Parcel detecting and sorting method and system based on machine vision

Abstract

The invention relates to the technical field of parcel sorting, in particular to a parcel detecting and sorting method and system based on machine vision. According to the method, historical parcel motion trajectory data and real-time video information are fused, a category and weight differentiation trajectory model is constructed, sudden offset is captured in combination with computer vision, the trajectory is accurately predicted, and the attitude offset is obtained. The initial fixed parameter of the actuator is dynamically adjusted to be a first action parameter based on the offset, a compensation coefficient is generated through trajectory tracking error verification, and a second action parameter is obtained; meanwhile, friction coefficient changes are calculated through the real-time track and pressure data, a dynamic response coefficient is determined, and finally a multi-sorting-port cooperation instruction is generated to achieve path dynamic distribution. According to the process, through historical and real-time data fusion, dynamic parameter adjustment, closed-loop feedback and multi-system cooperation, the problems of traditional sorting static presetting, parameter blind adjustment, resource allocation imbalance and the like are solved, and the sorting precision, adaptability and efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of parcel sorting technology, and in particular to a parcel detection and sorting method and system based on machine vision. Background Technology

[0002] In today's rapidly developing logistics industry, parcel sorting, as a core process in logistics transfer, directly impacts the operational efficiency of the entire logistics system. With the explosive growth of e-commerce, the number of parcels has surged, exhibiting diverse shapes, specifications, and materials. Traditional manual sorting methods are no longer sufficient to meet the demands for efficient sorting, making automated and intelligent parcel sorting technology an inevitable trend in the industry.

[0003] Currently, automated parcel sorting systems mostly use mechanical transmission combined with sensor recognition to achieve sorting operations. However, in practical applications, there are still many technical bottlenecks. For example, during the transportation of parcels on conveyor belts, they are prone to changes in posture due to collisions, vibrations, or shifts in their own center of gravity, such as positional shifts or angular deflections. Existing systems usually control the sorting actuators based on preset trajectories or fixed parameters, which cannot sense and dynamically compensate for the posture shifts of the parcels in real time. When a parcel is subjected to sudden external forces, such as squeezing from adjacent parcels or fluctuations in conveyor belt speed, the actual movement trajectory deviates significantly from the predicted trajectory, and the error accumulates and becomes larger and larger during transportation. This will cause the actuator's point of action or the direction of force to deviate from the expected direction, resulting in sorting errors or parcels falling, reducing the accuracy of sorting. Therefore, a parcel detection and sorting method based on machine vision is needed to solve the above problems. Summary of the Invention

[0004] The main objective of this invention is to provide a package detection and sorting method and system based on machine vision, aiming to solve the technical problems in the prior art.

[0005] This invention proposes a machine vision-based package detection and sorting method and system, including a sorting actuator, and further comprising: Acquire real-time video information and historical package movement trajectory data; use real-time video information and historical package movement trajectory data to obtain the package attitude offset for the prediction period. Obtain the initial parameters of the sorting executor, and obtain the first action parameter based on the initial parameters and the package posture offset during the predicted time period; The corresponding motion coordinate path is obtained by correcting the package's motion trajectory according to the first action parameter, and the trajectory tracking error is calculated based on the motion coordinate path and the predicted package motion trajectory. A compensation adjustment coefficient is generated based on the trajectory tracking error and the first action parameter, and a second action parameter is obtained based on the compensation adjustment coefficient. The real-time movement trajectory of the package is obtained. The change in the coefficient of friction of the package is obtained by monitoring the real-time movement trajectory of the corresponding package based on the pressure sensor. The dynamic response coefficient of the sorting actuator is calculated based on the change in the coefficient of friction of the package. Based on the second action parameter and the dynamic response coefficient, a collaborative control instruction for multiple sorting ports is generated, and dynamic allocation of package paths is performed according to the collaborative control instruction.

[0006] Preferably, the step of obtaining real-time video information and historical package motion trajectory data, and obtaining the package posture offset for the prediction period from the real-time video information and historical package motion trajectory data, includes: Extract multiple historical package movement trajectories from the package movement trajectory data of historical sorting scenarios, and obtain the average historical package movement trajectory based on the multiple historical package trajectories; The real-time video information is split according to a preset video frame sequence length to obtain multiple video frame sequence lengths. Multiple real-time motion coordinates of the corresponding packages are obtained according to the multiple video frame sequence lengths, and a real-time package motion trajectory is generated according to the multiple real-time motion coordinates of the packages. Obtain the preset standard package movement trajectory; The real-time dynamic offset coefficient of the package posture is obtained based on the real-time package posture motion trajectory and the preset standard package motion trajectory. Historical offset benchmarks are obtained based on the average historical parcel movement trajectory and the preset standard parcel movement trajectory. The predicted package offset is obtained based on the real-time dynamic offset coefficient of the package posture and the historical offset benchmark.

[0007] Preferably, the step of obtaining the initial parameters of the sorting actuator and obtaining the first action parameter based on the initial parameters and the package posture offset during the predicted time period includes: The initial thrust vector, initial angular components, and initial duration of action are obtained based on the initial parameters. Obtain the package attitude offset during the prediction period, wherein the attitude offset includes a position offset component and an angle offset component; The thrust correction is obtained based on the position offset component, and the angle correction is obtained based on the angle offset component; The corrected thrust vector is obtained based on the initial thrust vector and the thrust correction amount, and the corrected angle component is obtained based on the initial angle component and the angle correction amount. The first operating parameters of the sorting actuator are generated based on the corrected thrust vector, the corrected angle components, and the initial operating duration.

[0008] Preferably, the step of correcting the package's motion trajectory according to the first action parameter to obtain the corresponding motion coordinate path, and calculating the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory, includes: Obtain the first action parameter, wherein the first action parameter includes a thrust component, an angle component, and an action duration component; The force direction vector of the sorting actuator is obtained based on the thrust component and the angle component. Based on the force direction vector and duration component, a sorting actuator control command is generated; based on the actuator control command, the package movement trajectory is obtained; and based on the package movement trajectory, multiple continuous position coordinates are obtained. Generate a motion coordinate path based on multiple consecutive position coordinates; Obtain the predicted trajectory of the package, and calculate the trajectory tracking error based on the predicted trajectory and the movement coordinate path.

[0009] Preferably, the step of generating a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtaining the second action parameter based on the compensation adjustment coefficient, includes: The current thrust vector, current angular component, and current duration of action are obtained based on the first action parameter. The position deviation component and angle deviation component are obtained based on the current trajectory tracking error; The average position deviation is obtained for each of the position deviation components, and the position deviation coefficient is obtained for each of the average position deviation components. The corresponding average angle deviation is obtained for each of the angle deviation components, and the angle deviation coefficient is obtained for each of the average angle deviation components. Obtain the corresponding weighted normalization coefficient based on each of the position deviation coefficients and angle deviation coefficients; A position deviation correction coefficient is obtained based on each of the weight normalization coefficients and position deviation coefficients; an angle deviation correction coefficient is obtained based on each of the weight normalization coefficients and angle deviation coefficients. The basic compensation amount is obtained based on the thrust vector and angle components in the first action parameter; The compensation adjustment coefficient is obtained based on each of the aforementioned basic compensation amounts, position deviation correction coefficients, and angle deviation correction coefficients; The first action parameter is adjusted twice based on each of the compensation adjustment coefficients to obtain the second action parameter.

[0010] Preferably, the steps of acquiring the real-time movement trajectory of the package, obtaining the change in the package friction coefficient based on the real-time movement trajectory monitored by the pressure sensor, and calculating the dynamic response coefficient of the sorting actuator based on the change in the package friction coefficient include: Multiple real-time movement coordinates of packages are obtained based on the real-time movement trajectory of the packages. Multiple movement time series are obtained by differential calculation based on the multiple real-time movement coordinates of the packages. The corresponding movement speed series are obtained based on the multiple movement time series. Obtain the material of the package's outer packaging, and then determine the corresponding basic coefficient of friction based on that material. Obtain the normal pressure and tangential friction force of the target package based on the pressure data; The material's adaptive friction coefficient is obtained based on the motion velocity sequence, normal pressure, and tangential friction force. Obtain multiple material adaptive friction coefficients within a preset time period, and generate material adaptive friction dynamic change curves based on the multiple material adaptive friction coefficients; Multiple material adaptive friction dynamic peaks are extracted from the material adaptive friction dynamic change curve, and the time-varying friction coefficient change is obtained from the multiple material adaptive friction dynamic peaks and the basic friction coefficient. The dynamic response coefficient is generated based on the basic friction coefficient and the time-varying friction coefficient change.

[0011] Preferably, the step of generating collaborative control instructions for multiple sorting ports based on the second action parameter and the dynamic response coefficient includes: The preset target position of the corresponding sorting port is obtained according to each of the second action parameters, and the package trajectory correction amount is obtained according to each preset target position and dynamic response coefficient; Determine whether the deviation between the current package position and the preset target position at each sorting point exceeds a preset threshold range; If the current package position deviation at the sorting point exceeds a preset threshold range, the package trajectory is determined to be abnormal. Adjust the dynamic response coefficient until the package position deviation is within the preset threshold range, and determine the response coefficient at this point as the final response coefficient; The path allocation sensitivity of the corresponding sorting port is obtained based on each final response coefficient and trajectory correction amount; Based on the path allocation sensitivity, final response coefficient, and trajectory correction amount, collaborative control instructions for multiple sorting ports are generated.

[0012] This application also provides a machine vision-based package detection and sorting system, including a sorting actuator, and further comprising: The first acquisition module acquires real-time video information and historical package movement trajectory data, and uses the real-time video information and historical package movement trajectory data to obtain the package posture offset during the prediction period. The second acquisition module acquires the initial parameters of the sorting executor and acquires the first action parameter based on the initial parameters and the package posture offset during the predicted time period. The first correction module corrects the package's motion trajectory according to the first action parameter to obtain the corresponding motion coordinate path, and calculates the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory. The first generation module generates a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtains the second action parameter based on the compensation adjustment coefficient. The third acquisition module acquires the real-time movement trajectory of the package, obtains the change in the friction coefficient of the package based on the real-time movement trajectory monitored by the pressure sensor, and calculates the dynamic response coefficient of the sorting actuator based on the change in the friction coefficient of the package. The second generation module generates collaborative control instructions for multiple sorting ports based on the second action parameter and the dynamic response coefficient, and performs dynamic allocation of package paths according to the collaborative control instructions.

