A gas station video intelligent analysis and alarm system and method

Abstract

The application provides a gas station video intelligent analysis and alarm system and method, comprising a guardian identification module, a non-staff identification module, a crane pipe identification module, an oil unloading vehicle identification module, a staff tracking module, a high-risk area setting module and an off-duty time setting module; the guardian identification module is used for identifying guardians; the non-staff identification module is used for identifying the identity of non-staff; the crane pipe identification module is used for identifying crane pipes working in the oil filling area and the oil unloading area; the oil unloading vehicle identification module is used for identifying oil unloading vehicles working in the oil filling area and the oil unloading area; the staff tracking module is used for tracking staff and pedestrians working and walking in the gas station; the high-risk area setting module is used for processing, analyzing and detecting only the targets existing in the concerned area; the application uses a CNN network to identify multiple tasks, and jointly judges the identification results of isolated models, thereby greatly improving the accuracy of identification.

Description

Technical Field

[0001] This invention relates to the field of gas station safety management technology, specifically to a gas station video intelligent analysis and alarm system and method. Background Technology

[0002] Petroleum is the lifeblood of industry and plays a crucial role in the global energy sector. With the booming development of China's industry and economy, the demand for petroleum has increased year by year, leading to a corresponding increase in the number of gas stations. Due to the numerous locations and complex environments of gas stations, coupled with rising safety requirements and increased business volume, the flow of pedestrians and vehicles at gas stations has become more complex and demanding, making it difficult for management personnel to keep up. Traditional video surveillance requires real-time human supervision, which inevitably leads to oversights when dealing with massive amounts of image information. Furthermore, it requires hard disk recorders to store 24-hour video recordings, much of which is invalid, thus wasting significant hardware resources. The video storage devices also require substantial human and financial resources to maintain, and accessing video recordings on hard drives has many drawbacks; in the event of an accident, timely handling is often impossible, frequently missing the optimal opportunity for case resolution. Therefore, traditional video surveillance methods are no longer adequate for meeting the needs of multi-tasking monitoring. To ensure effective tracking of vehicles and personnel in the refueling and unloading areas of gas stations, monitor potentially dangerous activities such as refueling and unloading, and prevent accidents during daily operations, gas stations have begun to apply intelligent video analytics technology to tasks such as real-time monitoring of personnel loading and unloading and operations in the unloading and refueling areas, vehicle identification, and pedestrian tracking.

[0003] Currently, relevant research has been conducted on intelligent video recognition monitoring and alarm systems. CN 200972689Y discloses an intelligent video recognition monitoring and alarm system that analyzes discrete frame images of moving video, combining image brightness and color change information and regional change information to determine whether the features belong to people, animals, or fire, and finally determines whether to trigger an alarm. Existing technologies do not consider using isolated models to perform tasks such as identifying monitoring personnel, loading arms, and unloading vehicles, and jointly judging the results. This makes it impossible to optimize personnel tracking and implement repeated alarm cancellation mechanisms, and it also cannot adjust the allowed time interval for personnel departure, reduce the workload of manual verification, or improve the effectiveness of subsequent data statistics. Furthermore, it cannot incorporate background information into the system design or perform zoning, resulting in poor recognition performance. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes a smart video analysis and alarm system and method for gas stations. The system is reasonably designed, overcomes the shortcomings of the prior art, and has good results.

[0005] To achieve the above objective 1, the present invention adopts the following technical solution:

[0006] A smart video analysis and alarm system for gas stations includes a monitoring personnel identification module, a non-staff identification module, an loading arm identification module, an unloading vehicle identification module, a personnel tracking module, a high-risk area setting module, and an off-duty time setting module.

[0007] The monitoring personnel identification module identifies the identity of monitoring personnel, whether they are wearing work clothes, and whether they have left their posts. The non-staff identification module identifies the identity of non-staff personnel and determines whether they have entered high-risk areas. The loading arm identification module identifies loading arms working in the refueling and unloading areas to monitor refueling and unloading tasks at various points in the gas station. The unloading vehicle identification module identifies unloading vehicles working in the refueling and unloading areas to monitor unloading tasks at various points in the gas station. The personnel tracking module tracks staff and pedestrians working and walking in the gas station, obtaining their activity trajectories and identity IDs within the area. The high-risk area setting module allocates the areas of interest for different algorithms, processing, analyzing, and detecting only targets within the areas of interest. The off-duty time setting module controls alarm thresholds.

