Muck truck garbage dumping monitoring and warning method based on multi-stage image recognition

Abstract

The invention discloses a muck truck garbage dumping monitoring and warning method based on multi-stage image recognition, relates to the technical field of garbage dumping monitoring, and specifically comprises a multi-stage image recognition method and a multi-information fusion method. According to the multi-stage image recognition method, multi-type camera image data, a slag car exclusive feature library, camera GPS coordinate information and scenarized threshold parameters are combined. According to the muck truck garbage dumping monitoring and warning method based on multi-stage image recognition, multi-type camera image data, a muck truck exclusive feature library, camera GPS coordinate information and scenarized threshold parameters are matched with one another, and multi-type camera data, feature library data, GPS data, threshold data and a pre-training model library are cooperatively scheduled; the method achieves the triggering of the starting of the next link through the result of the previous link, guarantees the real-time performance and stability of the whole monitoring process through the seamless connection of data and a model between links, and achieves the unattended continuous operation of 7 * 24 hours.

Description

Technical Field

[0001] This invention relates to the field of waste dumping monitoring technology, specifically to a method for monitoring and alarming waste dumping by construction waste trucks based on multi-stage image recognition. Background Technology

[0002] Monitoring the dumping process of garbage trucks is a crucial part of urban management and environmental protection. It is not simply a matter of tracking the whereabouts of a vehicle, but a systematic project involving multiple aspects such as public safety, environmental governance, urban image, and economic benefits.

[0003] However, in the existing monitoring process of dump trucks dumping garbage, when it is necessary to rely on staff to view multiple camera feeds in real time, it is difficult to cover the entire time period, such as at night and in the early morning. This results in low efficiency of manual monitoring, and when a single screen monitors ≥8 feeds simultaneously, the missed detection rate of garbage dumping events is ≥35%.

[0004] Existing dump truck dumping monitoring technology can only identify whether a dump truck exists or whether an object is dumped, without linking the entire process of dump truck dumping, dumping action, and garbage distribution. This results in a false alarm rate (e.g., a truck unloading normally is judged as dumping) of ≥28% and a false alarm rate (e.g., a dump truck dumping but garbage accumulation is not detected) of ≥22%. This single identification link increases the risk of false alarms and false alarms.

[0005] Existing waste identification systems only use the presence of waste as an alarm condition, without setting differentiated volume / area thresholds based on regional attributes (such as main roads or demolition areas). This results in frequent alarms for small amounts of waste (such as 0.2 m³) on main roads, while large amounts of waste (such as 1.0 m³) in demolition areas are not alarmed in a timely manner, thus making the threshold settings unsuitable for different scenarios.

[0006] Existing technologies take an average of ≥15 seconds from identification to alarm information generation, and do not link key data such as camera GPS coordinates and dump truck license plates, which makes it impossible for regulatory personnel to quickly locate and handle the situation, delaying the best time for cleanup, resulting in poor linkage between identification and alarm. Summary of the Invention

[0007] This invention provides a method for monitoring and alarming the dumping of waste by construction waste trucks based on multi-stage image recognition, which solves the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: a method for monitoring and alarming the dumping of garbage by muck trucks based on multi-stage image recognition, including a multi-stage image recognition method and a multi-information fusion method;

[0009] The multi-stage image recognition method combines multi-type camera image data, a database of features specific to construction waste trucks, camera GPS coordinate information, and scenario-based threshold parameters to enable real-time monitoring and automatic alarm of illegal dumping of waste by construction waste trucks in the field of urban management.

[0010] The multi-information fusion method establishes differentiated alarm standards through pre-configured regional thresholds, and then uses multi-stage identification results to trigger precise push notifications during actual monitoring, so as to ensure that urban management personnel obtain full-dimensional information on location, target and magnitude.

[0011] Optionally, the multi-type cameras in the multi-stage image recognition method include high-definition cameras, infrared cameras, and PTZ cameras.

[0012] Optionally, the feature library specific to dump trucks in the multi-stage image recognition method includes structural, appearance, and identification features.