[0013] As a preferred embodiment, the first acquisition module includes: The first extraction unit extracts multiple historical package movement trajectories based on package movement trajectory data from historical sorting scenarios, and obtains the average historical package movement trajectory based on the multiple historical package trajectories. The first acquisition unit splits the real-time video information according to a preset video frame sequence length to obtain multiple video frame sequence lengths, obtains multiple real-time motion coordinates of the corresponding packages according to the multiple video frame sequence lengths, and generates a real-time package motion trajectory according to the multiple real-time motion coordinates of the packages. The second acquisition unit acquires the preset standard package movement trajectory; The third acquisition unit acquires the real-time dynamic offset coefficient of the package posture based on the real-time package posture motion trajectory and the preset standard package motion trajectory. The fourth acquisition unit acquires historical offset benchmarks based on the historical average movement trajectory of packages and the preset standard package movement trajectory; The fifth acquisition unit acquires the package offset for the predicted time period based on the real-time package attitude dynamic offset coefficient and the historical offset benchmark.

[0014] Preferably, the second acquisition module includes: The sixth acquisition unit acquires the initial thrust vector, initial angular components, and initial action duration based on the initial parameters; The seventh acquisition unit acquires the package attitude offset during the prediction period, wherein the attitude offset includes a position offset component and an angle offset component. The eighth acquisition unit acquires the thrust correction amount based on the position offset component and the angle correction amount based on the angle offset component; The ninth acquisition unit acquires the corrected thrust vector based on the initial thrust vector and the thrust correction amount, and acquires the corrected angle component based on the initial angle component and the angle correction amount. The first generation unit generates the first operating parameters of the sorting actuator based on the corrected thrust vector, the corrected angle component, and the initial operating duration.

[0015] The beneficial effects of this invention are as follows: First, by using historical parcel movement trajectory data from sorting scenarios, this invention constructs trajectory models for parcels of different categories, such as cardboard boxes and cloth bags, and of different weights. This clarifies their motion patterns, such as acceleration and turning angle, in scenarios like conveyor belts and sorting slots, providing a benchmark for prediction. Simultaneously, by combining real-time video information collected by high-definition cameras and 3D vision sensors, computer vision algorithms capture sudden changes such as offsets caused by collisions and vibrations. This dual coverage of patterns and sudden changes avoids the randomness of pure real-time data prediction and the static defects of traditional fixed parameter presets, making the trajectory prediction results more closely match the actual scenario and providing accurate basis for subsequent sorting actions. Based on the predicted parcel posture offset, the initial fixed parameters of the actuator, such as the robotic arm's gripping force and the guide wheel's turning angle, are transformed into dynamically adjustable primary action parameters, ensuring that the parameters directly match the actual parcel posture and avoiding blind parameter adjustments. Then, through trajectory tracking error quantification analysis, a compensation adjustment coefficient is generated to obtain the second action parameter, realizing progressive optimization of primary adjustment + secondary compensation. This solves the problem of insufficient precision in single adjustment, upgrading the actuator's actions from initial adaptation to precise adaptation, significantly improving the accuracy of grasping, guiding, and other actions. By comparing the corrected trajectory with the predicted trajectory, the trajectory tracking error is calculated, transforming the actuator parameter adjustment effect into a quantifiable indicator, forming a closed-loop logic of adjustment-verification-feedback. This breaks through the open-loop mode of traditional systems where parameter adjustments lack effect verification, ensuring the effectiveness of trajectory correction and providing a clear basis for continuous parameter optimization. Through real-time package movement trajectory and pressure sensor data, the differences caused by friction coefficient changes such as conveyor belt material and surface condition changes are calculated in real time, generating the dynamic response coefficient of the sorting actuator. This allows the actuator's response speed, force, and other parameters to adjust in real time according to environmental changes. This solves the problem that fixed response parameters cannot adapt to environmental fluctuations, avoiding motion deviations such as overspeeding and inaccurate stopping caused by changes in friction coefficient. Based on the second action parameter and dynamic response coefficient, multi-sorting port collaborative control commands are generated to achieve dynamic allocation of package paths. This breaks the traditional model of independent operation of each sorting station, solving the problems of resource imbalance, congestion in some sorting stations, idleness in others, and inability to adjust when path planning is out of sync with real-time status. It effectively avoids package accumulation and improves the overall throughput and anti-congestion capability of the sorting system. In summary, this invention comprehensively improves the accuracy, adaptability, and efficiency of package sorting through data fusion, dynamic adjustment, closed-loop feedback, environmental adaptation, and multi-system collaboration, and is suitable for various complex sorting scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] like Figure 1 As shown, this application provides a machine vision-based package detection and sorting method, including a sorting actuator, and further comprising: S1. Obtain real-time video information and historical package movement trajectory data. Obtain the package posture offset during the prediction period from the real-time video information and historical package movement trajectory data. S2. Obtain the initial parameters of the sorting executor, and obtain the first action parameter based on the initial parameters and the package posture offset during the predicted time period. S3. Correct the package's motion trajectory according to the first action parameter to obtain the corresponding motion coordinate path, and calculate the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory. S4. Generate a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtain the second action parameter based on the compensation adjustment coefficient; S5. Obtain the real-time movement trajectory of the package. Based on the pressure sensor monitoring the real-time movement trajectory of the corresponding package, obtain the change in the friction coefficient of the package. Calculate the dynamic response coefficient of the sorting actuator based on the change in the friction coefficient of the package. S6. Generate collaborative control instructions for multiple sorting ports based on the second action parameter and dynamic response coefficient, and perform dynamic allocation of package paths based on the collaborative control instructions.

[0021] As described in steps S1-S6 above, this invention first uses a dual-dimensional data input of package movement trajectory data from historical sorting scenarios and real-time video information to achieve dynamic prediction of package movement trajectories. Based on the comparison between the predicted trajectory and historical trajectory data, it accurately obtains the package posture offset during the prediction period. Its core advantages lie in the accuracy of dynamic prediction and the comprehensiveness of data fusion: on the one hand, historical data provides basic pattern support for prediction, such as common movement trajectory characteristics of different types of packages, avoiding the randomness of pure real-time data prediction; on the other hand, real-time video information can capture sudden changes in packages in the current scenario, such as trajectory offsets caused by collisions or conveyor belt vibrations, achieving dual coverage of patterns and sudden changes. This prediction model, which combines historical data with real-time correction, breaks through the static thinking of traditional sorting systems that rely on fixed parameters and preset trajectories. It makes the trajectory prediction results more closely match the actual scenario. Feature extraction is performed on package movement trajectory data from historical sorting scenarios to construct trajectory models for different categories such as cardboard boxes, cloth bags, foam boxes, and packages of varying weights. This clarifies their movement patterns, such as acceleration and turning angle ranges, in scenarios like conveyor belts and sorting slots, providing a baseline for prediction. High-definition cameras and 3D vision sensors are used to collect real-time video information. Computer vision algorithms, such as target detection and optical flow, are used to track the position and angle changes of packages in real time. Real-time data is compared with historical models to identify abnormal movements that deviate from the pattern, such as trajectory deviations caused by sudden collisions. This is then compared with the package posture offset obtained from S1 during the prediction period to generate the first action parameter. Its core advantages lie in the targeted nature of parameter adjustments and the dynamic adaptability of initial parameters: initial parameters, such as the gripping force of the robotic arm and the steering angle of the guide wheels, are usually preset based on the standard package posture. However, S2 incorporates the posture offset into the parameter adjustment logic, transforming the actuator parameters from fixed values ​​to dynamic values, ensuring that the parameters match the actual posture of the package. This direct correlation mechanism between offset and parameter avoids the blindness of parameter adjustments, improves the accuracy of actuator actions, corrects the package's motion trajectory based on the generated first action parameter, obtains the motion coordinate path, and calculates the trajectory tracking error by comparing this path with the predicted package motion trajectory. Its core advantages lie in the establishment of a closed-loop feedback mechanism and the quantitative presentation of trajectory deviation: by comparing the corrected trajectory with the predicted trajectory, the actual effect of the actuator parameter adjustment is transformed into a quantifiable error value, providing a clear basis for subsequent optimization. This adjustment-verification-feedback logic breaks through the open-loop mode of traditional systems where there is no effect verification after parameter adjustment, ensuring the effectiveness of trajectory correction. Based on the calculated trajectory tracking error and the generated first action parameter, a compensation adjustment coefficient is generated, thus obtaining the second action parameter. Its core advantages lie in the accuracy of error compensation and the progressiveness of parameter optimization: through quantitative analysis of trajectory tracking error, compensation coefficients are generated in a targeted manner, and the first action parameter is optimized in a second way, so that the actuator parameters are upgraded from initial adaptation to precise adaptation.This single adjustment + secondary compensation model solves the problem of insufficient precision that may exist in single parameter adjustment, and greatly improves the accuracy of actuator action. The precision of single adjustment is limited: the first action parameter is only initially adjusted based on the attitude offset, without considering the actual error after trajectory correction, such as residual error caused by actuator mechanical clearance and conveyor belt vibration, resulting in parameter deviation. Error and compensation are not related: even if trajectory error is detected, due to the lack of a quantitative model of error and compensation coefficient, the adjustment of compensation parameters relies on experience, such as manually increasing the steering angle of the guide wheel, which may lead to overcompensation, such as the package shifting in the opposite direction due to an excessively large steering angle, or insufficient compensation that does not eliminate the error. By acquiring the real-time movement trajectory of the package and pressure sensor data, the change in the package friction coefficient is calculated, and then the dynamic response coefficient of the sorting actuator is obtained. Its core advantages lie in the dynamic adaptation to environmental factors and the real-time response of the actuator: The coefficient of friction is a key environmental factor affecting the movement of packages. Changes in environmental factors such as the material of the conveyor belt and the surface roughness of the package will cause changes in the coefficient of friction. By monitoring the changes in the coefficient of friction in real time through pressure data, such as the pressure of the package on the conveyor belt, the actuator's response speed, force and other parameters can be dynamically adjusted with environmental changes, avoiding the problem that fixed response parameters cannot adapt to environmental fluctuations. Changes in the coefficient of friction cause movement deviations: When packages move on conveyor belts of different materials, such as rubber and metal mesh, the coefficients of friction are different. Rubber belts have a high coefficient of friction, while metal mesh has a low coefficient of friction. If the output power of the actuator, such as the drive motor, is fixed, it will cause the package movement speed to deviate from the expected speed. For example, on the metal mesh, due to the low friction, the package may exceed the speed limit. Fixed actuator response parameters: The dynamic response parameters of the actuator, such as acceleration time and braking force, are based on the standard coefficient of friction preset. When the coefficient of friction changes abruptly, such as when the conveyor belt gets wet and the coefficient of friction drops sharply, the response parameters cannot be adjusted in time, resulting in inaccurate package positioning, such as overshooting the target slot. Based on the second action parameter and the dynamic response coefficient of S5, collaborative control commands for multiple sorting ports are generated to realize the dynamic allocation of package paths. Its core advantages lie in the globalization of multi-device collaboration and path optimization: breaking the traditional independent operation mode of each sorting station in the sorting system, it realizes the dynamic allocation of resources across multiple stations through collaborative control commands, upgrading path allocation from fixed planning to real-time optimization. This global collaboration mechanism improves the overall efficiency and anti-congestion capability of the sorting system, addressing the imbalance in sorting station resource allocation: each sorting station receives packages independently, lacking global scheduling, leading to package accumulation and congestion at some sorting stations, such as those to popular destinations, while other sorting stations are idle, resulting in low overall efficiency and a disconnect between path planning and real-time status: package paths are fixed based on the initial plan, and when a sorting station suddenly malfunctions, such as grid blockage, the system cannot adjust the path in real time, causing packages to be delayed. Based on the above steps, this application can not only improve sorting accuracy but also realize dynamic adjustment of sorting across multiple sorting stations to avoid package accumulation.