[0008] Furthermore, the working process of the guardian identification module includes the following steps:

[0009] a. Collect video image data containing monitoring personnel in actual gas station scenarios;

[0010] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "guardian".

[0011] c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it.

[0012] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the guardian recognition module to obtain the guardian recognition model;

[0013] e. Input the real-time video of the refueling or unloading area to be identified into the monitoring personnel identification model to obtain the identification result.

[0014] Furthermore, the video image data includes videos of the refueling process in the refueling area and videos of the unloading process in the unloading area.

[0015] Furthermore, the non-staff identification module's operation includes the following steps:

[0016] a. Collect video image data of non-staff members in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area;

[0017] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "non-staff".

[0018] c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it.

[0019] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the non-worker recognition module to obtain the non-worker recognition model;

[0020] e. Input the real-time video of the refueling area or unloading area to be identified into the non-staff identification model to obtain the identification result.

[0021] Furthermore, the working process of the loading arm recognition module includes the following steps:

[0022] a. Collect video image data containing loading arms in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area;

[0023] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "arming".

[0024] c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it.

[0025] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module to obtain the loading arm recognition model;

[0026] e. Input the real-time video of the refueling area or unloading area to be identified into the loading arm recognition model to obtain the recognition result.

[0027] Furthermore, the working process of the oil unloading vehicle identification module includes the following steps:

[0028] a. Collect video image data containing unloading vehicles, including videos of the refueling process in the refueling area and the unloading process in the unloading area;

[0029] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "oil unloading vehicles";

[0030] c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it.

[0031] d. Using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module, the preprocessed images are trained to obtain the unloading vehicle recognition model;

[0032] e. Input the real-time video of the refueling or unloading area to be identified into the unloading vehicle identification model to obtain the identification result.

[0033] Furthermore, the personnel tracking module's workflow includes the following steps:

[0034] a. Use deep sort as a personnel tracker;

[0035] b. Use the pre-trained guardian identification module as a detector to correct the tracker's prediction results;

[0036] c. Input the real-time video streams of the refueling area and the unloading area to be tracked into the personnel tracking and detection model to obtain the tracking results.

[0037] Further, in step b, the personnel detection box is divided into high-scoring detection boxes and low-scoring detection boxes by setting high-scoring thresholds and low-scoring thresholds. The high-scoring detection boxes are matched with the trajectory predicted by the personnel tracker. If the high-scoring detection box and the tracking trajectory do not match, the low-scoring detection box is used to match the tracking trajectory.

[0038] Furthermore, the off-duty time setting module starts timing from the first video frame where the monitoring personnel are not detected until the latest frame where the monitoring personnel are not detected. The time difference is compared with the set off-duty time threshold to determine whether an alarm should be triggered.

[0039] To achieve the above objective 2, the present invention adopts the following technical solution:

[0040] A method for intelligent video analysis and alarm of gas stations, employing the aforementioned intelligent video analysis and alarm system for gas stations, specifically involves the following steps:

[0041] S1. Obtain video streams from cameras in the refueling and unloading areas of the gas station, and input the video streams into the gas station video intelligent analysis and alarm system mounted on the video intelligent analysis box;

[0042] S2. When the unloading vehicle identification module identifies the unloading vehicle and the loading arm identification module identifies the loading arm connection, the personnel tracking module and the monitoring personnel identification module are activated to time the departure of the monitoring personnel. When the time exceeds the time threshold required by the gas station management regulations, an alarm will be triggered to indicate that the monitoring personnel have left their post.

[0043] When the unloading vehicle identification module fails to identify the unloading vehicle, the high-risk area module and the non-staff identification module are activated to mark the high-risk oil tank area and oil and gas recovery device of the gas station. When non-staff are detected entering, an alarm for unauthorized personnel entering outside the unloading period is triggered.

[0044] S3. Push the alarm results to the smart terminal and update the data in real time.