[0013] Optionally, the specific operation steps of the multi-stage image recognition method are as follows:

[0014] S1, Adaptive frame skipping

[0015] The frame sampling frequency is dynamically adjusted according to the scene to minimize computing power consumption while ensuring no missed detection. The frame sampling frequency is dynamically adjusted based on the scene judgment. If there is a suspected target, the next step is triggered directly. If there is no suspected target, the frame sampling frequency is maintained and monitored in a loop.

[0016] S2, Multi-feature Dump Truck Detection

[0017] Accurate identification of dump trucks is achieved through multi-dimensional feature fusion. The initial judgment relies on structural features, and the final judgment is made by combining image detection of yellow license plate with black characters, side of the truck bed and roof light of the dump truck.

[0018] S3, Tomb raiding motion recognition

[0019] Accurately determine the dumping behavior through timing action analysis, receive valid dump truck data output from the multi-feature dump truck detection stage, and keep the target coordinates stable within 3 consecutive frames;

[0020] S4, Waste Quantification

[0021] Waste areas are segmented and their actual size is converted using parameters such as pixel count and camera installation location. Specifically, image detection and classification algorithms are used to mark waste areas, connect consecutive waste areas, and perform size conversion based on camera parameters. The volume of waste is estimated using standard waste height.

[0022] Optionally, the dynamic adjustment rule in S1 of the multi-stage image recognition method is as follows: 1 frame / second for static scenes to save computing power; 5 frames / second for dynamic scenes to capture target detection; and 10 frames / second for suspected target scenes to capture action details.

[0023] Optionally, in S4 of the multi-stage image recognition method, the scene transition threshold matching is performed by matching a preset scene threshold based on the camera's GPS positioning.

[0024] Optionally, in step S4 of the multi-stage image recognition method, when the alarm volume of the main road is not less than 0.3 m³, an alarm is triggered and pushed to the competent department.

[0025] When the alarm volume in the demolition area is not less than 1.0 m³, the alarm will be sent to the competent department after it is triggered.

[0026] When the alarm volume in the sensitive area is not less than 0.1m³, the alarm will be actively pushed to the superior department after it is triggered.

[0027] Optionally, the specific operation steps of the multi-information fusion method are as follows:

[0028] S1, Input Initialization

[0029] Multi-stage recognition preparation: real-time image stream reading from the camera;

[0030] GPS linkage operation: The camera's GPS information is bound to the camera ID, and can be directly accessed in subsequent stages;

[0031] S2, Adaptive Frame Drop

[0032] GPS linkage: Calls GPS information to query the attributes of the area, including main roads, demolition areas and sensitive areas;

[0033] Multi-stage recognition: Read cached images and determine scene dynamism;

[0034] S3, Multi-feature Dump Truck Detection

[0035] Multi-stage recognition: Use image algorithm models to detect frame-by-frame images, output the confidence score and target box coordinates of the dump truck, extract the structure, license plate and features, and output the confidence score of each category;

[0036] GPS linkage: GPS matches high-risk areas, with construction sites and suburbs being high-risk and urban areas and residential areas being low-risk, and calculates the final confidence level;

[0037] S4, Waste Quantification

[0038] Multi-stage identification: Using image detection and classification algorithms, garbage areas are marked, continuous garbage is connected into regions, the actual area of ​​the garbage is calculated and the overall volume is estimated;

[0039] GPS linkage: GPS matching scene thresholds, where the threshold is no less than 0.3m³ for main roads, no less than 1.0m³ for demolition areas, and no less than 0.1m³ for sensitive areas;

[0040] S5, Scene Adaptation Alarm Handling

[0041] Multi-stage identification: Integrating dump truck license plates and waste quantification data to generate identification results;

[0042] GPS linkage: GPS coordinate conversion query to the relevant law enforcement area, push notifications according to scenario level, with sensitive areas pushed to higher-level departments and ordinary areas pushed to directly subordinate departments.

[0043] The present invention has the following beneficial effects:

[0044] 1. This method for monitoring and alarming dump trucks dumping waste based on multi-stage image recognition, through the mutual coordination of multi-type camera image data, dump truck-specific feature library, camera GPS coordinate information, and scenario-based threshold parameters, coordinates and schedules multi-type camera data, feature library data, GPS data, threshold data, and pre-trained model library to achieve the result of the previous stage triggering the start of the next stage. Through the seamless connection of data and models between stages, the real-time performance and stability of the entire monitoring process are ensured, enabling 24 / 7 unattended continuous operation.