[0022] In one embodiment, the step of obtaining real-time video information and historical package motion trajectory data, and obtaining the package posture offset for the prediction period from the real-time video information and historical package motion trajectory data, includes: S201. Extract multiple historical package movement trajectories based on the package movement trajectory data of historical sorting scenarios, and obtain the average historical package movement trajectory based on the multiple historical package trajectories. S202. The real-time video information is split according to a preset video frame sequence length to obtain multiple video frame sequence lengths. Multiple real-time motion coordinates of the packages are obtained according to the multiple video frame sequence lengths, and a real-time package motion trajectory is generated according to the multiple real-time motion coordinates of the packages. S203. Obtain the preset standard package movement trajectory; S204. Obtain the real-time dynamic offset coefficient of the package posture based on the real-time package posture motion trajectory and the preset standard package motion trajectory. S205. Obtain historical offset benchmarks based on the historical average movement trajectory of packages and the preset standard movement trajectory of packages; S206. Obtain the package offset for the predicted time period based on the real-time package attitude dynamic offset coefficient and the historical offset benchmark.

[0023] As described in steps S201-S206 above, this invention analyzes a large number of package movement trajectories in historical sorting scenarios, extracts multiple historical package movement trajectories, and calculates their average value to form the historical package average movement trajectory. The core advantage of this method lies in utilizing the statistical characteristics of historical data to eliminate random noise and abnormal fluctuations in a single trajectory, thereby obtaining a more representative standard movement pattern. The accumulation of historical data can reflect the typical trajectory characteristics of the sorting system under normal operating conditions, providing a reliable benchmark for subsequent real-time trajectory comparison. Subsequently, the real-time video information is split according to a preset frame sequence length, and the package position in each video frame is extracted using computer vision technology to generate the real-time package movement trajectory. Compared with traditional direct sensor measurement, video processing has the advantages of non-contact, global field of view, and simultaneous monitoring of multiple targets. By analyzing continuous video frames, the complete movement process of the package in three-dimensional space can be captured, including complex posture changes such as translation and rotation. The real-time video is split into fixed-length frame sequences (e.g., 30 frames per sequence) to ensure that each sequence contains sufficient motion information. Then, object detection algorithms (such as YOLO or Faster R-CNN) are used to identify the position and bounding box of the package in each frame. Tracking algorithms (such as DeepSORT) are used to associate the same package in different frames, thus obtaining a continuous motion trajectory. For example, in a logistics sorting scenario, by analyzing a video stream of 30 frames per second, the system can accurately track the motion trajectory of the package, including its translation and slight rotation on the conveyor belt. By introducing a preset standard package motion trajectory as an ideal reference model, a clear target benchmark is provided for subsequent offset calculations. The standard trajectory can be based on theoretical analysis, simulation, or best practices optimized over a long period, representing the optimal motion path that the package should follow during the sorting process. This preset benchmark approach allows the system to quickly determine whether the current package movement deviates from the ideal state without relying on real-time adjustments, improving response speed. The preset standard package motion trajectory can be obtained in various ways. For example, during the system design phase, the optimal motion trajectory of the package under ideal conditions can be calculated through mechanical analysis and kinematic simulation; or during the system debugging phase, the best-performing actual trajectory can be recorded as a standard through multiple trials and optimizations. For example, for a package of a specific size and weight, simulation calculations determine its optimal trajectory on a conveyor belt is a straight line, with a speed maintained at 1.5 m / s and an acceleration not exceeding 0.2 m / s². These parameters are fixed as a standard trajectory. By comparing the real-time package trajectory with the preset standard trajectory, a dynamic offset coefficient for the real-time package posture is calculated. This coefficient quantifies the degree of deviation of the current package movement from the ideal state. Unlike simple position deviation calculations, the dynamic offset coefficient considers the time-series characteristics and motion trends of the trajectory, and can more comprehensively reflect the changes in the package posture.For example, even if two packages have the same positional deviation at a certain moment, if their trajectory trends are different—one returning to the standard trajectory and the other moving away—the dynamic offset coefficient will give different results. The dynamic offset coefficient can be calculated using various methods, such as Dynamic Time Warping (DTW) or Kalman filtering. Taking DTW as an example, this algorithm aligns the real-time trajectory with the standard trajectory on the time axis, calculates their similarity, and converts the similarity into an offset coefficient. For instance, by calculating the minimum cumulative distance between two trajectories in the time series, the distance value is normalized to obtain an offset coefficient between 0 and 1; a larger value indicates a more severe deviation. By comparing the historical average movement trajectory of packages with a preset standard trajectory, a historical offset baseline is calculated. The historical offset baseline reflects the systematic deviations that exist in the system during long-term operation, such as slight conveyor belt deviations or trajectory deviations caused by mechanical wear. By separating these systematic deviations, the system can more accurately determine whether the current package's deviation is caused by temporary factors such as improper package placement or by problems within the system itself. Combining the real-time package attitude dynamic offset coefficient and the historical offset baseline, the system can predict the package's attitude deviation in future periods. This predictive capability allows the system to make adjustments in advance, avoiding the discovery of anomalies only when packages arrive at the sorting execution position, thereby improving sorting accuracy and efficiency. By considering historical trends and real-time dynamics, the prediction results are more reliable and can adapt to changes in different package types and sorting scenarios. The calculation of package offset over a predicted period can be based on time series prediction models, such as ARIMA, LSTM, or Kalman filtering. Taking LSTM neural networks as an example, this model can learn the relationship between historical offset data and real-time dynamic offset coefficients to predict offset trends over a future period. For example, by inputting offset data from the past 10 seconds and the current dynamic offset coefficients, the LSTM model can predict the package offset within the next 2 seconds. In practical applications, the system can adjust the parameters of the sorting actuators, such as the gripping position and angle of the robotic arm, in advance based on the prediction results.

[0024] In one embodiment, the step of obtaining the initial parameters of the sorting actuator and obtaining the first action parameter based on the initial parameters and the package posture offset during the predicted time period includes: S301. Obtain the initial thrust vector, initial angular components, and initial duration of action based on the initial parameters. S302. Obtain the package attitude offset during the predicted time period, wherein the attitude offset includes a position offset component and an angle offset component; S303. Obtain the thrust correction amount based on the position offset component, and obtain the angle correction amount based on the angle offset component; S304. Obtain the corrected thrust vector based on the initial thrust vector and the thrust correction amount, and obtain the corrected angle component based on the initial angle component and the angle correction amount; S305. Generate the first operating parameters of the sorting actuator based on the corrected thrust vector, the corrected angle component, and the initial operating duration.

[0025] As described in steps S301-S305 above, this invention decomposes the initial parameters of the sorting actuator into an initial thrust vector, including the magnitude and direction of the thrust; an initial angular component, the interaction angle between the actuator and the package (e.g., the gripping angle of the robotic arm, the deflection angle of the guide plate); and an initial duration, the duration for which the actuator applies force to the package (e.g., the gripping time of the robotic arm, the jetting duration of the jetting device). Its advantage lies in the structured decomposition and quantitative representation of the parameters: by transforming the abstract initial parameters into three measurable and calculable physical quantities, a visual description of the actuator's actions is achieved. This decomposition breaks the limitation of the traditional system where "initial parameters are treated as a black box," making the physical meaning of each parameter clear and explicit, providing precise adjustment targets for subsequent targeted corrections, obtaining the package's attitude offset, and decomposing it into a positional offset component (e.g., linear offset distances in the X and Y axes) and an angular offset component (e.g., the rotation angle and tilt angle of the package around the Z-axis). Its advantage lies in the structured representation and dimensional subdivision of the offset: by clearly defining the position-angle dual-dimensional attributes of the offset, the ambiguity caused by the general description of the offset in the traditional system, such as the package being tilted to the left, is avoided. This detailed breakdown elevates the offset from a qualitative description to quantitative data, providing precise problem coordinates for subsequent targeted corrections. It establishes a direct correlation between position offset components and thrust corrections, and between angular offset components and angular corrections. Specifically, it calculates the required thrust adjustment (e.g., increasing / decreasing thrust, changing thrust direction) based on the direction and magnitude of the position offset, and calculates the required actuator angle adjustment (e.g., increasing / decreasing deflection angle) based on the degree of angular offset. Its advantages lie in the targeted and quantitative calculation of the corrections: the correction amount no longer relies on empirical estimation but is derived from mathematical models based on the specific values ​​of the offset components, ensuring that the correction is exactly what is expected, avoiding blind corrections. The calculated thrust and angular corrections are then superimposed on the acquired initial thrust vector and initial angular components, respectively, to obtain the corrected thrust vector (including the adjusted magnitude and direction) and the corrected angular components. Its advantages lie in the continuity and accuracy of the correction: the correction does not completely discard the initial parameters, but makes incremental adjustments based on them. It retains the parts of the initial parameters that are adapted to the characteristics of the package, such as the low initial thrust for fragile items, and makes up for the deviation caused by the offset through the correction amount. It avoids the instability of the actuator action caused by parameter mutation. It integrates the corrected thrust vector, the corrected angle component and the initial action duration to generate the first action parameter of the sorting actuator. Its advantages lie in the integrity and coordination of the parameters. The first action parameter contains the three elements of the actuator action: "force-angle-time", which ensures that the three are matched with each other. For example, when the thrust is increased, the action duration can be appropriately shortened to avoid over-pushing. It solves the problem of parameter fragmentation in traditional systems, such as adjusting only the thrust and ignoring the duration, which leads to the lack of coordination of actions.