[0045] The beneficial technical effects of this invention are as follows:

[0046] This invention proposes a smart video analysis and alarm system and method for gas stations. It combines the identification results of isolated models such as unloading vehicle identification, loading arm identification, and monitoring personnel identification to optimize personnel tracking and repeated alarm elimination mechanisms, achieving 100% accuracy in event identification. It also allows for adjustment of the allowed time interval for personnel to leave the site, reducing the workload of manual confirmation and improving the effectiveness of subsequent data statistics. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the intelligent video analysis and alarm system for gas stations proposed in this invention;

[0048] Figure 2 This is a schematic diagram of the various modules and alarm process proposed in this invention; Detailed Implementation

[0049] The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0050] The terminology used in this disclosure is for descriptive purposes only and is not intended to be limiting. The singular forms “a,” “the,” and “the” used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0051] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and specific examples:

[0052] Example 1

[0053] A smart video analysis and alarm system for gas stations includes a monitoring personnel identification module, a non-staff identification module, an loading arm identification module, an unloading vehicle identification module, a personnel tracking module, a high-risk area setting module, and an off-duty time setting module.

[0054] Specifically, the guardian identification module is used to identify the guardian's identity, whether they are wearing work clothes, and whether they have left their post. The working process of the guardian identification module includes the following steps:

[0055] a. Collect video image data of actual gas station scenes containing monitoring personnel, including videos of the refueling process in the refueling area and the unloading process in the unloading area;

[0056] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "guardian".

[0057] c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it.

[0058] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the guardian recognition module to obtain the guardian recognition model;

[0059] e. Input the real-time video of the refueling or unloading area to be identified into the monitoring personnel identification model to obtain the identification result.

[0060] Specifically, the non-staff identification module is used to identify the identity of non-staff members and determine whether a person has entered a high-risk area. The working process of the non-staff identification module includes the following steps:

[0061] a. Collect video image data of non-staff members in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area;

[0062] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "non-staff".

[0063] c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it.

[0064] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the non-worker recognition module to obtain the non-worker recognition model;

[0065] e. Input the real-time video of the refueling area or unloading area to be identified into the non-staff identification model to obtain the identification result.

[0066] Specifically, the loading arm identification module identifies loading arms operating in the refueling and unloading areas to monitor refueling and unloading tasks at various points in the gas station. The working process of the loading arm identification module includes the following steps:

[0067] a. Collect video image data containing loading arms in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area;

[0068] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "arming".

[0069] c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it.

[0070] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module to obtain the loading arm recognition model;

[0071] e. Input the real-time video of the refueling area or unloading area to be identified into the loading arm recognition model to obtain the recognition result.

[0072] Specifically, the unloading vehicle identification module is used to identify unloading vehicles working in the refueling and unloading areas, and to monitor unloading tasks at various points in the gas station. The working process of the unloading vehicle identification module includes the following steps:

[0073] a. Collect video image data containing unloading vehicles, including videos of the refueling process in the refueling area and the unloading process in the unloading area;

[0074] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "oil unloading vehicles";

[0075] c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it.

[0076] d. Using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module, the preprocessed images are trained to obtain the unloading vehicle recognition model;

[0077] e. Input the real-time video of the refueling or unloading area to be identified into the unloading vehicle identification model to obtain the identification result.

[0078] Specifically, the personnel tracking module is used to track staff and pedestrians working and walking in the gas station, obtaining their activity trajectories and identity IDs within the area. The working process of the personnel tracking module includes the following steps:

[0079] a. Use deep sort as a personnel tracker;

[0080] b. Use the pre-trained guardian identification module as a detector to correct the tracker's prediction results;

[0081] Furthermore, traditional tracking methods use only a single threshold to filter detection boxes. This approach is prone to missed detections and track interruptions under special circumstances such as occlusion or drastic changes in lighting, leading to repeated alarms in personnel tracking. To address the issue of repeated alarms in personnel tracking, this invention divides personnel detection boxes into high-scoring and low-scoring detection boxes by setting high-scoring and low-scoring thresholds. High-scoring detection boxes are matched against the trajectory predicted by the personnel tracker. If occlusion or other factors prevent a high-scoring detection box from matching the tracking trajectory, a low-scoring detection box is used for matching, thus mitigating the problem of track interruption and prioritizing the matching of high-scoring detection boxes, improving the efficiency and success rate of trajectory matching. Each detected personnel tracking trajectory represents a personnel ID. When identifying violations, only one image needs to be captured and pushed, triggering an alarm only once, thus optimizing personnel tracking and eliminating repeated alarms.

[0082] c. Input the real-time video streams of the refueling area and the unloading area to be tracked into the personnel tracking and detection model to obtain the tracking results.

