Household garbage throwing behavior early warning method and system based on computer vision

Abstract

The invention provides a household garbage throwing behavior early warning method and system based on computer vision, and relates to the technical field of computer vision garbage throwing early warning, and the method comprises the steps: deploying a vision capturing device, delimiting a recovery equipment body, normal garbage throwing and peripheral throwing monitoring areas, and configuring dynamic video analysis parameters. Outputting a standardized visual monitoring system; capturing a putting behavior video stream through a standardized visual monitoring system, executing personnel target detection, motion trail tracking, garbage object feature recognition, putting behavior mode analysis and putting result validity evaluation, and outputting putting behavior analysis data; constructing a putting behavior grading system, calculating putting behavior characteristic parameters, carrying out behavior grading, and matching early warning instruction parameters; and an early warning instruction is generated according to the early warning instruction parameter, so that real-time monitoring, intelligent identification and graded early warning of garbage throwing behaviors are realized, the garbage throwing behaviors of the public are effectively standardized, and the urban environmental health management level is improved.

Description

Technical Field

[0001] This invention relates to the field of computer vision-based waste disposal early warning technology, and in particular to a computer vision-based method and system for early warning of household waste disposal behavior. Background Technology

[0002] Existing household waste management systems primarily focus on waste sorting and recycling, with relatively weak oversight of waste disposal behavior. In public places, uncivilized waste disposal behaviors such as throwing waste from a distance and indiscriminate littering occur frequently, not only affecting environmental sanitation but also increasing the workload of cleaning and maintenance.

[0003] Traditional monitoring methods rely primarily on manual patrols or simple video playback, lacking real-time capabilities and intelligent analysis, making it difficult to intervene and effectively manage uncivilized waste disposal. Existing monitoring systems typically only passively record images, unable to actively identify and analyze waste disposal behavior, let alone provide tiered warnings and interventions based on the severity of the behavior. This passive monitoring approach is inefficient, fails to effectively guide and regulate behavior, and cannot meet the intelligent needs of modern urban waste management.

[0004] Therefore, it is necessary to provide a computer vision-based early warning method and system for household waste disposal behavior to solve the above-mentioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a computer vision-based method and system for early warning of household waste disposal behavior, which solves the problems of insufficient supervision of waste disposal behavior, lack of real-time analysis and intelligent early warning capabilities in existing waste management systems.

[0006] This invention provides a computer vision-based early warning method for household waste disposal behavior, the method comprising: Visual capture devices are deployed in a differentiated manner around the household waste recycling equipment. The recycling equipment body area, normal waste disposal area and peripheral disposal monitoring area are delineated and a spatial mapping relationship is established. Dynamic video analysis parameters adapted to the scene are configured to output a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. The standardized visual monitoring system captures video streams of household waste disposal behavior in real time, and performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, disposal behavior pattern analysis, and disposal result effectiveness evaluation on the video streams in sequence, outputting multi-dimensional disposal behavior analysis data that includes personnel spatiotemporal information, object motion trajectory, and disposal effectiveness judgment results. A three-dimensional delivery behavior classification standard system is constructed, the delivery behavior feature parameters in the multi-dimensional delivery behavior analysis data are quantitatively calculated, and the delivery behavior feature parameters are classified by fuzzy logic judgment rules. Based on the behavior classification results, corresponding customized early warning instruction parameters are matched. Structured warning instructions are generated based on the customized warning instruction parameters and distributed to visual prompting devices, sound warning devices, and behavior recording devices.

[0007] Preferably, the step of deploying visual capture devices in a differentiated manner around the household waste recycling equipment, delineating the recycling equipment body area, the normal waste disposal area, and the peripheral disposal monitoring area and establishing a spatial mapping relationship, configuring scene-adaptive dynamic video analysis parameters, and outputting a standardized visual monitoring system with full-scene coverage without blind spots and parameter self-adaptation capabilities, specifically includes: The visual capture devices are deployed in a differentiated manner around the household waste recycling equipment. Specifically, a single camera or a multi-camera array configuration is selected based on the complexity of the disposal scene around the household waste recycling equipment, and a verification report on the signal and coverage of the device deployment is output. Based on the coverage verification report, and combined with the mapping relationship between the image coordinate system of the visual capture device and the physical space where the household waste recycling equipment is located, the physical boundaries, coordinate ranges and spatial relationships of the recycling equipment body area, the normal waste disposal area and the peripheral disposal monitoring area are determined, and three-dimensional monitoring area model data containing area location, size and hierarchical relationship are output. Based on the regional characteristics of the three-dimensional monitoring area model data, the sensitivity of personnel detection, behavior analysis time window, ambient light adaptation coefficient and early warning triggering basic threshold set are set, and the dynamic video analysis parameters adapted to the scene are output, resulting in a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. Based on the detection accuracy, false alarm rate, and environmental change data fed back during the operation of the standardized visual monitoring system, the dynamic video analysis parameters are adjusted in real time.

[0008] Preferably, the personnel target detection process for the video stream of household waste disposal behavior includes: The video stream of the household waste disposal behavior is subjected to noise filtering, illumination equalization and contrast enhancement processing to output pre-processed video frames of the disposal behavior. A deep learning model combined with a region proposal mechanism is used to extract multi-scale image features from the preprocessed video frames of the delivery behavior, generate candidate regions for personnel targets, and calculate the region detection confidence of the candidate regions for personnel targets. The confidence threshold is adaptively adjusted based on the complexity of the delivery scenario to filter candidate areas of the personnel target that have a detection confidence level higher than the confidence threshold, thereby generating the personnel target area; Based on the bounding box coordinates of the personnel target in the personnel target area, the personnel target is located in the physical space where the municipal solid waste recycling equipment is located, and the personnel physical coordinate data of the personnel target is output.

[0009] Preferably, the continuous tracking process of the motion trajectory of the video stream of the household waste disposal behavior includes: Based on the physical coordinate data of the personnel and combined with the motion trajectory continuity constraint of the video stream of the household waste disposal behavior, a unique identifier is assigned to the personnel target, a binding relationship between the personnel target and the corresponding unique identifier is established, and a set of personnel target-identifier relationships is output. Extract the appearance features, motion features, and position features of the personnel target to generate initial motion trajectory data; A personnel target tracking model is constructed. In the continuous video frames of the video stream of the household waste disposal behavior, the personnel target is matched across frames. The appearance features, motion features and position features of the personnel target in the initial motion trajectory data are updated in real time, and the updated motion trajectory data is output. The occlusion determination conditions and the retention time of the disconnection status are set. In the updated motion trajectory data, combined with the personnel target-identifier relationship set, the trajectory prediction mechanism is used to maintain the corresponding unique identifier for the occluded personnel target and continuously track it. The corresponding unique identifier for the temporarily disconnected personnel target is temporarily stored until the temporarily disconnected personnel target reappears and the tracking is resumed. The personnel motion trajectory data of the personnel target is then output.