[0026] In one embodiment, the step of correcting the package's motion trajectory according to the first action parameter to obtain the corresponding motion coordinate path, and calculating the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory, includes: S401. Obtain the first action parameter, wherein the first action parameter includes a thrust component, an angle component, and an action duration component; S402. Obtain the force direction vector of the sorting actuator based on the thrust component and the angle component; S403. Generate sorting actuator control instructions based on the force direction vector and duration component, obtain package movement trajectory based on the actuator control instructions, and obtain multiple continuous position coordinates based on the package movement trajectory. S404. Generate a motion coordinate path based on the multiple continuous position coordinates; S405. Obtain the predicted package movement trajectory and obtain the trajectory tracking error based on the predicted package movement trajectory and movement coordinate path.

[0027] As described in steps S401-S405 above, this invention obtains the first action parameter and clarifies that it includes the magnitude and direction decomposition values ​​of the force applied by the actuator to the package, such as X-axis thrust 3N, Y-axis thrust 2N, the tilt angle or rotation angle when the actuator contacts the package, such as the robotic arm gripping angle 30°, the guide plate deflection angle 15°, and the duration of the actuator's continuous force application to the package, such as the clamping duration 0.5 seconds, the jetting duration 0.3 seconds. Its advantage lies in the structured and resolvable nature of the parameters: by decomposing the first action parameter into three components with clear physical meaning, it breaks the limitation of the traditional system where the action parameter is a fuzzy instruction set, making the function and influence range of each parameter clearly identifiable, providing precise input variables for subsequent force analysis and trajectory tracking. The obtained thrust and angle components are used to calculate the force direction vector, which includes the magnitude and direction of the applied force (e.g., magnitude 5N, direction with an angle of 30° to the X-axis). Its advantages lie in the quantification and precise representation of the force direction: integrating the magnitude of the thrust and the angle of action into a vector with a clear spatial orientation solves the directional ambiguity problem caused by the separation of thrust and angle description in traditional systems. This makes the force applied by the actuator to the package predictable and calculable. Control commands for the sorting actuator, such as motor drive signals and pneumatic valve switching signals, are generated based on the force direction vector and the duration component. The movement trajectory of the package under the control commands is acquired through a vision sensor or position encoder. Multiple continuous position coordinates are extracted from the trajectory, such as recording X and Y coordinates every 0.1 seconds. Its advantage lies in the closed-loop association between commands and trajectories: directly binding the actuator's control commands to the actual movement trajectory of the package, and achieving real-time verification of command output and trajectory feedback through continuous coordinate recording, solves the open-loop control problem of no motion tracking after command issuance in traditional systems. Based on the acquired multiple continuous position coordinates... The coordinate system generates a continuous and smooth trajectory curve for the motion coordinate path through interpolation algorithms such as linear interpolation and Bézier curve fitting. Its advantages lie in the continuity and completeness of the trajectory: transforming discrete coordinate points into a continuous path reflecting the overall movement trend of the package, solving the problem in traditional systems where discrete coordinates cannot reflect intermediate motion states. This upgrades trajectory analysis from point-based judgment to line-based analysis. The actual trajectory of the motion coordinate path is compared point-by-point with the predicted package motion trajectory (expected trajectory), calculating the trajectory tracking error, such as the position difference at a given moment and the root mean square value of the overall path deviation. Its advantages also lie in the quantification and precise positioning of errors: by quantifying the deviation between the actual and expected trajectories, the magnitude, direction, and distribution of the error are clarified (e.g., smaller error in the first half and larger error in the second half), providing a measurable optimization target for subsequent parameter compensation. This solves the ambiguity problem in traditional systems where errors rely solely on subjective judgment, such as apparent deviations.

[0028] In one embodiment, the step of generating a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtaining a second action parameter based on the compensation adjustment coefficient, includes: S501. Obtain the current thrust vector, current angular component, and current duration of action based on the first action parameter; S502. Obtain the position deviation component and angle deviation component based on the current trajectory tracking error; S503. Obtain the corresponding average position deviation for each of the position deviation components, and obtain the position deviation coefficient for each of the average position deviations, wherein the calculation formula is: = ; in This represents the position deviation coefficient of the i-th position. This represents the i-th positional deviation component. Indicates the average positional deviation. Indicates the position deviation coefficient; The position deviation coefficient is an indicator used to measure the degree of deviation of the i-th position deviation component from the average position deviation. For example, in a logistics sorting scenario, if the ratio of the position deviation component of a package to the average position deviation is large, it indicates that the position deviation of the package is significant and may require special handling. The position deviation component refers to the deviation of the i-th point or object from a reference position in a specific measurement or calculation. For example, when measuring the position of multiple points on a straight line, the perpendicular distance of each point to the line is the position deviation component of that point. The average position deviation is the average of all position deviation components, reflecting the overall position deviation level. For example, the average position deviation can be obtained by calculating the average deviation of each data point from the average value in a set of data.

[0029] S504. Obtain the corresponding average angle deviation for each angle deviation component, and obtain the angle deviation coefficient for each average angle deviation, wherein the calculation formula is: = in, This represents the deviation coefficient of the j-th angle. This represents the j-th angular deviation component. Indicates the average angular deviation. Indicates the angular deviation coefficient; The angle deviation coefficient is an indicator used to measure the degree of deviation of the j-th angle deviation component from the average angle deviation. For example, in machining, if the ratio of the angle deviation component of a part to the average angle deviation is large, it indicates that the angle deviation of the part is significant and may require remachining. The angle deviation component refers to the deviation of the j-th point or object from a reference angle in a specific measurement or calculation. For example, when measuring the interior angles of a polygon, the difference between each interior angle and the standard angle is the angle deviation component of that angle. The average angle deviation is the average of all angle deviation components, reflecting the overall angle deviation level. j is an index variable used to identify a specific angle measurement point or object.

[0030] S505. Obtain the corresponding weighted normalization coefficient based on each of the position deviation coefficients and angle deviation coefficients, wherein the calculation formula is: = in, Represents the weight normalization coefficient. This represents the deviation coefficient of the j-th angle. This represents the position deviation coefficient of the i-th position. This represents the summation of all positional deviation coefficients and angular deviation coefficients. This represents the position deviation coefficient. Indicates the angular deviation coefficient; in, This represents the weight normalization coefficient. It is obtained by weighted summation and normalization of the position deviation coefficient and angle deviation coefficient. The weight normalization coefficient is used in multi-sorting gate collaborative control to comprehensively consider the impact of position and angle deviations, and to assign weights to different sorting gates or objects to optimize sorting strategies. For example, in a logistics sorting system, packages with larger position and angle deviations may be assigned higher weights for priority processing. This step involves summing all position and angle deviation coefficients. The purpose of this step is to use the sum of position and angle deviations as a normalization benchmark, ensuring that the weighted normalization coefficients range from 0 to 1 for easier comparison and analysis. For example, when calculating the weighted normalization coefficients for multiple packages, the position and angle deviation coefficients of all packages are summed. Then, the sum of the position and angle deviation coefficients of each package is divided by this sum to obtain the weighted normalization coefficient for that package.

[0031] S506. Obtain a position deviation correction coefficient based on each of the weight normalization coefficients and position deviation coefficients, and obtain an angle deviation correction coefficient based on each of the weight normalization coefficients and angle deviation coefficients; S507. Obtain the basic compensation amount based on the thrust vector and angle component in the first action parameter, wherein the calculation formula is: = ; in, Indicates the basic compensation amount. Represents the thrust vector components. Represents angular components; in, This represents the basic compensation amount. It is obtained by multiplying the cosine of the thrust vector component and the angle component. The basic compensation amount is used in sorting systems to compensate for the movement of packages, correcting their positional or angular deviations. For example, in a logistics sorting system, the basic compensation amount is calculated based on the positional and angular deviations of the packages, and then the thrust or speed of the sorting equipment is adjusted to compensate for the packages, enabling them to accurately reach their target positions. This is the thrust vector component, which represents the magnitude of the thrust in a certain direction. In sorting systems, the thrust vector component can be used to describe the magnitude and direction of the force exerted on packages by sorting equipment (such as robotic arms and conveyor belts). For example, in a conveyor belt sorting system, Fk can represent the magnitude of the driving force of the conveyor belt in a certain direction, which can be adjusted... It can change the speed and direction of the package's movement. This is the angular component, which represents the angle between the thrust vector and a reference direction. The angular component is used in sorting systems to describe the direction of the thrust. For example, in a robotic arm sorting system... It can represent the angle between the direction of movement of the robotic arm and the horizontal direction, and can be adjusted... The direction of the robotic arm's movement can be changed, thereby enabling accurate grasping and placement of packages.