[0083] Specifically, the high-risk area setting module is used to allocate the areas of interest for different algorithms, and only processes, analyzes and detects targets such as people and objects that exist within the areas of interest;

[0084] Specifically, the off-duty time setting module is used to control the alarm threshold. The off-duty time setting module starts counting from the first video frame where the monitoring personnel are not detected until the latest frame where the monitoring personnel are not detected. In this embodiment, the time difference is set to 20 minutes. The time difference is compared with the set off-duty time threshold to determine whether an alarm is triggered, thereby reducing the workload of manual confirmation and improving the effectiveness of subsequent data statistics.

[0085] A method for intelligent video analysis and alarm of gas stations, employing the aforementioned intelligent video analysis and alarm system for gas stations, specifically involves the following steps:

[0086] S1. Obtain video streams from cameras in the refueling and unloading areas of the gas station, and input the video streams into the gas station video intelligent analysis and alarm system mounted on the video intelligent analysis box. The video intelligent analysis box is an edge computing device with image analysis capabilities and an embedded graphics card.

[0087] S2. When the unloading vehicle identification module identifies the unloading vehicle and the loading arm identification module identifies the loading arm connection, the personnel tracking module and the monitoring personnel identification module are activated to time the departure of the monitoring personnel. When the time exceeds the time threshold required by the gas station management regulations, an alarm will be triggered to indicate that the monitoring personnel have left their post.

[0088] When the unloading vehicle identification module fails to identify the unloading vehicle, the high-risk area module and the non-staff identification module are activated to mark the high-risk oil tank area and oil and gas recovery device of the gas station. When non-staff are detected entering, an alarm for unauthorized personnel entering outside the unloading period is triggered.

[0089] S3. Push the alarm results to the smart terminal and update the data in real time. The terminal can be a smart terminal such as a smartphone, desktop computer, tablet computer, or laptop computer, in order to achieve the purpose of intelligent video analysis and alarm at the gas station.

[0090] Example 2

[0091] A smart video analysis and alarm system for gas stations includes a monitoring personnel identification module, a non-staff identification module, an loading arm identification module, an unloading vehicle identification module, a personnel tracking module, a high-risk area setting module, and an off-duty time setting module.

[0092] Specifically, the guardian identification module is used to identify the guardian's identity, whether they are wearing work clothes, and whether they have left their post. The working process of the guardian identification module includes the following steps:

[0093] a. Collect video image data of actual gas station scenes containing monitoring personnel, including videos of the refueling process in the refueling area and the unloading process in the unloading area;

[0094] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "guardian".

[0095] c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it.

[0096] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the guardian recognition module to obtain the guardian recognition model;

[0097] e. Input the real-time video of the refueling or unloading area to be identified into the monitoring personnel identification model to obtain the identification result.

[0098] Specifically, the non-staff identification module is used to identify the identity of non-staff members and determine whether a person has entered a high-risk area. The working process of the non-staff identification module includes the following steps:

[0099] a. Collect video image data of non-staff members in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area;

[0100] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "non-staff".

[0101] c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it.

[0102] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the non-worker recognition module to obtain the non-worker recognition model;

[0103] e. Input the real-time video of the refueling area or unloading area to be identified into the non-staff identification model to obtain the identification result.

[0104] Specifically, the loading arm identification module identifies loading arms operating in the refueling and unloading areas to monitor refueling and unloading tasks at various points in the gas station. The working process of the loading arm identification module includes the following steps:

[0105] a. Collect video image data containing loading arms in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area;

[0106] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "arming".

[0107] c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it.

[0108] d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module to obtain the loading arm recognition model;

[0109] e. Input the real-time video of the refueling area or unloading area to be identified into the loading arm recognition model to obtain the recognition result.

[0110] Specifically, the unloading vehicle identification module is used to identify unloading vehicles working in the refueling and unloading areas, and to monitor unloading tasks at various points in the gas station. The working process of the unloading vehicle identification module includes the following steps:

[0111] a. Collect video image data containing unloading vehicles, including videos of the refueling process in the refueling area and the unloading process in the unloading area;

[0112] b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "oil unloading vehicles";

[0113] c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it.

[0114] d. Using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module, the preprocessed images are trained to obtain the unloading vehicle recognition model;

[0115] e. Input the real-time video of the refueling or unloading area to be identified into the unloading vehicle identification model to obtain the identification result.