[0010] Preferably, the process of identifying waste object features in the video stream of the household waste disposal behavior includes: Based on the motion characteristics of the target person in the motion trajectory data, the hand region of the target person is located by combining skin color feature and contour detection technology and a hand tracking window is established. The position change and shape characteristics of the hand region are monitored in real time, and the real-time coordinate data of the hand region and the hand motion state data are output. Based on the real-time coordinate data of the hand, the association between the person's hand and the garbage object is identified, and the shape features, texture features, and color features of the garbage object in the hand area are extracted and summarized to generate garbage object features; Based on the hand movement data and the characteristics of the garbage object, the relative movement speed, contour separation degree and shape change between the hand area and the garbage object are analyzed to determine whether the garbage object has separated from the hand of the person target. If the garbage object has separated from the hand of the person target, the corresponding separation time and separation position are extracted and the separation feature information is generated. For the waste object that separates from the hand of the person being tracked, an independent waste tracking mechanism is established. Combining the separation feature information, the movement path, speed, and trend of the waste object are recorded in real time and aggregated to generate waste object movement data.

[0011] Preferably, the analysis process of the disposal behavior pattern of the video stream of household waste disposal behavior includes: Based on the three-dimensional monitoring area model data and the physical coordinate data of the personnel, the actual physical distance between the personnel target and the domestic waste recycling equipment is calculated, the area where the personnel target is located is determined to be the normal waste disposal area or the peripheral disposal monitoring area, and the personnel area attribution determination result is output. The human key point detection technology is used to extract 21 core limb key points of the target person. Combined with the result of the person’s regional affiliation determination, the angle change rate and distance change rate between the core limb key points are analyzed to identify typical person’s delivery posture and output the delivery posture feature set. The motion path, speed, and trend of the garbage object are analyzed to determine whether the trajectory pattern of the garbage object belongs to the normal placement mode or the throwing mode, and the garbage trajectory pattern determination result is output. By integrating the actual physical distance, the disposal posture feature set, and the garbage trajectory pattern determination results, four types of disposal behaviors are comprehensively identified: normal disposal, close-range disposal, long-range disposal, and random littering. The disposal behavior determination results and the corresponding disposal behavior determination confidence levels are output.

[0012] Preferably, the process for evaluating the effectiveness of the video stream of household waste disposal behavior includes: Based on the motion data of the garbage object, the predicted landing point of the garbage object and the corresponding reliability of the predicted landing point are predicted. The movement of the garbage object is tracked in real time by a visual monitoring system. The actual landing point of the garbage object is located based on the predicted landing point. The consistency between the actual landing point and the predicted landing point is compared, and the landing point matching result is output. Based on the actual landing point and the range of the recycling equipment body area in the three-dimensional monitoring area model data, a distance judgment threshold is set. If the minimum distance between the actual landing point and the boundary of the recycling equipment body area is less than the distance judgment threshold, the garbage object is judged to be successfully disposed of; otherwise, it is considered a failure, and the garbage disposal judgment result is output. Based on the landing point matching results, the clarity of the visual monitoring image, and the continuity of the garbage object's movement, the reliability of the garbage disposal determination result is quantitatively evaluated, and the confidence level of the garbage disposal determination is output.

[0013] Preferably, the process of constructing the three-dimensional delivery behavior grading standard system includes: The criteria for judging normal placement behavior and three levels of uncivilized placement behavior are set. Level 1 uncivilized placement behavior includes short-distance throwing and mild long-distance throwing; Level 2 uncivilized placement behavior includes severe long-distance throwing and mild placement failure; and Level 3 uncivilized placement behavior includes severe placement failure and malicious throwing. The three-dimensional placement behavior classification standard system including behavior distance, placement method and placement result is constructed.

[0014] A computer vision-based early warning system for household waste disposal behavior, the system comprising: The monitoring system output module is used to deploy visual capture devices in a differentiated manner around the household waste recycling equipment, delineate the recycling equipment body area, normal waste disposal area and peripheral disposal monitoring area and establish spatial mapping relationship, configure scene-adaptive dynamic video analysis parameters, and output a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. The waste disposal behavior analysis module is used to capture video streams of waste disposal behavior in real time through the standardized visual monitoring system. The module performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, waste disposal behavior pattern analysis, and waste disposal result effectiveness evaluation on the video streams of waste disposal behavior in sequence. It outputs multi-dimensional waste disposal behavior analysis data that includes personnel spatiotemporal information, object motion trajectory, and waste disposal effectiveness judgment results. The early warning instruction matching module is used to construct a three-dimensional delivery behavior classification standard system, quantitatively calculate the delivery behavior feature parameters in the multi-dimensional delivery behavior analysis data, classify the delivery behavior feature parameters according to fuzzy logic judgment rules, and match the corresponding customized early warning instruction parameters based on the behavior classification results. The early warning instruction distribution module is used to generate structured early warning instructions based on the customized early warning instruction parameters, and distribute them to visual prompting devices, sound warning devices, and behavior recording devices.

[0015] Compared with related technologies, the computer vision-based early warning method and system for household waste disposal behavior provided by this invention has the following beneficial effects: This invention utilizes differentiated visual capture devices deployed around household waste recycling equipment to delineate the equipment itself, the normal waste disposal area, and the surrounding monitoring area, establishing a spatial mapping relationship. It configures scene-adaptive dynamic video analysis parameters and outputs a standardized visual monitoring system with full-scene coverage and parameter self-adaptation capabilities. The system captures real-time video streams of household waste disposal behavior and sequentially performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, disposal behavior pattern analysis, and disposal result effectiveness evaluation on these video streams. The output includes multi-dimensional disposal data encompassing personnel spatiotemporal information, object motion trajectories, and disposal effectiveness assessment results. To analyze the data, a three-dimensional classification standard system for waste disposal behavior is constructed. The characteristic parameters of waste disposal behavior in the multi-dimensional waste disposal behavior analysis data are quantitatively calculated, and the behavior characteristic parameters are classified according to fuzzy logic judgment rules. Based on the behavior classification results, corresponding customized early warning instruction parameters are matched. Structured early warning instructions are generated based on the customized early warning instruction parameters and distributed to visual prompting devices, sound warning devices, and waste disposal behavior recording devices. By deploying visual capture devices around waste recycling equipment and combining computer vision and behavior analysis technologies, real-time monitoring, intelligent identification, and classified early warning of waste disposal behavior can be achieved, effectively regulating public waste disposal behavior and improving the level of urban environmental sanitation management.