[0032] S508. Obtain the compensation adjustment coefficient based on each of the basic compensation amounts, position deviation correction coefficients, and angle deviation correction coefficients, wherein the calculation formula is: = *( ); in, This represents the compensation adjustment coefficient. Indicates the basic compensation amount. This represents the position deviation correction factor. This represents the angle deviation correction factor; in, The compensation adjustment coefficient is expressed as a result of the basic compensation amount. With position deviation correction factor and angle deviation correction factor The compensation adjustment coefficient is obtained by multiplying the sums. It is used to adjust the primary action parameter in a sorting system to more precisely control the movement of packages. For example, in a logistics sorting system, the compensation adjustment coefficient is calculated based on the position and angular deviation of the package, and then parameters such as the thrust, angle, or duration of the sorting equipment are adjusted to achieve accurate sorting of the package. It is the position deviation correction coefficient, which is obtained by normalizing the weights. With position deviation coefficient The result is obtained by multiplication. The position deviation correction factor is used to correct for position deviations, more accurately reflecting the importance of different position deviations in the overall deviation. For example, in a logistics sorting system, packages with large position deviations and high weighting normalization coefficients will also have larger position deviation correction factors, indicating that the position deviation of this package is more important in the overall deviation and requires special attention and handling. It is the angle deviation correction coefficient, which is obtained by normalizing the weights. With angle deviation coefficient The angle deviation correction coefficient is obtained by multiplication. It is used to correct for angle deviations to more accurately reflect the importance of different angle deviations in the overall deviation. For example, in a logistics sorting system, packages with large angle deviations and high weighting normalization coefficients will also have larger angle deviation correction coefficients, indicating that the angle deviation of this package is more important in the overall deviation and requires special attention and handling. This formula is obtained by multiplying the basic compensation amount. With position deviation correction factor and angle deviation correction factor Multiplying the sums, we get the compensation adjustment coefficient. Compensation adjustment coefficients can be used to adjust the primary action parameters in a sorting system to more precisely control the movement of packages, thereby improving sorting accuracy and efficiency. For example, in a logistics sorting system, compensation adjustment coefficients are calculated based on the position and angular deviations of packages. Then, by adjusting parameters such as the thrust, angle, or duration of the sorting equipment, accurate sorting of packages can be achieved. S509. The first action parameter is adjusted a second time according to each of the compensation adjustment coefficients to obtain the second action parameter, wherein the calculation formula is: = + ; in, Indicates the second active parameter. This represents the compensation adjustment coefficient. Indicates the first active parameter; in, This refers to the second action parameter, which is obtained by adjusting the first action parameter. It is used to more precisely control the movement of packages in the sorting system. For example, in a logistics sorting system, the second action parameter could be an adjusted thrust, angle, or duration, to adapt to the sorting requirements of different packages. This refers to the first action parameter, which is the initial control parameter in the sorting system, such as thrust, angle, or duration. These parameters are preset based on the basic information of the package and the sorting requirements, and are used to control the actions of the sorting equipment. For example, in a robotic arm sorting system, the first action parameter could be the initial thrust, angle of movement, and duration of the robotic arm. This represents the sum of adjustments to the first action parameter, where, These are the weights or coefficients for each adjustment amount, used to adjust different parts of the primary action parameter. For example, in a logistics sorting system, It can be an adjustment coefficient calculated based on factors such as the positional deviation and angular deviation of the package. By summing these coefficients, the total adjustment amount to the first action parameter can be obtained.

[0033] As described in steps S501-S509 above, this invention extracts the current thrust vector, including its magnitude and direction (e.g., 8N horizontally to the right), the current angular component (the angle between the actuator and the package, such as the robotic arm's grasping angle of 25°), and the current duration of the actuator's force application (e.g., 0.6 seconds) from the first action parameter. Its advantages lie in the clarity and traceability of the parameter states: by accurately extracting the core physical quantities from the first action parameter, the abstract action parameter is transformed into measurable and analyzable specific values, providing a reference coordinate for subsequent error compensation. This decomposition breaks the limitation of the traditional system where the "first action parameter is treated as a black box," making the current state of each parameter clearly visible and ensuring that subsequent adjustments are "based on evidence." Position deviation components (e.g., X-axis offset 4cm, Y-axis offset 2cm) and angular deviation components (e.g., the package rotates 8° around the Z-axis, tilt angle 5°) are extracted from the trajectory tracking error. Its advantages lie in the structured decomposition of errors: breaking down the general trajectory tracking error into specific deviations in two dimensions, position and angle, clarifying whether the error space is a positional shift or an attitude tilt, solving the problem of fuzzy error description in traditional systems, and providing precise targets for subsequent targeted compensation. Then, the average position deviation is calculated for the position deviation component, and a position deviation coefficient is generated based on the ratio of the average deviation to an allowable threshold, such as 2cm. Its advantages also lie in the statistical representation of position errors: averaging the deviation eliminates the randomness of single measurements, such as jumps caused by instantaneous sensor errors, and quantifies the severity of the error through the deviation coefficient, upgrading the position error from an instantaneous value to a statistical value, providing a stable quantitative basis for compensation. Then, the average angle deviation is calculated for the angle deviation component, and an angle deviation coefficient is generated based on the ratio of the average deviation to an allowable threshold. Finally, its advantages lie in the statistical quantification of angle errors: averaging eliminates instantaneous fluctuations, such as angle jumps caused by package vibration, and clarifies the severity of the error through the coefficient, transforming the angle error from a random value into a decisionable statistical quantity, ensuring the stability and targeting of angle compensation. Based on the position deviation coefficient and the angle deviation coefficient, a normalization algorithm is used to generate weighted normalization coefficients.Its advantage lies in the quantitative balance of the impact of deviations: by clearly defining the contribution ratio of position deviation and angle deviation to the total error through weight allocation, it avoids the overall error optimization imbalance caused by overcompensation for a single deviation, such as focusing only on position and ignoring angle. It ensures that compensation resources, such as thrust adjustment and angle correction, are reasonably allocated according to their degree of impact. The effectiveness of the position deviation coefficient and angle deviation coefficient is verified, outlier coefficients are eliminated, and the weights are calculated using linear normalization formulas: Position weight = Position deviation coefficient / (Position deviation coefficient + Angle deviation coefficient), Angle weight = Angle deviation coefficient / (Position deviation coefficient + Angle deviation coefficient), ensuring that position weight + angle weight = 1, thus realizing the balance of the impact of deviations. The quantitative allocation of the impact is based on the package type, with a preset sensitivity factor (e.g., the angle sensitivity factor of a precision instrument is 1.2, and the weight of the angle deviation coefficient is amplified), and the weight is corrected (e.g., position weight = position weight × 1.2 / (position weight + angle weight × 1.2)) to make the weight adapt to the needs of the scenario. The weight normalization coefficient (e.g., position weight = 0.6, angle weight = 0.4) is multiplied by the position deviation coefficient (position deviation coefficient = 2) and the angle deviation coefficient (angle deviation coefficient = 1.5) respectively to generate the position deviation correction coefficient (e.g., 0.57 × 2 = 1.14) and the angle deviation correction coefficient (e.g., 0.43 × 1.5 = 0.65). Its advantages lie in the precise quantification of compensation intensity: the correction coefficient reflects both the severity of the deviation (the larger the coefficient, the more compensation is needed) and its contribution to the total error (the higher the weight, the larger the correction coefficient). This provides a clear basis for calculating the compensation amount, such as thrust adjustment = basic compensation × position correction coefficient, avoiding insufficient or excessive compensation. Based on the thrust vector and angle component in the first action parameter, the basic compensation amount is calculated through a physical model, such as effective thrust = thrust × cos(angle deviation). Its advantages also lie in the physical correlation of compensation: the basic compensation amount is not set arbitrarily, but is based on the actual physical relationship between thrust and angle. For example, angle deviation reduces the effective component of thrust, ensuring that the compensation amount conforms to the laws of mechanics. For example, the larger the angle deviation, the larger the basic compensation amount, avoiding conflicts between compensation and physical laws (e.g., a 10° angle deviation reducing thrust). The basic compensation amount is multiplied by the correction coefficient to generate the compensation adjustment coefficient. Its advantage lies in the precise progression of compensation: the compensation adjustment coefficient = basic compensation amount × correction coefficient, which not only retains the physical rationality of the basic compensation and conforms to the laws of mechanics, but also adds the degree of influence of the deviation through the correction coefficient. The more serious the deviation, the larger the coefficient, so that the compensation is upgraded from basic correction to precise optimization, ensuring that the compensation amount is matched with the severity of the error and the physical laws. The compensation adjustment coefficient is applied to the first action parameter to make secondary adjustments to the thrust vector and angle components to generate the second action parameter.Its advantage lies in the closed-loop optimization of parameters: the second action parameter is the result of secondary optimization of the first action parameter after error analysis, compensation calculation and physical verification. It not only corrects the deficiencies of the first parameter, such as insufficient thrust and angle deviation, but also retains its reasonable parts, such as the initial thrust that adapts to the package quality. It realizes the progression from preliminary adjustment to precise optimization, and solves the problem of insufficient precision in traditional systems where adjustment ends after one adjustment. The thrust adjustment coefficient is superimposed with the thrust vector of the first action parameter, and the angle adjustment coefficient is superimposed with the angle component. The action time is adjusted according to the thrust and angle. The package movement trajectory is simulated based on the second action parameter and compared with the predicted trajectory. If the residual error is ≤ the allowable threshold, the parameter is effective; otherwise, the compensation adjustment coefficient is recalculated, such as increasing the thrust adjustment coefficient or the angle adjustment coefficient. For multi-actuator systems, the second action parameters of each actuator are adjusted synchronously, such as increasing the thrust of the front push plate by 2N and correcting the angle of the rear push plate by 3°, to ensure that the package trajectory is continuous, such as the smooth tangent of the front and rear trajectories.

[0034] In one embodiment, the steps of acquiring the real-time movement trajectory of the package, obtaining the change in the package friction coefficient based on the real-time movement trajectory monitored by the pressure sensor, and calculating the dynamic response coefficient of the sorting actuator based on the change in the package friction coefficient include: S601. Obtain multiple real-time moving coordinates of the package based on the real-time moving trajectory of the package; obtain multiple moving time sequences by differential calculation based on the multiple real-time moving coordinates of the package; and obtain the corresponding motion speed sequence based on the multiple moving time sequences. S602. Obtain the material of the outer packaging of the package, and obtain the corresponding basic coefficient of friction based on the material of the outer packaging of the package; S603. Obtain the normal pressure and tangential friction force of the target package based on the pressure data; S604. Obtain the material's adaptive friction coefficient based on the motion velocity sequence, normal pressure, and tangential friction force, wherein the calculation formula is: = *v(t); in, This indicates the material's adaptive coefficient of friction. This represents tangential friction. Let v(t) represent the normal pressure and v(t) represent the velocity sequence. The damping coefficient is represented by t, and time is represented by t. in This represents the adaptive friction coefficient. It is a time-varying quantity used to describe the relationship between the frictional force and normal pressure experienced by an object during motion, taking into account the effect of velocity. For example, in logistics sorting systems, the adaptive friction coefficient can be used to more accurately simulate the movement of packages on a conveyor belt, thereby optimizing sorting strategies. It is tangential friction, measured by a pressure sensor. Tangential friction refers to the frictional force experienced by an object in the direction of its motion; it is opposite to the direction of motion. For example, in a conveyor belt sorting system... It can represent the frictional force experienced by the package as it moves on the conveyor belt, and the magnitude of this force can be measured in real time using a pressure sensor. This is normal pressure, also known as normal force. It refers to the pressure exerted on an object perpendicular to the direction of motion, usually caused by the object's weight and other external forces. For example, in a conveyor belt sorting system... The pressure exerted by the package on the conveyor belt can be represented by the force exerted by the conveyor belt, which is equal in magnitude and opposite in direction to the supporting force exerted by the conveyor belt on the package. k is the damping coefficient, a constant that needs to be determined through calibration. The damping coefficient describes the degree to which speed affects friction. For example, in a logistics sorting system, different conveyor belt materials and package types may result in different damping coefficients. v(t) is the absolute value of the object's velocity, representing the instantaneous speed of the object during its movement. For example, in a conveyor belt sorting system, v(t) can represent the speed at which the package moves on the conveyor belt, and this speed can be measured in real time using a speed sensor.

