7. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The system also includes a vehicle identification subsystem, which triggers a high-definition snapshot camera or RFID reader through a ground sensor coil to obtain vehicle identity information and automatically retrieve pre-stored vehicle file data, which includes standard load capacity and water tank position parameters.

8. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The training data of the deep learning action recognition module is constructed by manually annotating the key action sequences of tap operations, expanding the training samples using data augmentation techniques such as random rotation, brightness adjustment, and noise addition, and optimizing the model training process using an improved loss function.

9. The intelligent system for checking and weighing the water level in a vehicle water tank according to claim 1, characterized in that: The system uses multi-camera video stream stitching technology to eliminate detection blind spots, and stitches monitoring images from multiple angles into a complete view through feature point matching and image fusion algorithms. At the same time, it combines deep learning models to achieve full-view real-time action recognition and water flow status analysis.

10. The intelligent system for checking and weighing water in a vehicle water tank according to claim 1, characterized in that: The system is equipped with a remote management interface. When the system detects that the driver fails to perform operations as required or an abnormal situation occurs, it automatically generates an alarm message and transmits it to the management platform via the network. At the same time, it saves relevant video evidence to support management personnel to remotely view the on-site situation and handle it.

Images