[0016] This invention achieves seamless coverage of the entire waste disposal process by deploying visual capture devices and modeling 3D monitoring areas through differentiated deployment and parameter self-adaptation optimization. This effectively solves the problems of poor environmental adaptability and incomplete coverage inherent in traditional monitoring methods. It maintains stable monitoring performance regardless of changes in lighting, pedestrian flow, or complex terrain, improving detection accuracy and reducing false alarm rates. Furthermore, by integrating multi-feature extraction and cross-frame matching technologies, this invention accurately identifies four types of behaviors: normal disposal, throwing, and indiscriminate littering. It overcomes the problems of fuzzy judgment and high false alarm rates in traditional methods by assessing the confidence level of behavior. Finally, by constructing a 3D classification standard system for disposal behavior and using fuzzy logic to achieve precise behavior classification, this invention matches customized early warning strategies, avoiding the inefficiency of traditional single-warning methods. Level 1 behaviors receive a light prompt, while level 3 behaviors receive a stronger warning, effectively curbing uncivilized disposal behavior while also considering user experience and improving intervention response efficiency. This invention combines multi-device collaborative early warning with environmental adaptive optimization to ensure low-latency transmission and accurate execution of early warning commands, improve the effective reach of early warnings in special scenarios such as strong light and high noise, and achieve closed-loop management through behavior recording, providing data traceability support for waste classification management, significantly reducing the cost of manual inspections, and helping to improve the level of refined and intelligent waste classification management. Attached Figure Description

[0017] Figure 1A flowchart illustrating a computer vision-based early warning method for household waste disposal behavior provided in an embodiment of the present invention; Figure 2 A system block diagram of a computer vision-based early warning system for household waste disposal behavior is provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0019] like Figure 1 The diagram shown is a flowchart of a computer vision-based early warning method for household waste disposal behavior provided in an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S4 are detailed as follows: S1. Differentiatedly deploy visual capture devices around the household waste recycling equipment, delineate the recycling equipment body area, normal waste disposal area and peripheral disposal monitoring area and establish spatial mapping relationship, configure scene-adaptive dynamic video analysis parameters, and output a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. S2, the standardized visual monitoring system captures video streams of household waste disposal behavior in real time, and sequentially performs personnel target detection, continuous tracking of motion trajectory, identification of waste object features, analysis of disposal behavior patterns, and evaluation of the effectiveness of disposal results on the video streams of household waste disposal behavior, and outputs multi-dimensional disposal behavior analysis data containing personnel spatiotemporal information, object motion trajectory, and disposal effectiveness judgment results. S3, construct a three-dimensional delivery behavior classification standard system, quantitatively calculate the delivery behavior feature parameters in the multi-dimensional delivery behavior analysis data, classify the delivery behavior feature parameters according to the fuzzy logic judgment rules, and match the corresponding customized early warning instruction parameters according to the behavior classification results. S4. Generate a structured early warning instruction based on the customized early warning instruction parameters, and distribute it to the visual prompting device, the sound warning device, and the behavior recording device.

[0020] Understandably, visual capture devices are deployed in a differentiated manner based on the complexity of the surrounding environment of the waste recycling equipment. Multi-camera arrays are used in densely populated areas with complex terrain, while single-camera configurations are used in simpler scenarios. After deployment, a deployment signal and coverage verification report is output to ensure comprehensive monitoring coverage. Then, based on the coverage verification report, a precise mapping relationship between the visual capture device's image coordinate system and physical space is established. This clarifies the physical boundaries, coordinate ranges, and spatial relationships of the recycling equipment's main area, the normal waste disposal area, and the surrounding monitoring area, generating a 3D monitoring area model data that includes the area's location, size, and hierarchical relationships. Subsequently, combined with the regional characteristics of the 3D monitoring area model, dynamic video analysis parameters are configured to adapt to the scenario, such as personnel detection sensitivity, behavior analysis time window, ambient light adaptation coefficient, and early warning trigger threshold set. Finally, based on the detection accuracy, false alarm rate, and environmental change data fed back during system operation, the dynamic video analysis parameters are dynamically adjusted in real time. The final output is a standardized visual monitoring system with full-scene, blind-spot-free coverage and parameter self-adaptation characteristics, ensuring the stability of monitoring performance in complex environments.

[0021] Furthermore, a standardized visual monitoring system is used to capture real-time video streams of household waste disposal behavior. The video streams undergo noise filtering, illumination equalization, and contrast enhancement preprocessing to improve image quality before outputting preprocessed disposal behavior video frames. A deep learning model combined with a region proposal mechanism is used to detect personnel targets. Personnel target areas are dynamically generated based on confidence levels, and their physical coordinates are located. Based on personnel physical coordinate data and trajectory continuity constraints, unique identifiers are assigned to personnel, and a tracking model is built to achieve continuous tracking of movement trajectories and output personnel movement trajectory data. Skin color features and contour detection technology are used to locate the personnel's hand area, extract waste object features, and determine the separation action between the object and the hand, establishing an independent waste tracking mechanism to output waste object movement data. The actual physical distance between personnel and equipment, disposal posture feature sets, and waste trajectory pattern determination results are integrated to identify four types of disposal behavior. The waste object movement data is combined to predict and detect the disposal landing point, determine the success of the disposal, and quantify the confidence level of the results. Finally, the above analysis results are integrated to output multi-dimensional disposal behavior analysis data containing personnel spatiotemporal information, object movement trajectories, and disposal effectiveness determination results.

[0022] Secondly, a three-dimensional classification standard system for delivery behavior is constructed, encompassing three dimensions: distance, delivery method, and delivery result. The criteria for distinguishing between normal delivery behavior and three levels of uncivilized delivery behavior are clearly defined: Level 1 uncivilized behavior includes close-range throwing and mild long-range throwing; Level 2 includes severe long-range throwing and mild delivery failure; and Level 3 includes severe delivery failure and malicious throwing. Based on multi-dimensional delivery behavior analysis data, core behavioral characteristic parameters such as distance characteristics, intensity characteristics, and result impact characteristics of delivery behavior are quantitatively calculated. Subsequently, fuzzy logic judgment rules are used to analyze the quantified delivery behavior characteristic parameters, completing the behavior classification judgment and outputting the behavior level and judgment confidence level. Finally, based on different behavior classification results, corresponding warning intensity and execution methods are matched to generate customized warning instruction parameters containing information such as warning type, response intensity, and execution logic, ensuring the targeted and reasonable nature of warning intervention.

[0023] Finally, based on customized warning command parameters, structured warning commands are generated in a unified data format, including warning type, warning level, list of executing devices, parameter configuration, and timestamp, ensuring the universality and executability of the commands. An efficient transmission protocol is used to distribute the structured warning commands to visual prompting devices, audio warning devices, and behavior recording devices. A data verification mechanism verifies the integrity and accuracy of the command transmission, preventing command loss or tampering, and outputs the command transmission results and verification status. All executing devices respond collaboratively: visual prompting devices display customized civilized waste disposal information, audio warning devices play tiered voice warnings, and behavior recording devices store behavior videos and analysis data, enabling real-time intervention and data retention for uncivilized waste disposal behavior.