[0035] S605. Obtain multiple material adaptive friction coefficients within a preset time period, and generate a material adaptive friction dynamic change curve based on the multiple material adaptive friction coefficients. S606. Extract multiple material adaptive friction dynamic peak values ​​from the material adaptive friction dynamic change curve, and obtain the time-varying friction coefficient change based on the multiple material adaptive friction dynamic peak values ​​and the basic friction coefficient. The calculation formula is as follows: = G; in, This represents the change in the time-varying coefficient of friction. This represents the maximum value of the adaptive friction coefficient over a period of time, while G represents the basic friction coefficient. in, This represents the time-varying change in the coefficient of friction. It is obtained by extracting the difference between the peak value and the reference coefficient of friction from the dynamic change curve. The time-varying coefficient of friction describes how the coefficient of friction changes over time, reflecting the dynamic changes in frictional force during the movement of an object. For example, in a logistics sorting system, the time-varying coefficient of friction can be used to monitor the wear of conveyor belts; when the change exceeds a certain threshold, it indicates that the conveyor belt needs to be replaced. It is an adaptive friction coefficient The maximum value over a period of time. The adaptive coefficient of friction is a time-varying quantity used to describe the relationship between the frictional force and normal pressure experienced by an object during motion, taking into account the effect of velocity. For example, in a conveyor belt sorting system, The coefficient of friction (G) represents the ratio of the maximum frictional force to the normal pressure experienced by a package moving on a conveyor belt, plus the effect of velocity. It is the reference friction coefficient, which is the coefficient of friction of an object under ideal conditions (such as at rest or in uniform motion), and is usually a pre-defined constant. The reference friction coefficient is used as a reference and compared with the actual measured maximum adaptive friction coefficient to calculate the time-varying change in the friction coefficient. For example, in a logistics sorting system, the reference friction coefficient can be pre-set based on factors such as the material of the conveyor belt and the package. S607. A dynamic response coefficient is generated based on the basic friction coefficient and the time-varying friction coefficient change, wherein the calculation formula is: = + *G; in, Represents the dynamic response coefficient. This represents the change in the time-varying coefficient of friction. Indicates the basic coefficient of friction. The weighting coefficient of the time-varying friction coefficient change The weighting coefficient represents the basic friction coefficient; in This represents the dynamic response coefficient. It is a parameter that comprehensively considers the material type and the time-varying change in the coefficient of friction, used in sorting systems to describe the degree of response of an object to external forces. For example, in logistics sorting systems, the dynamic response coefficient can be used to adjust the action parameters of sorting equipment to adapt to the sorting requirements of packages made of different materials. The time-varying coefficient of friction is the change in friction coefficient over time. It is obtained by extracting the difference between the peak value and the reference friction coefficient from the dynamic change curve. The time-varying coefficient of friction describes how the friction coefficient changes over time, reflecting the dynamic characteristics of frictional force during an object's motion. For example, in a logistics sorting system, the time-varying coefficient of friction can be used to monitor the wear of conveyor belts. When the change exceeds a certain threshold, it indicates that the conveyor belt needs to be replaced. G represents the reference friction coefficient, which is the friction coefficient of an object under ideal conditions (such as at rest, in uniform motion, or in an undisturbed environment), and is usually a pre-calibrated constant. In a logistics sorting system, it can serve as a reference benchmark to measure the deviation between the actual friction coefficient and the ideal state, helping to determine whether the condition of conveyor belts, packages, etc., is normal. This is the weighting coefficient for the change in the time-varying coefficient of friction. It is used to adjust the contribution of the change in the time-varying coefficient of friction to the dynamic response coefficient. For example, in a logistics sorting system, if the change in the time-varying coefficient of friction has a significant impact on the sorting effect, the weighting coefficient can be increased. The value of is adjusted to improve the sensitivity of the dynamic response coefficient to changes in the time-varying friction coefficient. The weighting coefficient of the base friction coefficient. It is used to adjust the degree of influence of the base friction coefficient on the dynamic response coefficient. Different materials have different base friction coefficients; by setting an appropriate weighting coefficient... This value allows the dynamic response coefficient to more accurately adapt to the sorting needs of objects made of different materials. For example, a smaller value can be set for plastic packages. Value; for cardboard box packaging, a larger value can be set. Values ​​(such as plastics) =0.8, cardboard box =1.2).

[0036] As described in steps S601-S607 above, this invention extracts multiple real-time moving coordinates from the real-time movement trajectory of the package (e.g., recording X and Y coordinates every 0.1 seconds), transforms the coordinate sequence into a movement time sequence (time points corresponding to the coordinates) through differential calculation, and then calculates the motion speed sequence (e.g., v1=0.8m / s, v2=0.9m / s, v3=0.7m / s) through the rate of change of the time sequence and coordinates. Its advantages lie in the dynamic real-time nature and accuracy of the speed: through differential calculation of continuous coordinates, it captures the instantaneous changes in the package's movement speed, such as acceleration, deceleration, and constant speed switching, avoiding the limitations of traditional systems based on fixed speed presets or single-point speed sampling. Through the chain derivation of coordinates, time, and speed, speed is upgraded from a static assumption to a dynamic measurement, providing a precise kinematic basis for subsequent friction coefficient calculation. By identifying the outer packaging material of the package, such as cardboard boxes, plastic bags, foam boxes, and cloth bags, and retrieving the corresponding basic friction coefficients from a preset material friction coefficient library, such as the basic friction coefficient of 0.6 for cardboard boxes and conveyor belts, and 0.3 for plastic bags. Its advantages lie in the material adaptability of the friction coefficient: different materials have vastly different surface roughness and hardness, such as the rough surface of a foam box and the smooth surface of a plastic bag. The basic friction coefficient is retrieved based on the material characteristics, avoiding the coarse setting of a uniform friction coefficient such as 0.5 in traditional systems. This provides a material baseline for subsequent friction coefficient calculations. Furthermore, the material friction coefficient library is built based on historical material friction values. The corresponding friction coefficient is extracted from the material friction coefficient library. Pressure sensors installed on the conveyor belt surface collect the pressure data of the package on the conveyor belt, from which the normal pressure (the pressure perpendicular to the conveyor belt surface, i.e., the weight component of the package, FN=mg×cosθ, where θ is the conveyor belt inclination angle) and tangential friction force (the force parallel to the conveyor belt surface, i.e., the force that opposes the relative motion of the package, Ff) are separated. Its advantage lies in the precise separation of force components: normal pressure is the basis for calculating friction (friction force = friction coefficient × normal pressure), while tangential friction directly reflects the actual interaction between the package and the conveyor belt. The separation of the two avoids the error in friction calculation caused by mixing pressure data in traditional systems, such as including tangential force in normal force. The motion speed sequence reflects the motion state of the package, normal pressure, and tangential friction. The material adaptive friction coefficient is calculated using the friction formula (friction coefficient = tangential friction / normal pressure) to reflect the actual friction coefficient between the package and the conveyor belt in the current state.Its advantages lie in the real-time and dynamic nature of the friction coefficient: the value is not a fixed base friction coefficient, but is dynamically adjusted according to the movement state of the package, such as the frictional heat generated by changes in speed, which reduces the friction coefficient of the plastic surface, and the pressure changes, such as the increase in normal pressure caused by stacking, which slightly increases the friction coefficient. This solves the static problem of the base friction coefficient being unchanging in traditional systems. Within a preset time period, such as 10 seconds, it covers the entire cycle of the package from entering the sorting area to leaving, and collects multiple material adaptive friction coefficients, such as a value every 0.5 seconds. These values ​​are arranged in chronological order, and a dynamic change curve of material adaptive friction is plotted, with time on the horizontal axis and friction coefficient on the vertical axis. Its advantage lies in the visualization of the trend of friction coefficient change: the curve can intuitively reflect the fluctuation law of friction coefficient, such as gradual increase, sudden decrease, and periodic fluctuation, avoiding the limitation of traditional systems that only focus on the friction coefficient value at a single point, such as μ=0.4 at a certain moment, while ignoring the overall trend of change, such as μ decreasing from 0.6 to 0.4 in the first 5 seconds, indicating that the friction coefficient is continuously decreasing. In the material adaptive friction dynamic change curve, the material adaptive friction dynamic peak value is extracted to reflect the extreme change of friction coefficient. By the difference between the peak value and the basic friction coefficient, the time-varying friction coefficient change is calculated (e.g., the difference between the peak value of 0.7 and the basic value of 0.5 is +0.2, that is, the change is +0.2). Its advantage lies in the quantitative capture of extreme changes: the peak value of the friction coefficient often corresponds to key anomalies in the sorting process, such as the highest point being due to package jamming, and the lowest point being due to slippage. The calculation of the change can accurately quantify the impact of these extreme situations on sorting, avoiding the inadequacy of traditional systems that only use the average change (such as 0.5→0.6, average change of 0.1) and ignore the peak value (such as instantaneous 0.8). Combined with the type of packaging material of the package, such as cardboard box, plastic bag and time-varying friction coefficient change (such as +0.2, -0.3), a dynamic response coefficient is generated to reflect the adjustment range that the sorting actuator needs to adjust, such as the thrust adjustment coefficient and the pressure adjustment coefficient. Its advantages are: adaptability to material changes and sensitivity to changes. Different materials are sensitive to changes in the coefficient of friction. For example, cloth bags are more sensitive to increased friction and are more prone to wear; plastic bags are more sensitive to decreased friction and are more prone to slipping. The dynamic response coefficient is set differently according to the material characteristics and the magnitude of the change. For example, the response coefficient of plastic bags is larger for negative changes. This avoids the inefficiency of the traditional system's uniform response strategy, such as adjusting the thrust by 20% regardless of the material and a change of 0.2.