[0045] 2. This multi-stage image recognition-based method for monitoring and alarming dump trucks dumping garbage relies on the fusion of multiple information sources, including GPS coordinates, dump truck license plates, and garbage heat maps. It first establishes differentiated alarm standards using pre-configured regional thresholds, and then uses the multi-stage recognition results to trigger precise push notifications during actual monitoring. This ensures that urban management personnel obtain comprehensive information on location, target, and scale, thereby achieving a highly efficient monitoring effect that balances 24 / 7 unattended operation with low false alarms and missed alarms.

[0046] 3. This method for monitoring and alarming the dumping of garbage by construction waste trucks based on multi-stage image recognition only requires the connection of an integrated image analysis and detection terminal to the existing cameras, without the need for additional hardware. Moreover, the model training data can reuse the existing image library of the urban management bureau. Compared with the traditional multi-system deployment, the overall cost is reduced by 60%. At the same time, it realizes data interoperability, process linkage and result unification, further improving the collaborative efficiency of urban management law enforcement. Attached Figure Description

[0047] Figure 1 This is a flowchart of the multi-stage image recognition method in this invention;

[0048] Figure 2 This is a flowchart of the multi-information fusion method in this invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 and Figure 2 This invention provides a technical solution: a monitoring and alarm method for dump trucks dumping garbage based on multi-stage image recognition, including a multi-stage image recognition method and a multi-information fusion method, which takes into account both "24 / 7 unattended operation" and "low false alarm rate" for efficient monitoring. The collaborative method of multi-stage image recognition improves the accuracy and real-time performance of the monitoring system, and the full-process linkage of adaptive frame extraction, multi-feature dump truck detection, dumping bucket recognition and garbage quantification is the core technical support.

[0051] The multi-stage image recognition method combines multi-type camera image data, a database of features specific to construction waste trucks, camera GPS coordinate information, and scenario-based threshold parameters to enable real-time monitoring and automatic alarms for illegal dumping of waste by construction waste trucks in urban management. This solution addresses scenarios with traditional manual monitoring (missed detection rate ≥35%), single-stage identification (false alarm rate ≥28%), undifferentiated thresholds (many invalid alarms on main roads / missed alarms in demolition areas), and delayed alarm linkage (time ≥15 seconds) by employing "adaptive frame extraction." The strategy, which integrates multi-feature fusion for dump truck detection, bucket tipping action recognition, waste quantification, and rapid alarm linkage, achieves 24 / 7 unattended operation, high accuracy (dump truck recognition ≥98.2%, bucket tipping action ≥96.5%), low false alarms and missed alarms (overall false alarms ≤2.5%, missed alarms ≤1.8%), scenario adaptability (regionally differentiated thresholds), and rapid response (alarm push ≤8 seconds) for dump truck waste dumping monitoring. This provides integrated "identification-location-disposal" technical support for urban management.

[0052] The multi-information fusion method establishes differentiated alarm standards by pre-configuring regional thresholds, and then uses multi-stage identification results to trigger precise push notifications in actual monitoring, so as to ensure that urban management personnel obtain full-dimensional information on location, target and scale.

[0053] The multi-stage image recognition method utilizes various camera types, including high-definition cameras, infrared cameras, and PTZ cameras. It only requires connecting an integrated image analysis and detection terminal to existing cameras, without the need for additional hardware. Furthermore, the model training data can reuse the existing image library of the urban management bureau. Compared to traditional multi-system deployments, the overall cost is reduced by 60%, while simultaneously achieving "data interoperability, process linkage, and unified results," further improving the collaborative efficiency of urban management law enforcement.

[0054] The multi-stage image recognition method's dedicated feature library for dump trucks includes structural, appearance, and identification features.