[0024] The aforementioned method involves deploying visual capture devices in a differentiated manner around the household waste recycling equipment, delineating the recycling equipment itself area, the normal waste disposal area, and the surrounding disposal monitoring area, establishing a spatial mapping relationship, configuring scene-adaptive dynamic video analysis parameters, and outputting a standardized visual monitoring system with full-scene coverage without blind spots and parameter self-adaptation capabilities. Specifically, this includes: The visual capture devices are deployed in a differentiated manner around the household waste recycling equipment. Specifically, a single camera or a multi-camera array configuration is selected based on the complexity of the disposal scene around the household waste recycling equipment, and a verification report on the signal and coverage of the device deployment is output. Based on the coverage verification report, and combined with the mapping relationship between the image coordinate system of the visual capture device and the physical space where the household waste recycling equipment is located, the physical boundaries, coordinate ranges and spatial relationships of the recycling equipment body area, the normal waste disposal area and the peripheral disposal monitoring area are determined, and three-dimensional monitoring area model data containing area location, size and hierarchical relationship are output. Based on the regional characteristics of the three-dimensional monitoring area model data, the sensitivity of personnel detection, behavior analysis time window, ambient light adaptation coefficient and early warning triggering basic threshold set are set, and the dynamic video analysis parameters adapted to the scene are output, resulting in a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. Based on the detection accuracy, false alarm rate, and environmental change data fed back during the operation of the standardized visual monitoring system, the dynamic video analysis parameters are adjusted in real time.

[0025] First, visual capture devices are deployed in a differentiated manner. Depending on the complexity of the surrounding waste collection environment, either a single camera or a multi-camera array configuration is selected to ensure coverage of the core monitoring area. After deployment, a deployment confirmation signal and a coverage verification report are output.

[0026] Based on the coverage verification report, a precise mapping relationship between the image coordinate system and physical space is established, clarifying the physical boundaries, coordinate ranges, and spatial relationships of the recycling equipment body area, the normal waste disposal area, and the peripheral disposal monitoring area. This generates three-dimensional monitoring area model data that includes the location, size, and hierarchical relationship of the area, thus realizing the digital definition of the monitoring area.

[0027] By combining the regional characteristics of the 3D monitoring area model data, we can specifically set the sensitivity of personnel detection, the time window for behavior analysis, the environmental light adaptation coefficient, and the basic threshold set for early warning triggering, forming dynamic video analysis parameters that are adapted to the scene, and then outputting a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation characteristics.

[0028] Based on the detection accuracy, false alarm rate and environmental change data fed back during the operation of the standardized visual monitoring system, the dynamic video analysis parameters are adjusted in real time to continuously optimize the system's adaptability to different environments and pedestrian conditions, ensuring the stability and reliability of monitoring performance.

[0029] The personnel target detection process in the video stream of household waste disposal behavior includes: The video stream of the household waste disposal behavior is subjected to noise filtering, illumination equalization and contrast enhancement processing to output pre-processed video frames of the disposal behavior. A deep learning model combined with a region proposal mechanism is used to extract multi-scale image features from the preprocessed video frames of the delivery behavior, generate candidate regions for personnel targets, and calculate the region detection confidence of the candidate regions for personnel targets. The confidence threshold is adaptively adjusted based on the complexity of the delivery scenario to filter candidate areas of the personnel target that have a detection confidence level higher than the confidence threshold, thereby generating the personnel target area; Based on the bounding box coordinates of the personnel target in the personnel target area, the personnel target is located in the physical space where the municipal solid waste recycling equipment is located, and the personnel physical coordinate data of the personnel target is output.

[0030] It should be noted that, to address potential issues such as uneven lighting, environmental noise interference, and insufficient image contrast in household waste disposal scenarios, a combined preprocessing operation is performed on the captured video stream of household waste disposal behavior. Noise filtering technology is used to remove random interference signals from the image, an illumination equalization algorithm is employed to correct for differences in lighting conditions at different times and in different environments, and contrast enhancement technology is combined to improve the distinction between the person and the background environment. The final output is a preprocessed video frame of the disposal behavior with high clarity, complete detail retention, and strong anti-interference capabilities.

[0031] Based on preprocessed video frames of the delivery behavior, a technical solution combining a deep learning model and a region proposal mechanism is adopted. The deep learning model possesses powerful feature learning capabilities, which, combined with the region proposal mechanism, can accurately locate potential human target regions in the image. Simultaneously, through a multi-scale feature extraction strategy, it adapts to human targets at different distances and in different postures, such as close-range delivery personnel and distant observers, comprehensively capturing both shallow texture features and deep semantic features of the targets. After feature extraction, multiple candidate regions for human targets are generated, and the region detection confidence of each candidate region is obtained through quantitative calculation. This confidence level characterizes the reliability of the candidate region as a real human target.

[0032] Due to varying complexity of deployment scenarios, such as densely populated areas, open areas, and low-light areas at night, the confidence threshold can be adaptively adjusted based on the complexity index of the current scenario. For complex scenarios, such as densely populated areas and cluttered backgrounds, the threshold can be appropriately increased to reduce false detections; for simple scenarios, the threshold can be reasonably decreased to avoid missed detections. By selecting candidate areas for personnel targets with a detection confidence level higher than this dynamic threshold, and eliminating background interference and falsely detected non-personnel areas, such as clutter and equipment shadows, a precise personnel target area is ultimately generated, ensuring the accuracy of subsequent positioning.

[0033] Based on the bounding box coordinates of the personnel target area output, and combined with the preset calibration parameters of the physical space where the waste recycling equipment is located, a mapping relationship between the image coordinate system and the physical space coordinate system is established. Through spatial coordinate transformation technology, the bounding box coordinates in the image are converted into the specific location information of the personnel in the actual physical space, and finally, the physical coordinate data of the personnel containing their real-time location is output, realizing the accurate positioning of the personnel target in the disposal scenario.

[0034] The continuous tracking process of the motion trajectory of the video stream of household waste disposal behavior includes: Based on the physical coordinate data of the personnel and combined with the motion trajectory continuity constraint of the video stream of the household waste disposal behavior, a unique identifier is assigned to the personnel target, a binding relationship between the personnel target and the corresponding unique identifier is established, and a set of personnel target-identifier relationships is output. Extract the appearance features, motion features, and position features of the personnel target to generate initial motion trajectory data; A personnel target tracking model is constructed. In the continuous video frames of the video stream of the household waste disposal behavior, the personnel target is matched across frames. The appearance features, motion features and position features of the personnel target in the initial motion trajectory data are updated in real time, and the updated motion trajectory data is output. The occlusion determination conditions and the retention time of the disconnection status are set. In the updated motion trajectory data, combined with the personnel target-identifier relationship set, the trajectory prediction mechanism is used to maintain the corresponding unique identifier for the occluded personnel target and continuously track it. The corresponding unique identifier for the temporarily disconnected personnel target is temporarily stored until the temporarily disconnected personnel target reappears and the tracking is resumed. The personnel motion trajectory data of the personnel target is then output.

[0035] In practical applications, physical coordinate data of individuals is used as the core input. Combined with the continuity constraint of the movement trajectory of individuals in the video stream of household waste disposal behavior (meaning the movement trajectory will not jump or interrupt without reason), a unique and persistent identifier is assigned to each detected individual. By establishing a fixed binding relationship between individual targets and their corresponding identifiers, it is ensured that each individual is uniquely identified throughout the monitoring process, avoiding confusion between different individual targets. The final output is a set of individual target-identifier relationships containing the correspondence between individual targets and their identifiers, providing an identification basis for subsequent trajectory tracking.