[0037] In one embodiment, the step of generating collaborative control instructions for multiple sorting ports based on the second action parameter and the dynamic response coefficient includes: S701. Obtain the preset target position of the corresponding sorting port according to each of the second action parameters, and obtain the package trajectory correction amount according to each preset target position and dynamic response coefficient, wherein the calculation formula is: - ) in, This indicates the amount of correction to the package's trajectory. Represents the dynamic response coefficient. Indicates the first The target location coordinates of each sub-interface Indicates the first The current package location coordinates for each sub-interface; in, This represents the trajectory correction amount for the i-th sorting interface. It is obtained by multiplying the dynamic response coefficient λ by the difference between the target position coordinates and the current package position coordinates. The trajectory correction amount is used to correct the movement trajectory of packages in a sorting system to ensure that packages accurately reach their target locations. For example, in a logistics sorting system, the trajectory correction amount is calculated based on the current and target positions of the packages. Then, by adjusting the actions of the sorting equipment, such as changing the speed of the conveyor belt or the movement trajectory of the robotic arm, accurate sorting of packages is achieved. The dynamic response coefficient is derived by comprehensively considering the material type and the time-varying change in the coefficient of friction. It describes the degree to which an object responds to external forces. In sorting systems, it can be used to adjust the operational parameters of sorting equipment to adapt to the sorting needs of packages made of different materials. For example, in a logistics sorting system, the dynamic response coefficient is calculated based on the package's material and the change in its coefficient of friction. Then, by adjusting parameters such as the pushing force, angle, or duration of the sorting equipment, accurate sorting of packages can be achieved. It is the first The target location coordinates of each sorting interface. These target location coordinates are the positions the packages need to reach within the sorting system, typically preset based on sorting requirements and system layout. For example, in a logistics sorting system, each sorting interface corresponds to a specific target location coordinate; packages need to be accurately sorted to their corresponding interface. It is the first The current package location coordinates are displayed at each interface. These coordinates represent the package's real-time position within the sorting system, obtained through real-time monitoring by sensors. For example, in a logistics sorting system, sensors such as cameras and laser rangefinders can acquire the package's location coordinates in real time to calculate trajectory corrections and adjust the sorting equipment's actions.

[0038] S702. Determine whether the deviation between the current package position and the preset target position at each sorting point exceeds a preset threshold range. S703. If the current package position deviation at the sorting station exceeds a preset threshold range, the package trajectory is determined to be abnormal. The calculation formula is as follows: abnormal ; Here, "abnormal" represents the judgment result, with a value of "yes" or "no". It is used to indicate whether there is an abnormality in the position of the current package at the i-th sorting interface. For example, in a logistics sorting system, if the judgment result is "yes", it means that the position deviation of the package exceeds the allowable range, which may require triggering an alarm or adjusting the sorting strategy. It is obtained through real-time monitoring using sensors (such as cameras, laser rangefinders, etc.). This represents the current package location coordinates of the i-th sub-interface. It is a fixed threshold range used to determine whether the package position deviation exceeds the allowable range. It is a constant preset based on the accuracy requirements and actual conditions of the sorting system. For example, in a logistics sorting system, if the system's sorting accuracy requirements are high, it can... Set it to a smaller value (e.g., 4.5-5cm); if lower precision is required, set it to a larger value (e.g., 13-15cm).

[0039] S704. Adjust the dynamic response coefficient until the package position deviation is within the preset threshold range, and determine the response coefficient at this point as the final response coefficient. The calculation formula is as follows: ; in, This represents the final dynamic response coefficient, which is the value of the original dynamic response coefficient after an anomaly is detected. The parameters are obtained by scaling up the scale and are used to further adjust the operating parameters of the sorting equipment. This represents the original dynamic response coefficient, obtained by comprehensively considering the material type and the time-varying coefficient of friction, and is used to describe the degree of response of an object to external forces. The trajectory correction amount for the i-th sub-interface is represented by the dynamic response coefficient. The value is obtained by multiplying the difference between the target location coordinates and the current package location coordinates, and is used to correct the package's movement trajectory. This represents a fixed threshold, similar to the one in the anomaly detection formula. It is used to measure the severity of an abnormal situation.

[0040] S705. Obtain the path allocation sensitivity of the corresponding sorting port based on each final response coefficient and trajectory correction amount, wherein the calculation formula is: ; in, This represents the path allocation sensitivity of the i-th sub-interface. Path allocation sensitivity measures the amount of trajectory correction required by the sub-interface under a unit final response coefficient, reflecting the sub-interface's sensitivity to changes in the response coefficient and its response efficiency. For example, in a logistics sorting system, a sub-interface with higher path allocation sensitivity will have a larger change in trajectory correction for the same change in the final response coefficient, indicating that the sub-interface is more sensitive to adjustments in the response coefficient and can respond more quickly to changes in sorting requirements.

[0041] S706. Based on the path allocation sensitivity, final response coefficient, and trajectory correction amount, a collaborative control command for multiple sorting ports is generated, wherein the calculation formula is: ZL ; ZL stands for Coordination Control Command, which coordinates the actions of multiple sorting interfaces or devices to achieve accurate sorting and path allocation of packages. It is the first The target location coordinates of each sorting interface. These target location coordinates are the positions the packages need to reach within the sorting system, typically preset based on sorting requirements and system layout. For example, in a logistics sorting system, each sorting interface corresponds to a specific target location coordinate; packages need to be accurately sorted to their corresponding interface. This part relates to the path assignment sensitivity formula mentioned earlier. It represents the ratio of trajectory correction to the final dynamic response coefficient, used to measure the interface's sensitivity to changes in the response coefficient and its response efficiency. This represents the path assignment sensitivity of the i-th sub-interface, which measures the amount of trajectory correction required by the sub-interface under a unit final response coefficient. It reflects the sub-interface's sensitivity to changes in the response coefficient and its response efficiency.

[0042] As described in steps S701-S706 above, this invention combines physical properties such as package weight and size with a dynamic response coefficient to reflect the influence of frictional characteristics on motion, calculating the preset target position and trajectory correction amount for each sorting point. Its core advantage lies in combining physical characteristics with dynamic response to achieve personalized trajectory planning. The dynamic response coefficient considers the influence of factors such as package material and surface condition on the motion trajectory, making the preset target position no longer a fixed value but dynamically adjusted according to real-time frictional characteristics. This method can adapt to the differentiated motion needs of different types of packages in the same sorting system, improving the accuracy of trajectory planning. By comparing the deviation between the current package position and the preset target position with a preset threshold range, real-time monitoring of the package's motion status is achieved. The preset threshold range considers system errors and normal fluctuations, distinguishing between normal phenomena with acceptable slight deviations and abnormal situations requiring adjustment due to severe deviations. This hierarchical judgment mechanism avoids adjusting all deviations, reducing unnecessary system actions and improving control efficiency. When the position deviation exceeds the threshold, the abnormal package trajectory is accurately determined, providing a basis for subsequent adjustments. Timely and accurate anomaly detection is crucial for ensuring the efficient operation of the sorting system. Through clearly defined threshold standards and real-time monitoring, the system can quickly identify packages deviating from their normal trajectories, preventing them from having a cascading impact on subsequent sorting processes. By iteratively adjusting the dynamic response coefficient, the package position deviation is gradually reduced to within a preset threshold range, and the final stable response coefficient is used as the control basis. This closed-loop feedback control mechanism can adaptively compensate for system errors and external disturbances, ensuring that the package movement trajectory remains within an acceptable range. Compared to open-loop control, closed-loop feedback control has stronger robustness and adaptability, capable of handling various uncertainties. Path allocation sensitivity is calculated based on the final response coefficient and trajectory correction amount, quantifying the sensitivity of each sorting station to package path adjustments. Path allocation sensitivity reflects the matching relationship between the adjustment capabilities of sorting actuators such as diverters and robotic arms and the physical characteristics of the packages, enabling the system to optimize the allocation of sorting resources for different package types and movement states. For example, for lightweight and fragile packages, highly sensitive sorting stations can be assigned. Precise path control can be achieved through minor adjustments, preventing damage caused by excessive adjustments. The path allocation sensitivity, final response coefficient, and trajectory correction amount are integrated to generate collaborative control instructions for multiple sorting stations. This integrated control strategy considers the interaction between multiple factors, achieving globally optimized sorting scheduling. Through collaborative control, different sorting stations can coordinate with each other, avoiding conflicts and resource competition, ensuring smooth package movement throughout the sorting system.

[0043] like Figure 2 As shown, this application also provides a machine vision-based package detection and sorting system, including a sorting actuator, and further comprising: The first acquisition module 1 acquires real-time video information and historical package movement trajectory data, and uses the real-time video information and historical package movement trajectory data to obtain the package posture offset during the prediction period. The second acquisition module 2 acquires the initial parameters of the sorting executor and acquires the first action parameter based on the initial parameters and the package posture offset during the predicted time period. The first correction module 3 corrects the package's motion trajectory according to the first action parameter to obtain the corresponding motion coordinate path, and calculates the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory. The first generation module 4 generates a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtains the second action parameter based on the compensation adjustment coefficient. The third acquisition module 5 acquires the real-time movement trajectory of the package, obtains the change in the friction coefficient of the package based on the real-time movement trajectory monitored by the pressure sensor, and calculates the dynamic response coefficient of the sorting actuator based on the change in the friction coefficient of the package. The second generation module 6 generates collaborative control instructions for multiple sorting ports based on the second action parameter and dynamic response coefficient, and performs dynamic allocation of package paths based on the collaborative control instructions.

[0044] As a preferred embodiment, the first acquisition module includes: The first extraction unit extracts multiple historical package movement trajectories based on package movement trajectory data from historical sorting scenarios, and obtains the average historical package movement trajectory based on the multiple historical package trajectories. The first acquisition unit splits the real-time video information according to a preset video frame sequence length to obtain multiple video frame sequence lengths, obtains multiple real-time motion coordinates of the corresponding packages according to the multiple video frame sequence lengths, and generates a real-time package motion trajectory according to the multiple real-time motion coordinates of the packages. The second acquisition unit acquires the preset standard package movement trajectory; The third acquisition unit acquires the real-time dynamic offset coefficient of the package posture based on the real-time package posture motion trajectory and the preset standard package motion trajectory. The fourth acquisition unit acquires historical offset benchmarks based on the historical average movement trajectory of packages and the preset standard package movement trajectory; The fifth acquisition unit acquires the package offset for the predicted time period based on the real-time package attitude dynamic offset coefficient and the historical offset benchmark.

[0045] Preferably, the second acquisition module includes: The sixth acquisition unit acquires the initial thrust vector, initial angular components, and initial action duration based on the initial parameters; The seventh acquisition unit acquires the package attitude offset during the prediction period, wherein the attitude offset includes a position offset component and an angle offset component. The eighth acquisition unit acquires the thrust correction amount based on the position offset component and the angle correction amount based on the angle offset component; The ninth acquisition unit acquires the corrected thrust vector based on the initial thrust vector and the thrust correction amount, and acquires the corrected angle component based on the initial angle component and the angle correction amount. The first generation unit generates the first operating parameters of the sorting actuator based on the corrected thrust vector, the corrected angle component, and the initial operating duration.