[0055] The specific steps of the multi-stage image recognition method are as follows:

[0056] S1, Adaptive frame skipping

[0057] The frame sampling frequency is dynamically adjusted according to the scene to minimize computing power consumption while ensuring no missed detections. The frame sampling frequency is dynamically adjusted based on the scene judgment of the image. If there is a suspected target, the next step is triggered directly. If there is no suspected target, the frame sampling frequency is maintained and the monitoring is carried out in a loop. This solves the problem of traditional "fixed frame sampling" (such as high missed detection rate of 1 frame / second and wasted computing power of 10 frames / second).

[0058] S2, Multi-feature Dump Truck Detection

[0059] The system accurately identifies dump trucks by fusing multi-dimensional features. It relies on structural features for initial identification and combines image detection of yellow license plates with black characters, the side of the truck bed, and the roof lights of the dump truck for further identification. This solves the problem of traditional "single appearance recognition". The trigger condition is to receive adaptive frame extraction output of "suspected target frames".

[0060] S3, Tomb raiding motion recognition

[0061] By analyzing the timing of actions, the dump truck can accurately determine the dumping behavior and receive valid dump truck data output from the multi-feature dump truck detection process. The target coordinates remain stable within three consecutive frames, which solves the problem of traditional "single-frame static recognition" (such as misjudging a stationary, raised dump truck as a dumping).

[0062] S4, Waste Quantification

[0063] The garbage area is segmented and its actual size is converted by using parameters of pixels and camera installation position. Specifically, image detection and classification algorithms are used to mark garbage areas, connect continuous garbage areas, and convert the size by combining camera parameters. The garbage volume is estimated by using standard garbage height, which solves the problem of traditional "no quantification". The trigger condition is to receive "valid bucket-turning action" data output by the bucket-turning action recognition link and the bucket-turning action ends.

[0064] The dynamic adjustment rules in S1 of the multi-stage image recognition method are as follows: 1 frame / second for static scenes to save computing power; 5 frames / second for dynamic scenes to capture target detection; and 10 frames / second for suspected target scenes to capture action details.

[0065] In the S4 stage of the multi-stage image recognition method, the scene transition threshold matching is based on the camera's GPS positioning and a preset scene threshold is matched.

[0066] In the S4 of the multi-stage image recognition method, when the alarm volume of the main road is not less than 0.3m³, the alarm is triggered and pushed to the competent department.

[0067] When the alarm volume in the demolition area is not less than 1.0 m³, the alarm will be sent to the competent department after it is triggered.

[0068] When the alarm volume in the sensitive area is not less than 0.1m³, the alarm will be actively pushed to the superior department after it is triggered.

[0069] The specific operational steps of the multi-information fusion method are as follows:

[0070] S1, Input Initialization

[0071] Multi-stage recognition preparation: real-time image stream reading from the camera;

[0072] GPS linkage operation: The camera's GPS information is bound to the camera ID, and can be directly accessed in subsequent stages;

[0073] S2, Adaptive Frame Drop

[0074] GPS linkage: Calls GPS information to query the attributes of the area, including main roads, demolition areas and sensitive areas;

[0075] Multi-stage recognition: Read cached images and determine scene dynamism;

[0076] S3, Multi-feature Dump Truck Detection

[0077] Multi-stage recognition: Use image algorithm models to detect frame-by-frame images, output the confidence score and target box coordinates of the dump truck, extract the structure, license plate and features, and output the confidence score of each category;

[0078] GPS linkage: GPS matches high-risk areas, with construction sites and suburbs being high-risk and urban areas and residential areas being low-risk, and calculates the final confidence level;

[0079] S4, Waste Quantification

[0080] Multi-stage identification: Using image detection and classification algorithms, garbage areas are marked, continuous garbage is connected into regions, the actual area of ​​the garbage is calculated and the overall volume is estimated;

[0081] GPS linkage: GPS matching scene thresholds, where the threshold is no less than 0.3m³ for main roads, no less than 1.0m³ for demolition areas, and no less than 0.1m³ for sensitive areas;

[0082] S5, Scene Adaptation Alarm Handling

[0083] Multi-stage identification: Integrating dump truck license plates and waste quantification data to generate identification results;

[0084] GPS linkage: GPS coordinate conversion query to the relevant law enforcement area, push notifications according to scenario level, with sensitive areas pushed to higher-level departments and ordinary areas pushed to directly subordinate departments.