[0036] Based on the established set of personnel target-identifier relationships, for each personnel target bound to a unique identifier, the system extracts its appearance features, such as clothing color and body shape, and its movement features, such as movement speed and direction, as well as its location features, i.e., its real-time spatial location based on the personnel's physical coordinate data. These three types of features are then fused and summarized to form initial motion trajectory data that comprehensively characterizes the personnel's movement state, providing basic data support for subsequent trajectory updates and tracking, and realizing the association between identifiers and personnel movement features.

[0037] Using initial motion trajectory data as input, a personnel target tracking model specifically adapted to waste disposal scenarios is constructed. This model achieves cross-frame matching of personnel targets through feature similarity comparison within continuous video frames of a video stream of waste disposal behavior, capturing real-time changes in the personnel's motion state. Based on the matching results, the model dynamically updates the appearance, motion, and positional features in the initial motion trajectory data, correcting trajectory deviations during personnel movement to ensure consistency between the trajectory data and the actual movement state of the personnel. Finally, updated motion trajectory data is output, maintaining the continuity and accuracy of the trajectory.

[0038] Using updated motion trajectory data and a set of personnel target-identifier relationships as dual inputs, the system pre-sets occlusion judgment conditions, such as a threshold for the number of frames in which the target is occluded by other objects, and a retention period for the out-of-connection state, such as the maximum retention time after the target temporarily leaves the monitoring range. For occluded personnel targets, a trajectory prediction mechanism is used to continuously calculate their position based on their historical motion trajectory patterns, maintaining the corresponding unique identifier and ensuring that tracking can be quickly resumed after the occlusion is removed. For temporarily out-of-connection personnel targets, the system temporarily stores their unique identifier and historical trajectory data. After the target re-enters the monitoring range, tracking is quickly resumed through identifier matching to avoid trajectory breaks. Finally, the system outputs complete and unbroken personnel motion trajectory data, providing continuous spatiotemporal motion data for subsequent waste object feature identification and disposal behavior pattern analysis.

[0039] The process of identifying the garbage object features in the video stream of household waste disposal behavior includes: Based on the motion characteristics of the target person in the motion trajectory data, the hand region of the target person is located by combining skin color feature and contour detection technology and a hand tracking window is established. The position change and shape characteristics of the hand region are monitored in real time, and the real-time coordinate data of the hand region and the hand motion state data are output. Based on the real-time coordinate data of the hand, the association between the person's hand and the garbage object is identified, and the shape features, texture features, and color features of the garbage object in the hand area are extracted and summarized to generate garbage object features; Based on the hand movement data and the characteristics of the garbage object, the relative movement speed, contour separation degree and shape change between the hand area and the garbage object are analyzed to determine whether the garbage object has separated from the hand of the person target. If the garbage object has separated from the hand of the person target, the corresponding separation time and separation position are extracted and the separation feature information is generated. For the waste object that separates from the hand of the person being tracked, an independent waste tracking mechanism is established. Combining the separation feature information, the movement path, speed, and trend of the waste object are recorded in real time and aggregated to generate waste object movement data.

[0040] Using the motion features of human targets extracted from human movement trajectory data as the core input, a fusion scheme combining skin color feature detection and contour detection technology is adopted. This fully utilizes the unique spectral characteristics of skin color in images and the morphological features of hand contours to achieve accurate locking of the human hand region against complex backgrounds. A dedicated hand tracking window is established for the locked hand region, monitoring the spatial position changes and morphological feature evolution of the hand within the window in real time, such as clenching a fist or extending an arm. The final output includes real-time hand coordinate data containing information such as real-time spatial coordinates, direction of movement, and movement speed, as well as hand motion state data representing the hand's action state, providing accurate spatial anchor points for subsequent garbage object association and identification.

[0041] Based on real-time hand coordinate data, the spatial range of the hand area is located. Pixel correlation analysis technology is used to identify the attachment relationship between the hand and objects within the area, accurately filtering out waste objects physically associated with the hand. For each waste object, a multi-feature fusion extraction strategy is employed. The system collects its shape features, such as geometric contours and aspect ratios; texture features, such as surface texture density and texture distribution patterns; and color features, such as dominant hue and color distribution ratio. After standardization of these three types of features, they are summarized and integrated to generate waste object features that uniquely characterize the attributes of the waste object, providing feature support for subsequent separation actions.

[0042] A separation action determination model is constructed using hand movement data and debris object features as dual inputs. By analyzing the relative motion velocity between the hand region and the debris object (i.e., the difference in their motion rates), contour separation degree (i.e., the overlap ratio and morphological changes of the pixel contours of the hand region and the debris object), and the relative position and pose evolution of the hand region and the debris object, multi-dimensional determination criteria are established to accurately identify whether a separation action has occurred between the debris object and the hand. When a separation action is determined, the separation time (specific time node) and separation location (spatial location) are simultaneously extracted. These two types of information are then combined to generate separation feature information, providing initial triggering conditions and basic data for subsequent independent tracking of the debris object.

[0043] For waste objects that have been determined to have separated, an independent tracking mechanism is established to prevent them from being confused with personnel targets or background objects. Using separation feature information as the initial tracking benchmark, and combining the waste object's own physical attributes, such as shape and color features, the waste object is dynamically tracked in continuous video frames. Its motion path, i.e., spatial coordinate sequence, motion speed (including instantaneous and average speed), and motion trend (including trajectory curvature and changes in motion direction), are recorded in real time. The above tracking data is then structured and integrated to ultimately output waste object motion data that comprehensively reflects the motion state of the waste object after separation.

[0044] The analysis process of the behavior pattern of the video stream of household waste disposal includes: Based on the three-dimensional monitoring area model data and the physical coordinate data of the personnel, the actual physical distance between the personnel target and the domestic waste recycling equipment is calculated, the area where the personnel target is located is determined to be the normal waste disposal area or the peripheral disposal monitoring area, and the personnel area attribution determination result is output. The human key point detection technology is used to extract 21 core limb key points of the target person. Combined with the result of the person’s regional affiliation determination, the angle change rate and distance change rate between the core limb key points are analyzed to identify typical person’s delivery posture and output the delivery posture feature set. The motion path, speed, and trend of the garbage object are analyzed to determine whether the trajectory pattern of the garbage object belongs to the normal placement mode or the throwing mode, and the garbage trajectory pattern determination result is output. By integrating the actual physical distance, the disposal posture feature set, and the garbage trajectory pattern determination results, four types of disposal behaviors are comprehensively identified: normal disposal, close-range disposal, long-range disposal, and random littering. The disposal behavior determination results and the corresponding disposal behavior determination confidence levels are output.

[0045] Using 3D monitoring area model data and personnel physical coordinate data as dual inputs, the actual physical distance between the personnel target and the municipal solid waste recycling equipment is obtained through spatial distance calculation technology. Combined with preset area division standards, the system determines whether the area where the personnel target is located is a normal waste disposal area or an outer disposal monitoring area, and finally outputs the personnel area attribution result, providing spatial scene basis for subsequent posture analysis and behavior determination.