[0046] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus RAM (RDRAM), direct memory bus RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0047] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0048] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A machine vision-based package detection and sorting method, comprising a sorting actuator, characterized in that, Also includes: Acquire real-time video information and historical package movement trajectory data; use real-time video information and historical package movement trajectory data to obtain the package attitude offset for the prediction period. Obtain the initial parameters of the sorting executor, and obtain the first action parameter based on the initial parameters and the package posture offset during the predicted time period; The corresponding motion coordinate path is obtained by correcting the package's motion trajectory according to the first action parameter, and the trajectory tracking error is calculated based on the motion coordinate path and the predicted package motion trajectory. A compensation adjustment coefficient is generated based on the trajectory tracking error and the first action parameter, and a second action parameter is obtained based on the compensation adjustment coefficient. The real-time movement trajectory of the package is obtained. The change in the coefficient of friction of the package is obtained by monitoring the real-time movement trajectory of the corresponding package based on the pressure sensor. The dynamic response coefficient of the sorting actuator is calculated based on the change in the coefficient of friction of the package. Based on the second action parameter and the dynamic response coefficient, a collaborative control instruction for multiple sorting ports is generated, and dynamic allocation of package paths is performed according to the collaborative control instruction.

2. The package detection and sorting method based on machine vision according to claim 1, characterized in that, The step of obtaining real-time video information and historical package motion trajectory data, and obtaining the package posture offset for the prediction period from the real-time video information and historical package motion trajectory data, includes: Extract multiple historical package movement trajectories from the package movement trajectory data of historical sorting scenarios, and obtain the average historical package movement trajectory based on the multiple historical package trajectories; The real-time video information is split according to a preset video frame sequence length to obtain multiple video frame sequence lengths. Multiple real-time motion coordinates of the corresponding packages are obtained according to the multiple video frame sequence lengths, and a real-time package motion trajectory is generated according to the multiple real-time motion coordinates of the packages. Obtain the preset standard package movement trajectory; The real-time dynamic offset coefficient of the package posture is obtained based on the real-time package posture motion trajectory and the preset standard package motion trajectory. Historical offset benchmarks are obtained based on the average historical parcel movement trajectory and the preset standard parcel movement trajectory. The predicted package offset is obtained based on the real-time dynamic offset coefficient of the package posture and the historical offset benchmark.

3. The machine vision-based package detection and sorting method according to claim 1, characterized in that, The step of obtaining the initial parameters of the sorting actuator and obtaining the first action parameter based on the initial parameters and the package posture offset during the predicted time period includes: The initial thrust vector, initial angular components, and initial duration of action are obtained based on the initial parameters. Obtain the package attitude offset during the prediction period, wherein the attitude offset includes a position offset component and an angle offset component; The thrust correction is obtained based on the position offset component, and the angle correction is obtained based on the angle offset component; The corrected thrust vector is obtained based on the initial thrust vector and the thrust correction amount, and the corrected angle component is obtained based on the initial angle component and the angle correction amount. The first operating parameters of the sorting actuator are generated based on the corrected thrust vector, the corrected angle components, and the initial operating duration.

4. The package detection and sorting method based on machine vision according to claim 1, characterized in that, The step of correcting the package's motion trajectory based on the first action parameter to obtain the corresponding motion coordinate path, and calculating the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory, includes: Obtain the first action parameter, wherein the first action parameter includes a thrust component, an angle component, and an action duration component; The force direction vector of the sorting actuator is obtained based on the thrust component and the angle component. Based on the force direction vector and duration component, a sorting actuator control command is generated; based on the actuator control command, the package movement trajectory is obtained; and based on the package movement trajectory, multiple continuous position coordinates are obtained. Generate a motion coordinate path based on multiple consecutive position coordinates; Obtain the predicted trajectory of the package, and calculate the trajectory tracking error based on the predicted trajectory and the movement coordinate path.

5. The machine vision-based package detection and sorting method according to claim 1, characterized in that, The step of generating a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtaining a second action parameter based on the compensation adjustment coefficient, includes: The current thrust vector, current angular component, and current duration of action are obtained based on the first action parameter. The position deviation component and angle deviation component are obtained based on the current trajectory tracking error; The average position deviation is obtained for each of the position deviation components, and the position deviation coefficient is obtained for each of the average position deviation components. The corresponding average angle deviation is obtained for each of the angle deviation components, and the angle deviation coefficient is obtained for each of the average angle deviation components. Obtain the corresponding weighted normalization coefficient based on each of the position deviation coefficients and angle deviation coefficients; A position deviation correction coefficient is obtained based on each of the weight normalization coefficients and position deviation coefficients; an angle deviation correction coefficient is obtained based on each of the weight normalization coefficients and angle deviation coefficients. The basic compensation amount is obtained based on the thrust vector and angle components in the first action parameter; The compensation adjustment coefficient is obtained based on each of the aforementioned basic compensation amounts, position deviation correction coefficients, and angle deviation correction coefficients; The first action parameter is adjusted twice based on each of the compensation adjustment coefficients to obtain the second action parameter.

6. The package detection and sorting method based on machine vision according to claim 1, characterized in that, The steps of acquiring the real-time movement trajectory of the package, obtaining the change in the package friction coefficient based on the real-time movement trajectory monitored by the pressure sensor, and calculating the dynamic response coefficient of the sorting actuator based on the change in the package friction coefficient include: Multiple real-time movement coordinates of packages are obtained based on the real-time movement trajectory of the packages. Multiple movement time series are obtained by differential calculation based on the multiple real-time movement coordinates of the packages. The corresponding movement speed series are obtained based on the multiple movement time series. Obtain the material of the package's outer packaging, and then determine the corresponding basic coefficient of friction based on that material. Obtain the normal pressure and tangential friction force of the target package based on the pressure data; The material's adaptive friction coefficient is obtained based on the motion velocity sequence, normal pressure, and tangential friction force. Obtain multiple material adaptive friction coefficients within a preset time period, and generate material adaptive friction dynamic change curves based on the multiple material adaptive friction coefficients; Multiple material adaptive friction dynamic peaks are extracted from the material adaptive friction dynamic change curve, and the time-varying friction coefficient change is obtained from the multiple material adaptive friction dynamic peaks and the basic friction coefficient. The dynamic response coefficient is generated based on the basic friction coefficient and the time-varying friction coefficient change.

7. The package detection and sorting method based on machine vision according to claim 1, characterized in that, The step of generating collaborative control instructions for multiple sorting ports based on the second action parameter and the dynamic response coefficient includes: The preset target position of the corresponding sorting port is obtained according to each of the second action parameters, and the package trajectory correction amount is obtained according to each preset target position and dynamic response coefficient; Determine whether the deviation between the current package position and the preset target position at each sorting point exceeds a preset threshold range; If the current package position deviation at the sorting point exceeds a preset threshold range, the package trajectory is determined to be abnormal. Adjust the dynamic response coefficient until the package position deviation is within the preset threshold range, and determine the response coefficient at this point as the final response coefficient; The path allocation sensitivity of the corresponding sorting port is obtained based on each final response coefficient and trajectory correction amount; Based on the path allocation sensitivity, final response coefficient, and trajectory correction amount, collaborative control instructions for multiple sorting ports are generated.

8. The machine vision-based package detection and sorting system according to claim 1, comprising a sorting actuator, characterized in that, Also includes: The first acquisition module acquires real-time video information and historical package movement trajectory data, and uses the real-time video information and historical package movement trajectory data to obtain the package posture offset during the prediction period. The second acquisition module acquires the initial parameters of the sorting executor and acquires the first action parameter based on the initial parameters and the package posture offset during the predicted time period. The first correction module corrects the package's motion trajectory according to the first action parameter to obtain the corresponding motion coordinate path, and calculates the trajectory tracking error based on the motion coordinate path and the predicted package motion trajectory. The first generation module generates a compensation adjustment coefficient based on the trajectory tracking error and the first action parameter, and obtains the second action parameter based on the compensation adjustment coefficient. The third acquisition module acquires the real-time movement trajectory of the package, obtains the change in the friction coefficient of the package based on the real-time movement trajectory monitored by the pressure sensor, and calculates the dynamic response coefficient of the sorting actuator based on the change in the friction coefficient of the package. The second generation module generates collaborative control instructions for multiple sorting ports based on the second action parameter and the dynamic response coefficient, and performs dynamic allocation of package paths according to the collaborative control instructions.

9. The machine vision-based package detection and sorting system according to claim 1, characterized in that, The first acquisition module includes: The first extraction unit extracts multiple historical package movement trajectories based on package movement trajectory data from historical sorting scenarios, and obtains the average historical package movement trajectory based on the multiple historical package trajectories. The first acquisition unit splits the real-time video information according to a preset video frame sequence length to obtain multiple video frame sequence lengths, obtains multiple real-time motion coordinates of the corresponding packages according to the multiple video frame sequence lengths, and generates a real-time package motion trajectory according to the multiple real-time motion coordinates of the packages. The second acquisition unit acquires the preset standard package movement trajectory; The third acquisition unit acquires the real-time dynamic offset coefficient of the package posture based on the real-time package posture motion trajectory and the preset standard package motion trajectory. The fourth acquisition unit acquires historical offset benchmarks based on the historical average movement trajectory of packages and the preset standard package movement trajectory; The fifth acquisition unit acquires the package offset for the predicted time period based on the real-time package attitude dynamic offset coefficient and the historical offset benchmark.

10. The machine vision-based package detection and sorting system according to claim 1, characterized in that, The second acquisition module includes: The sixth acquisition unit acquires the initial thrust vector, initial angular components, and initial action duration based on the initial parameters; The seventh acquisition unit acquires the package attitude offset during the prediction period, wherein the attitude offset includes a position offset component and an angle offset component. The eighth acquisition unit acquires the thrust correction amount based on the position offset component and the angle correction amount based on the angle offset component; The ninth acquisition unit acquires the corrected thrust vector based on the initial thrust vector and the thrust correction amount, and acquires the corrected angle component based on the initial angle component and the angle correction amount. The first generation unit generates the first operating parameters of the sorting actuator based on the corrected thrust vector, the corrected angle component, and the initial operating duration.

Images