[0085] Traditional technologies (such as those that can only identify dump trucks or detect garbage) are essentially fragmented, single-stage monitoring without a fully integrated process. This inevitably leads to "identification drift." For example, the accuracy rate is acceptable during the day (about 85%), but it drops below 60% at night or in rainy conditions due to noise interference. Furthermore, as the monitoring duration increases (such as after 12 hours of continuous monitoring), the false negative rate rises from 22% to 35%.

[0086] Through "multi-model collaborative scheduling", the system achieves "identification drift"-free operation. During the day, the accuracy rate of identifying dump trucks on main roads remains stable at 99.0%, while at night, it can still maintain 97.5% in demolition areas. The false alarm rate has been ≤2.5% for 72 consecutive hours of monitoring. Furthermore, by associating the camera with GPS coordinates, the system directly outputs the geodetic coordinates of the garbage dumping point (error ≤5m).

[0087] Because the "monitoring results" output by traditional manual monitoring are only subjective judgments of "whether there is garbage", there is no quantitative data or location information;

[0088] The coordinates output by a single recognition algorithm are "relative coordinates within the image" (such as x1, y1, x2, y2 of a dump truck in the image), not the geographic coordinates required by urban management enforcement, and therefore cannot be directly used for rapid location and disposal. For urban management bureaus, the core objective of monitoring is "rapid detection - accurate location - efficient disposal," and traditional technologies have gaps in these three areas: the missed detection rate of manual monitoring is ≥35% (almost ineffective at night / early morning), the false alarm rate of a single algorithm is ≥28% (such as misjudging a truck unloading as garbage dumping), and the alarm push takes ≥15 seconds and lacks license plate / volume information, causing law enforcement officers to spend more than 15 minutes searching for disposal points, resulting in a garbage cleanup timeliness rate of only 55%.

[0089] To fill the gaps mentioned above, firstly, the multi-stage identification process improves the overall monitoring accuracy to over 96.5% through "adaptive frame extraction (solving computational waste and missed detection) → multi-feature fusion (solving misidentification of dump trucks) → action features (solving missed judgment of dumping) → garbage detection quantification (solving garbage lack of quantification)," completely eliminating the efficiency bottleneck of manual monitoring. The monitoring efficiency of a single camera is 20 times higher than that of manual monitoring.

[0090] The alarm linkage mechanism integrates "GPS + license plate + heat map" to reduce the location processing time from 15 minutes to within 5 minutes, push notification time is ≤8 seconds, and the timely clearing rate is increased to 100%.

[0091] Thirdly, the scenario-based threshold design (0.3m³ for main roads, 1.0m³ for demolition areas, and 0.1m³ for sensitive areas) reduces invalid alarms on main roads by 55% and missed alarms in demolition areas by 70%, perfectly meeting the urban management bureau's need for "differentiated management of different areas".

[0092] In addition, the present invention also has the advantage of "low cost adaptability": traditional high-precision monitoring requires the deployment of multiple independent systems (such as dump truck identification system, garbage detection system, alarm system), which has high hardware costs and data is not interoperable;

[0093] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. The terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, unless otherwise explicitly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Moreover, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring and alarming the dumping of waste by construction waste trucks based on multi-stage image recognition, characterized in that: This includes multi-stage image recognition methods and multi-information fusion methods; The multi-stage image recognition method combines multi-type camera image data, a database of features specific to construction waste trucks, camera GPS coordinate information, and scenario-based threshold parameters to enable real-time monitoring and automatic alarm for illegal dumping of waste by construction waste trucks in the field of urban management. The multi-information fusion method establishes differentiated alarm standards through pre-configured regional thresholds, and then uses multi-stage identification results to trigger precise push notifications during actual monitoring, so as to ensure that urban management personnel obtain full-dimensional information on location, target and magnitude.

2. The method for monitoring and alarming dumping of construction waste trucks based on multi-stage image recognition according to claim 1, characterized in that: The multi-stage image recognition method uses various types of cameras, including high-definition cameras, infrared cameras, and PTZ cameras.

3. The method for monitoring and alarming dumping of construction waste trucks based on multi-stage image recognition according to claim 1, characterized in that: The multi-stage image recognition method's dedicated feature library for dump trucks includes structural, appearance, and identification features.