[0046] Based on the personnel location determination results, human key point detection technology is used to accurately extract 21 core limb key points of the personnel target, covering the head, torso, and key joints of the limbs. By analyzing the rate of change of angles between key points, such as the change of arm extension angle and the rate of change of distance, such as the change of shoulder-elbow distance, typical throwing posture features such as bending over to throw, standing to throw, and long-distance throwing are captured. After standardizing and integrating the feature parameters, a throwing posture feature set containing posture type, movement range, and movement rhythm is output, providing human movement support for behavioral pattern differentiation.

[0047] Using the motion data of garbage objects as input, a trajectory pattern analysis model is constructed. By analyzing the motion path shape, peak velocity, and motion trend stability of garbage objects, a binary judgment criterion is established to accurately distinguish between normal placement and throwing patterns, outputting the garbage trajectory pattern judgment result and clarifying the motion characteristics of garbage objects.

[0048] A multi-feature fusion judgment model is constructed by integrating actual physical distance, disposal posture feature set, and waste trajectory pattern judgment results. If a person is in the normal disposal area, in a bent-over disposal posture, and the waste is placed normally, it is judged as normal disposal; if a person is in the normal disposal area but the waste is thrown, it is judged as close-range disposal; if a person is in the outer monitoring area, in a throwing posture, and the waste is thrown at high speed, it is judged as long-distance disposal; if the waste trajectory has no fixed direction, the disposal posture has no clear direction, and it does not fall into the equipment area, it is judged as random disposal. The final output includes disposal behavior judgment results for the four categories of normal disposal, close-range disposal, long-distance disposal, and random disposal, as well as the disposal behavior judgment confidence level, which characterizes the reliability of the judgment.

[0049] The process for evaluating the effectiveness of the video stream of household waste disposal behavior includes: Based on the motion data of the garbage object, the predicted landing point of the garbage object and the corresponding reliability of the predicted landing point are predicted. The movement of the garbage object is tracked in real time by a visual monitoring system. The actual landing point of the garbage object is located based on the predicted landing point. The consistency between the actual landing point and the predicted landing point is compared, and the landing point matching result is output. Based on the actual landing point and the range of the recycling equipment body area in the three-dimensional monitoring area model data, a distance judgment threshold is set. If the minimum distance between the actual landing point and the boundary of the recycling equipment body area is less than the distance judgment threshold, the garbage object is judged to be successfully disposed of; otherwise, it is considered a failure, and the garbage disposal judgment result is output. Based on the landing point matching results, the clarity of the visual monitoring image, and the continuity of the garbage object's movement, the reliability of the garbage disposal determination result is quantitatively evaluated, and the confidence level of the garbage disposal determination is output.

[0050] Using the motion data of litter objects as the core input, and combining the physical laws of object motion with the characteristics of the scene environment, a landing point prediction model is constructed. By analyzing the motion state parameters of the litter objects, the final landing area after they leave human control is predicted, generating the spatial coordinate information of the predicted landing point. Simultaneously, based on the completeness of the litter object motion data, the stability of the motion trajectory, and environmental interference factors such as wind and obstacles, the reliability of the predicted landing point is quantitatively calibrated, outputting the predicted landing point and its corresponding credibility, providing accurate guidance for subsequent actual landing point location.

[0051] Using the predicted landing point as the positioning benchmark, a standardized visual monitoring system tracks the movement of waste objects in real time, accurately capturing the actual spatial position of the object after it stops moving, and outputting the actual landing point coordinates. The actual landing point is compared with the predicted landing point in space, and key indicators such as coordinate deviation and regional overlap are analyzed to determine the degree of consistency between the two. Finally, the landing point matching result is output to verify the accuracy of the landing point prediction.

[0052] Using both the actual landing point coordinates and the area of ​​the recycling equipment in the 3D monitoring area model as dual inputs, a scientifically reasonable distance judgment threshold is set. By calculating the minimum spatial distance between the actual landing point and the boundary of the recycling equipment area, a judgment rule is established: if this minimum distance is less than the set distance judgment threshold, it indicates that the waste object has successfully landed in the equipment area, and the disposal is judged as successful; if it is greater than or equal to the threshold, it indicates that the disposal has not met expectations, and the disposal is judged as a failure. The final output is a clear waste disposal judgment result, providing a core result indicator for behavior classification.

[0053] A multi-dimensional confidence assessment model is constructed using the landing point matching result, the clarity of the visual monitoring image, and the continuity of the waste object's movement as inputs. Higher landing point matching results, better monitoring image clarity, and more continuous object movement trajectories indicate stronger reliability of the waste disposal judgment. By standardizing and weighting the three types of indicators, the reliability of the waste disposal judgment is quantitatively assessed, and a waste disposal judgment confidence score is output. This confidence score provides a reliable reference for subsequent behavior grading, avoiding misgrading due to data bias and ensuring the accuracy of early warning decisions.

[0054] The process of constructing the three-dimensional delivery behavior grading standard system includes: The criteria for judging normal placement behavior and three levels of uncivilized placement behavior are set. Level 1 uncivilized placement behavior includes short-distance throwing and mild long-distance throwing; Level 2 uncivilized placement behavior includes severe long-distance throwing and mild placement failure; and Level 3 uncivilized placement behavior includes severe placement failure and malicious throwing. The three-dimensional placement behavior classification standard system including behavior distance, placement method and placement result is constructed.

[0055] It is important to clarify that, firstly, the core dimensions of the behavior classification are clearly defined, using distance, delivery method, and delivery result as three-dimensional evaluation benchmarks. These three factors are interconnected and judged collaboratively to ensure the comprehensiveness and accuracy of the classification. Secondly, classification judgment conditions are set to distinguish between normal delivery behavior and three levels of uncivilized delivery behavior: Level 1 uncivilized delivery behavior includes close-range throwing and mild long-range throwing, which have a relatively small impact; Level 2 includes severe long-range throwing and mild delivery failure, which significantly increase the interference of the behavior; Level 3 includes severe delivery failure and malicious throwing, which are uncivilized behaviors with a serious impact. Finally, the three-dimensional dimensions and classification conditions are integrated to form a complete classification standard system. By clarifying the quantitative boundaries and judgment rules for distance, method, and result of each level of behavior, the accurate classification of different delivery behaviors can be achieved.

[0056] like Figure 2 The diagram shown is a system block diagram of a computer vision-based early warning system for household waste disposal behavior provided in an embodiment of the present invention. The system includes: The monitoring system output module is used to deploy visual capture devices in a differentiated manner around the household waste recycling equipment, delineate the recycling equipment body area, normal waste disposal area and peripheral disposal monitoring area and establish spatial mapping relationship, configure scene-adaptive dynamic video analysis parameters, and output a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. The waste disposal behavior analysis module is used to capture video streams of waste disposal behavior in real time through the standardized visual monitoring system. The module performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, waste disposal behavior pattern analysis, and waste disposal result effectiveness evaluation on the video streams of waste disposal behavior in sequence. It outputs multi-dimensional waste disposal behavior analysis data that includes personnel spatiotemporal information, object motion trajectory, and waste disposal effectiveness judgment results. The early warning instruction matching module is used to construct a three-dimensional delivery behavior classification standard system, quantitatively calculate the delivery behavior feature parameters in the multi-dimensional delivery behavior analysis data, classify the delivery behavior feature parameters according to fuzzy logic judgment rules, and match the corresponding customized early warning instruction parameters based on the behavior classification results. The early warning instruction distribution module is used to generate structured early warning instructions based on the customized early warning instruction parameters, and distribute them to visual prompting devices, sound warning devices, and behavior recording devices.