[0116] Specifically, the personnel tracking module is used to track staff and pedestrians working and walking in the gas station, obtaining their activity trajectories and identity IDs within the area. The working process of the personnel tracking module includes the following steps:

[0117] a. Use deep sort as a personnel tracker;

[0118] b. Use the pre-trained guardian identification module as a detector to correct the tracker's prediction results;

[0119] Furthermore, traditional tracking methods use only a single threshold to filter detection boxes. This approach is prone to missed detections and track interruptions under special circumstances such as occlusion or drastic changes in lighting, leading to repeated alarms in personnel tracking. To address the issue of repeated alarms in personnel tracking, this invention divides personnel detection boxes into high-scoring and low-scoring detection boxes by setting high-scoring and low-scoring thresholds. High-scoring detection boxes are matched against the trajectory predicted by the personnel tracker. If occlusion or other factors prevent a high-scoring detection box from matching the tracking trajectory, a low-scoring detection box is used for matching, thus mitigating the problem of track interruption and prioritizing the matching of high-scoring detection boxes, improving the efficiency and success rate of trajectory matching. Each detected personnel tracking trajectory represents a personnel ID. When identifying violations, only one image needs to be captured and pushed, triggering an alarm only once, thus optimizing personnel tracking and eliminating repeated alarms.

[0120] c. Input the real-time video streams of the refueling area and the unloading area to be tracked into the personnel tracking and detection model to obtain the tracking results.

[0121] Specifically, the high-risk area setting module is used to allocate the areas of interest for different algorithms, and only processes, analyzes and detects targets such as people and objects that exist within the areas of interest;

[0122] Specifically, the off-duty time setting module is used to control the alarm threshold. The off-duty time setting module starts counting from the first video frame where the monitoring personnel are not detected until the latest frame where the monitoring personnel are not detected. In this embodiment, the time difference is set to 10 minutes. The time difference is compared with the set off-duty time threshold to determine whether an alarm is triggered, thereby reducing the workload of manual confirmation and improving the effectiveness of subsequent data statistics.

[0123] A method for intelligent video analysis and alarm of gas stations, employing the aforementioned intelligent video analysis and alarm system for gas stations, specifically involves the following steps:

[0124] S1. Obtain video streams from cameras in the refueling and unloading areas of the gas station, and input the video streams into the gas station video intelligent analysis and alarm system mounted on the video intelligent analysis box. The video intelligent analysis box is an edge computing device with image analysis capabilities and an embedded graphics card.

[0125] S2. When the unloading vehicle identification module identifies the unloading vehicle and the loading arm identification module identifies the loading arm connection, the personnel tracking module and the monitoring personnel identification module are activated to time the departure of the monitoring personnel. When the time exceeds the time threshold required by the gas station management regulations, an alarm will be triggered to indicate that the monitoring personnel have left their post.

[0126] When the unloading vehicle identification module fails to identify the unloading vehicle, the high-risk area module and the non-staff identification module are activated to mark the high-risk oil tank area and oil and gas recovery device of the gas station. When non-staff are detected entering, an alarm for unauthorized personnel entering outside the unloading period is triggered.

[0127] S3. Push the alarm results to the smart terminal and update the data in real time. The terminal can be a smart terminal such as a smartphone, desktop computer, tablet computer, or laptop computer, in order to achieve the purpose of intelligent video analysis and alarm at the gas station.

[0128] For any parts not mentioned in this invention, existing technologies can be used or referenced.