4. The method for monitoring and alarming dumping of construction waste trucks based on multi-stage image recognition according to claim 1, characterized in that: The specific operation steps of the multi-stage image recognition method are as follows: S1, Adaptive frame skipping The frame sampling frequency is dynamically adjusted according to the scene to minimize computing power consumption while ensuring no missed detection. Specifically, the frame sampling frequency is dynamically adjusted based on the scene judgment. If there is a suspected target, the next step is triggered directly. If there is no suspected target, the frame sampling frequency is maintained and monitored in a loop. S2, Multi-feature Dump Truck Detection Accurate identification of dump trucks is achieved through multi-dimensional feature fusion. The initial judgment relies on structural features, and the final judgment is made by combining image detection of yellow license plate with black characters, side of the truck bed and roof light of the dump truck. S3, Tomb raiding motion recognition Accurately determine the dumping behavior through timing action analysis, receive valid dump truck data output from the multi-feature dump truck detection stage, and keep the target coordinates stable within 3 consecutive frames; S4, Waste Quantification Waste areas are segmented and their actual size is converted using parameters such as pixel count and camera installation location. Specifically, image detection and classification algorithms are used to mark waste areas, connect consecutive waste areas, and perform size conversion based on camera parameters. The volume of waste is estimated using standard waste height.

5. The method for monitoring and alarming the dumping of construction waste by dump trucks based on multi-stage image recognition according to claim 4, characterized in that: The dynamic adjustment rules in S1 of the multi-stage image recognition method are as follows: 1 frame / second for static scenes to save computing power; 5 frames / second for dynamic scenes to capture target detection; and 10 frames / second for suspected target scenes to capture action details.

6. The method for monitoring and alarming the dumping of construction waste by dump trucks based on multi-stage image recognition according to claim 4, characterized in that: In S4 of the multi-stage image recognition method, the scene transition threshold matching is performed by matching a preset scene threshold based on the camera's GPS positioning.

7. The method for monitoring and alarming dumping of construction waste trucks based on multi-stage image recognition according to claim 6, characterized in that: In step S4 of the multi-stage image recognition method, when the alarm volume of the main road is not less than 0.3m³, an alarm is triggered and pushed to the competent department. When the alarm volume in the demolition area is not less than 1.0m³, the alarm will be sent to the competent department after it is triggered. When the alarm volume in the sensitive area is not less than 0.1m³, the alarm will be actively pushed to the superior department after it is triggered.

8. The method for monitoring and alarming dumping of construction waste trucks based on multi-stage image recognition according to claim 1, characterized in that: The specific operation steps of the multi-information fusion method are as follows: S1, Input Initialization Multi-stage recognition preparation: real-time image stream reading from the camera; GPS linkage operation: The camera's GPS information is bound to the camera ID, and can be directly accessed in subsequent stages; S2, Adaptive Frame Drop GPS linkage: Calls GPS information to query the attributes of the area, including main roads, demolition areas and sensitive areas; Multi-stage recognition: Read cached images and determine scene dynamism; S3, Multi-feature Dump Truck Detection Multi-stage recognition: Use image algorithm models to detect frame-by-frame images, output the confidence score and target box coordinates of the dump truck, extract the structure, license plate and features, and output the confidence score of each category; GPS linkage: GPS matches high-risk areas, with construction sites and suburbs being high-risk and urban areas and residential areas being low-risk, and calculates the final confidence level; S4, Waste Quantification Multi-stage identification: Using image detection and classification algorithms, garbage areas are marked, continuous garbage is connected into regions, the actual area of ​​the garbage is calculated and the overall volume is estimated; GPS linkage: GPS matching scene thresholds, where the threshold is no less than 0.3m³ for main roads, no less than 1.0m³ for demolition areas, and no less than 0.1m³ for sensitive areas; S5, Scene Adaptation Alarm Handling Multi-stage identification: Integrating dump truck license plates and waste quantification data to generate identification results; GPS linkage: GPS coordinate conversion query to the relevant law enforcement area, push notifications according to scenario level, with sensitive areas pushed to higher-level departments and ordinary areas pushed to directly subordinate departments.

Images