[0057] Figure 2 The apparatus of the illustrated embodiment can be used to perform corresponding actions. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0058] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the steps of a computer vision-based early warning method for household waste disposal behavior as described above.

[0059] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device according to an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32, and a computer program; wherein... The memory 32 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0060] Processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0061] Alternatively, the memory 32 can be either standalone or integrated with the processor 31.

[0062] When the memory 32 is a device independent of the processor 31, the device may further include: Bus 33 is used to connect the memory 32 and the processor 31.

[0063] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of a computer vision-based early warning method for household waste disposal behavior as described above.

[0064] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0065] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0066] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0067] Through the above embodiments, this invention, through a computer vision-based method and system for early warning of household waste disposal behavior, deploys visual capture devices differentiatedly around household waste recycling equipment, delineates the recycling equipment body area, normal waste disposal area, and peripheral disposal monitoring area, establishes a spatial mapping relationship, configures scene-adaptive dynamic video analysis parameters, and outputs a standardized visual monitoring system with full-scene coverage without blind spots and parameter self-adaptation capabilities. The standardized visual monitoring system captures real-time video streams of household waste disposal behavior, and sequentially performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, disposal behavior pattern analysis, and disposal result effectiveness evaluation on the video streams, outputting information including personnel spatiotemporal information, object movement... This system analyzes multi-dimensional waste disposal behavior data, including movement trajectories and effectiveness assessments. It constructs a three-dimensional waste disposal behavior grading standard system, quantifies and calculates the behavioral characteristic parameters in the multi-dimensional data, and uses fuzzy logic rules to classify these parameters. Based on the grading results, it matches corresponding customized early warning command parameters. Structured early warning commands are generated from these parameters and distributed to visual prompting devices, audible warning devices, and waste disposal behavior recording devices. By deploying visual capture devices around waste collection equipment and combining computer vision and behavioral analysis technologies, the system achieves real-time monitoring, intelligent identification, and tiered early warning of waste disposal behavior, effectively regulating public waste disposal behavior and improving urban environmental sanitation management.

[0068] This invention achieves seamless coverage of the entire waste disposal process by deploying visual capture devices and modeling 3D monitoring areas through differentiated deployment and parameter self-adaptation optimization. This effectively solves the problems of poor environmental adaptability and incomplete coverage inherent in traditional monitoring methods. It maintains stable monitoring performance regardless of changes in lighting, pedestrian flow, or complex terrain, improving detection accuracy and reducing false alarm rates. Furthermore, by integrating multi-feature extraction and cross-frame matching technologies, this invention accurately identifies four types of behaviors: normal disposal, throwing, and indiscriminate littering. It overcomes the problems of fuzzy judgment and high false alarm rates in traditional methods by assessing the confidence level of behavior. Finally, by constructing a 3D classification standard system for disposal behavior and using fuzzy logic to achieve precise behavior classification, this invention matches customized early warning strategies, avoiding the inefficiency of traditional single-warning methods. Level 1 behaviors receive a light prompt, while level 3 behaviors receive a stronger warning, effectively curbing uncivilized disposal behavior while also considering user experience and improving intervention response efficiency. This invention combines multi-device collaborative early warning with environmental adaptive optimization to ensure low-latency transmission and accurate execution of early warning commands, improve the effective reach of early warnings in special scenarios such as strong light and high noise, and achieve closed-loop management through behavior recording, providing data traceability support for waste classification management, significantly reducing the cost of manual inspections, and helping to improve the level of refined and intelligent waste classification management.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A computer vision-based early warning method for household waste disposal behavior, characterized in that, The method includes: Visual capture devices are deployed in a differentiated manner around the household waste recycling equipment. The recycling equipment body area, normal waste disposal area and peripheral disposal monitoring area are delineated and a spatial mapping relationship is established. Dynamic video analysis parameters adapted to the scene are configured to output a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. The standardized visual monitoring system captures video streams of household waste disposal behavior in real time, and performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, disposal behavior pattern analysis, and disposal result effectiveness evaluation on the video streams in sequence, outputting multi-dimensional disposal behavior analysis data that includes personnel spatiotemporal information, object motion trajectory, and disposal effectiveness judgment results. A three-dimensional delivery behavior classification standard system is constructed, the delivery behavior feature parameters in the multi-dimensional delivery behavior analysis data are quantitatively calculated, and the delivery behavior feature parameters are classified by fuzzy logic judgment rules. Based on the behavior classification results, corresponding customized early warning instruction parameters are matched. Structured warning instructions are generated based on the customized warning instruction parameters and distributed to visual prompting devices, sound warning devices, and behavior recording devices.

2. The method for early warning of household waste disposal behavior based on computer vision according to claim 1, characterized in that, The aforementioned method involves deploying visual capture devices in a differentiated manner around the household waste recycling equipment, delineating the recycling equipment itself area, the normal waste disposal area, and the surrounding disposal monitoring area, establishing a spatial mapping relationship, configuring scene-adaptive dynamic video analysis parameters, and outputting a standardized visual monitoring system with full-scene coverage without blind spots and parameter self-adaptation capabilities. Specifically, this includes: The visual capture devices are deployed in a differentiated manner around the household waste recycling equipment. Specifically, a single camera or a multi-camera array configuration is selected based on the complexity of the disposal scene around the household waste recycling equipment, and a verification report on the signal and coverage of the device deployment is output. Based on the coverage verification report, and combined with the mapping relationship between the image coordinate system of the visual capture device and the physical space where the household waste recycling equipment is located, the physical boundaries, coordinate ranges and spatial relationships of the recycling equipment body area, the normal waste disposal area and the peripheral disposal monitoring area are determined, and three-dimensional monitoring area model data containing area location, size and hierarchical relationship are output. Based on the regional characteristics of the three-dimensional monitoring area model data, the sensitivity of personnel detection, behavior analysis time window, ambient light adaptation coefficient and early warning triggering basic threshold set are set, and the dynamic video analysis parameters adapted to the scene are output, resulting in a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. Based on the detection accuracy, false alarm rate, and environmental change data fed back during the operation of the standardized visual monitoring system, the dynamic video analysis parameters are adjusted in real time.

3. The method for early warning of household waste disposal behavior based on computer vision according to claim 1, characterized in that, The personnel target detection process in the video stream of household waste disposal behavior includes: The video stream of the household waste disposal behavior is subjected to noise filtering, illumination equalization and contrast enhancement processing to output pre-processed video frames of the disposal behavior. A deep learning model combined with a region proposal mechanism is used to extract multi-scale image features from the preprocessed video frames of the delivery behavior, generate candidate regions for personnel targets, and calculate the region detection confidence of the candidate regions for personnel targets. The confidence threshold is adaptively adjusted based on the complexity of the delivery scenario to filter candidate areas of the personnel target that have a detection confidence level higher than the confidence threshold, thereby generating the personnel target area; Based on the bounding box coordinates of the personnel target in the personnel target area, the personnel target is located in the physical space where the municipal solid waste recycling equipment is located, and the personnel physical coordinate data of the personnel target is output.