[0129] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A video intelligent analysis and alarm system for gas stations, characterized in that, It includes modules for identifying supervisors, identifying non-staff members, identifying loading arms, identifying unloading vehicles, tracking personnel, setting up high-risk areas, and setting off time. The guardian identification module is used to identify the identity of the guardian, whether they are wearing work clothes, and to determine whether the person has left their post; the non-staff identification module is used to identify the identity of non-staff and to determine whether the person has entered a high-risk area. The loading arm identification module identifies loading arms operating in the refueling and unloading areas to monitor refueling and unloading tasks at various points within the gas station; the unloading vehicle identification module identifies unloading vehicles operating in the refueling and unloading areas to monitor unloading tasks at various points within the gas station; the personnel tracking module tracks staff and pedestrians working and walking within the gas station, obtaining their activity trajectories and identity IDs; the high-risk area setting module allocates areas of interest for different algorithms, processing, analyzing, and detecting only targets within those areas; and the off-duty time setting module controls alarm thresholds. The working process of the loading arm recognition module includes the following steps: a. Collect video image data containing loading arms in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area; b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "arming". c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it. d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module to obtain the loading arm recognition model; e. Input the real-time video of the refueling area or unloading area to be identified into the loading arm recognition model to obtain the recognition result; The working process of the oil unloading vehicle identification module includes the following steps: a. Collect video image data containing unloading vehicles, including videos of the refueling process in the refueling area and the unloading process in the unloading area; b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "oil unloading vehicles"; c. Perform data augmentation on the training set, adjusting contrast and brightness, scaling it to the same size, and normalizing it. d. Using the classic object detection network YOLOv3 as the baseline network for the loading arm recognition module, the preprocessed images are trained to obtain the unloading vehicle recognition model; e. Input the real-time video of the refueling area or unloading area to be identified into the unloading vehicle identification model to obtain the identification result; The operation of the personnel tracking module includes the following steps: a. Use deep sort as a personnel tracker; b. Use the pre-trained monitoring personnel identification module as a detector to correct the prediction results of the tracker; this step b specifically includes: dividing the personnel detection box into high-scoring detection boxes and low-scoring detection boxes by setting high-scoring thresholds and low-scoring thresholds, matching the high-scoring detection boxes with the trajectory predicted by the personnel tracker, and if the high-scoring detection boxes and the tracking trajectory do not match, then matching the tracking trajectory with low-scoring detection boxes. c. Input the real-time video streams of the refueling area and the unloading area to be tracked into the personnel tracking and detection model to obtain the tracking results.

2. The intelligent video analysis and alarm system for gas stations according to claim 1, characterized in that, The operation of the guardian identification module includes the following steps: a. Collect video image data containing monitoring personnel in actual gas station scenarios; b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them with "guardian". c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it. d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the guardian recognition module to obtain the guardian recognition model; e. Input the real-time video of the refueling or unloading area to be identified into the monitoring personnel identification model to obtain the identification result.

3. The intelligent video analysis and alarm system for gas stations according to claim 2, characterized in that, The video image data includes videos of the refueling process in the refueling area and videos of the unloading process in the unloading area.

4. The intelligent video analysis and alarm system for gas stations according to claim 1, characterized in that, The non-staff identification module operates through the following steps: a. Collect video image data of non-staff members in the actual scene of the gas station, including videos of the refueling process in the refueling area and videos of the unloading process in the unloading area; b. Decompose the video into static frames, extract the images containing the objects to be identified, and divide the images into training and validation sets; use LabelImg annotation software to mark the detection boxes of the identified targets on the training and validation sets, and label them as "non-staff". c. Perform data augmentation on the training set, adjust the contrast and brightness, scale it to the same size, and normalize it. d. The preprocessed images were trained using the classic object detection network YOLOv3 as the baseline network for the non-worker recognition module to obtain the non-worker recognition model; e. Input the real-time video of the refueling area or unloading area to be identified into the non-staff identification model to obtain the identification result.

5. The intelligent video analysis and alarm system for gas stations according to claim 1, characterized in that, The off-duty time setting module starts timing from the first video frame where the monitoring personnel are not detected until the latest frame where the monitoring personnel are not detected. The time difference is compared with the set off-duty time threshold to determine whether an alarm should be triggered.

6. A method for intelligent video analysis and alarm at gas stations, characterized in that, The specific process of using the intelligent video analysis and alarm system for gas stations as described in any one of claims 1-5 is as follows: S1. Obtain video streams from cameras in the refueling and unloading areas of the gas station, and input the video streams into the gas station video intelligent analysis and alarm system mounted on the video intelligent analysis box; S2. When the unloading vehicle identification module identifies the unloading vehicle and the loading arm identification module identifies the loading arm connection, the personnel tracking module and the monitoring personnel identification module are activated to time the departure of the monitoring personnel. When the time exceeds the time threshold required by the gas station management regulations, an alarm will be triggered to indicate that the monitoring personnel have left their post. When the unloading vehicle identification module fails to identify the unloading vehicle, the high-risk area module and the non-staff identification module are activated to mark the high-risk oil tank area and oil and gas recovery device of the gas station. When non-staff are detected entering, an alarm for unauthorized personnel entering outside the unloading period is triggered. S3. Push the alarm results to the smart terminal and update the data in real time.

Images