4. The method for early warning of household waste disposal behavior based on computer vision according to claim 3, characterized in that, The continuous tracking process of the motion trajectory of the video stream of household waste disposal behavior includes: Based on the physical coordinate data of the personnel and combined with the motion trajectory continuity constraint of the video stream of the household waste disposal behavior, a unique identifier is assigned to the personnel target, a binding relationship between the personnel target and the corresponding unique identifier is established, and a set of personnel target-identifier relationships is output. Extract the appearance features, motion features, and position features of the personnel target to generate initial motion trajectory data; A personnel target tracking model is constructed. In the continuous video frames of the video stream of the household waste disposal behavior, the personnel target is matched across frames. The appearance features, motion features and position features of the personnel target in the initial motion trajectory data are updated in real time, and the updated motion trajectory data is output. The occlusion determination conditions and the retention time of the disconnection status are set. In the updated motion trajectory data, combined with the personnel target-identifier relationship set, the trajectory prediction mechanism is used to maintain the corresponding unique identifier for the occluded personnel target and continuously track it. The corresponding unique identifier for the temporarily disconnected personnel target is temporarily stored until the temporarily disconnected personnel target reappears and the tracking is resumed. The personnel motion trajectory data of the personnel target is then output.

5. The method for early warning of household waste disposal behavior based on computer vision according to claim 4, characterized in that, The process of identifying the garbage object features in the video stream of household waste disposal behavior includes: Based on the motion characteristics of the target person in the motion trajectory data, the hand region of the target person is located by combining skin color feature and contour detection technology and a hand tracking window is established. The position change and shape characteristics of the hand region are monitored in real time, and the real-time coordinate data of the hand region and the hand motion state data are output. Based on the real-time coordinate data of the hand, the association between the person's hand and the garbage object is identified, and the shape features, texture features, and color features of the garbage object in the hand area are extracted and summarized to generate garbage object features; Based on the hand movement data and the characteristics of the garbage object, the relative movement speed, contour separation degree and shape change between the hand area and the garbage object are analyzed to determine whether the garbage object has separated from the hand of the person target. If the garbage object has separated from the hand of the person target, the corresponding separation time and separation position are extracted and the separation feature information is generated. For the waste object that separates from the hand of the person being tracked, an independent waste tracking mechanism is established. Combining the separation feature information, the movement path, speed, and trend of the waste object are recorded in real time and aggregated to generate waste object movement data.

6. The method for early warning of household waste disposal behavior based on computer vision according to claim 5, characterized in that, The analysis process of the behavior pattern of the video stream of household waste disposal includes: Based on the three-dimensional monitoring area model data and the physical coordinate data of the personnel, the actual physical distance between the personnel target and the domestic waste recycling equipment is calculated, the area where the personnel target is located is determined to be the normal waste disposal area or the peripheral disposal monitoring area, and the personnel area attribution determination result is output. The human key point detection technology is used to extract 21 core limb key points of the target person. Combined with the result of the person’s regional affiliation determination, the angle change rate and distance change rate between the core limb key points are analyzed to identify typical person’s delivery posture and output the delivery posture feature set. The motion path, speed, and trend of the garbage object are analyzed to determine whether the trajectory pattern of the garbage object belongs to the normal placement mode or the throwing mode, and the garbage trajectory pattern determination result is output. By integrating the actual physical distance, the disposal posture feature set, and the garbage trajectory pattern determination results, four types of disposal behaviors are comprehensively identified: normal disposal, close-range disposal, long-range disposal, and random littering. The disposal behavior determination results and the corresponding disposal behavior determination confidence levels are output.

7. The method for early warning of household waste disposal behavior based on computer vision according to claim 6, characterized in that, The process for evaluating the effectiveness of the video stream of household waste disposal behavior includes: Based on the motion data of the garbage object, the predicted landing point of the garbage object and the corresponding reliability of the predicted landing point are predicted. The movement of the garbage object is tracked in real time by a visual monitoring system. The actual landing point of the garbage object is located based on the predicted landing point. The consistency between the actual landing point and the predicted landing point is compared, and the landing point matching result is output. Based on the actual landing point and the range of the recycling equipment body area in the three-dimensional monitoring area model data, a distance judgment threshold is set. If the minimum distance between the actual landing point and the boundary of the recycling equipment body area is less than the distance judgment threshold, the garbage object is judged to be successfully disposed of; otherwise, it is considered a failure, and the garbage disposal judgment result is output. Based on the landing point matching results, the clarity of the visual monitoring image, and the continuity of the garbage object's movement, the reliability of the garbage disposal determination result is quantitatively evaluated, and the confidence level of the garbage disposal determination is output.

8. The method for early warning of household waste disposal behavior based on computer vision according to claim 1, characterized in that, The process of constructing the three-dimensional delivery behavior grading standard system includes: The criteria for judging normal placement behavior and three levels of uncivilized placement behavior are set. Level 1 uncivilized placement behavior includes short-distance throwing and mild long-distance throwing; Level 2 uncivilized placement behavior includes severe long-distance throwing and mild placement failure; and Level 3 uncivilized placement behavior includes severe placement failure and malicious throwing. The three-dimensional placement behavior classification standard system including behavior distance, placement method and placement result is constructed.

9. A computer vision-based early warning system for household waste disposal behavior, applied to the computer vision-based early warning method for household waste disposal behavior as described in any one of claims 1-8, characterized in that, The system includes: The monitoring system output module is used to deploy visual capture devices in a differentiated manner around the household waste recycling equipment, delineate the recycling equipment body area, normal waste disposal area and peripheral disposal monitoring area and establish spatial mapping relationship, configure scene-adaptive dynamic video analysis parameters, and output a standardized visual monitoring system with full scene coverage without blind spots and parameter self-adaptation capability. The waste disposal behavior analysis module is used to capture video streams of waste disposal behavior in real time through the standardized visual monitoring system. The module performs personnel target detection, continuous motion trajectory tracking, waste object feature recognition, waste disposal behavior pattern analysis, and waste disposal result effectiveness evaluation on the video streams of waste disposal behavior in sequence. It outputs multi-dimensional waste disposal behavior analysis data that includes personnel spatiotemporal information, object motion trajectory, and waste disposal effectiveness judgment results. The early warning instruction matching module is used to construct a three-dimensional delivery behavior classification standard system, quantitatively calculate the delivery behavior feature parameters in the multi-dimensional delivery behavior analysis data, classify the delivery behavior feature parameters according to fuzzy logic judgment rules, and match the corresponding customized early warning instruction parameters based on the behavior classification results. The early warning instruction distribution module is used to generate structured early warning instructions based on the customized early warning instruction parameters, and distribute them to visual prompting devices, sound warning devices, and behavior recording devices.

Images