AGV environment sensing method and system based on laser beams

Abstract

The invention belongs to the technical field of AGVs, and provides an AGV environment sensing method and system based on laser beams. The method comprises the following steps that: a vehicle-mounted laser radar continuously scans a surrounding environment to obtain a laser point cloud data frame set with a continuous time sequence; a plurality of dynamic targets are obtained based on the laser point cloud data frame set, dynamic characteristic parameters of the dynamic targets are extracted, and light dynamic targets are identified from the dynamic targets; performing motion center analysis on the corresponding light dynamic target based on the motion trail features to obtain a motion center analysis result; and determining a plurality of obstacle targets based on a motion center analysis result, planning a driving track in the surrounding environment based on each obstacle target, and executing. According to the method, the false stop rate of the AGV caused by the interference of flying catkins can be reduced, the safety risk that the AGV and the floating yarn are pulled is effectively avoided, the sensing reliability and the operation intelligence level in a complex spinning environment are improved, and then the accuracy of trajectory planning is improved.

Description

Technical Field

[0001] This invention relates to the field of AGV technology, and more specifically, to an AGV environmental perception method and system based on laser beams. Background Technology

[0002] Automated Guided Vehicles (AGVs) play a crucial role in automatically transferring yarn spindles between different workstations such as spinning machines, winding machines, and warehouses, which is of great significance for improving production efficiency and reducing labor costs.

[0003] The autonomous navigation and obstacle avoidance capabilities of AGVs heavily rely on their environmental perception systems. Currently, LiDAR-based navigation and obstacle avoidance technology has become the mainstream perception solution for AGVs due to its advantages such as high ranging accuracy, wide scanning range, and no need to modify the ground. This technology generates point cloud data of the surrounding environment by emitting laser beams and receiving reflected signals. Algorithms then enable the AGV to perform self-localization, Single-Lane Mapping (SLAM), obstacle detection, and ultimately path planning.

[0004] However, the air in spinning workshops is filled with a large amount of lightweight lint, such as cotton and synthetic fibers. As these lint particles fall, they reflect the laser beams emitted by the AGVs, creating numerous discrete, transient noise points in the point cloud data. Existing conventional obstacle recognition algorithms struggle to effectively distinguish these lint noise points from real, stable obstacles (such as personnel, equipment, and walls), often misclassifying the lint as obstacles. This leads to frequent unnecessary emergency braking or detours by the AGVs, severely impacting the efficiency and continuity of material handling, disrupting production rhythms, and exacerbating wear and tear on AGV mechanical components and unnecessary energy consumption.

[0005] Therefore, how to accurately filter out interference from flying fluff in the environment, accurately identify real obstacles, and improve the accuracy and rationality of AGV path planning are technical problems that urgently need to be solved. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an AGV environmental perception method and system based on laser beams.

[0007] This invention provides an AGV environmental perception method based on laser beams, comprising the following steps: An AGV in a spinning workshop controls an onboard LiDAR to continuously scan the surrounding environment, obtaining a time-series continuous set of laser point cloud data frames; based on the laser point cloud data frame set, several dynamic targets are identified, and dynamic feature parameters of each dynamic target are extracted, including at least the duration of the dynamic target's existence and its motion trajectory characteristics; a lightweight dynamic target is identified from among the dynamic targets based on the dynamic feature parameters; based on the motion trajectory characteristics, a motion center analysis is performed on the corresponding lightweight dynamic target to obtain the motion center analysis result, including whether the lightweight dynamic target is in a motion state based on a constraint center, and the number of constraint centers; based on the motion center analysis result, several obstacle targets are determined, and a driving trajectory is planned in the surrounding environment based on each obstacle target and executed.

[0008] This invention also provides an AGV environmental perception system based on laser beams. The system includes: an onboard lidar configured to continuously scan the surrounding environment of the AGV in a spinning workshop to obtain a time-series continuous set of laser point cloud data frames; a target initial recognition unit configured to: derive several dynamic targets based on the laser point cloud data frame set, and extract dynamic feature parameters of each dynamic target, including at least the duration of the dynamic target's existence and motion trajectory features, and identify lightweight dynamic targets from among the dynamic targets based on the dynamic feature parameters; a target differentiation unit configured to: perform motion center analysis on the corresponding lightweight dynamic targets based on the motion trajectory features, and obtain motion center analysis results, including whether the lightweight dynamic target is in a motion state based on a constraint center, and the number of constraint centers; and a trajectory planning unit configured to: determine several obstacle targets based on the motion center analysis results, plan a driving trajectory in the surrounding environment based on each obstacle target, and execute the plan.

[0009] This invention utilizes time-series analysis of laser point cloud data to extract the motion characteristics of dynamic targets, effectively distinguishing between lightweight flying fluff and fallen yarn in spinning workshops. Furthermore, by introducing motion center analysis, it accurately determines the physical motion constraint state of targets, filtering out unconstrained free-flying fluff and identifying single / double-constrained yarns that may pose an entanglement risk as real obstacles. This invention significantly reduces the false stop rate of AGVs caused by flying fluff interference, ensuring the continuity of material handling and production efficiency. It also effectively avoids the safety risks of AGVs pulling on fallen yarn, improves the reliability of perception and the level of intelligent operation in complex textile environments, and ultimately enhances the accuracy of trajectory planning. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of an AGV environmental perception method based on a laser beam, as disclosed in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the structure of an AGV environmental perception system based on a laser beam, as disclosed in an embodiment of the present invention.

[0012] Figure 3 This is another structural schematic diagram of an AGV environmental perception system based on a laser beam disclosed in an embodiment of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0016] like Figure 1 As shown, this embodiment of the invention discloses an AGV environmental perception method based on laser beams, including the following steps: Step 100, the AGV in the spinning workshop controls the on-board lidar to continuously scan the surrounding environment and obtain a time-series continuous set of laser point cloud data frames.

[0017] In this step, within the specific operating environment of the spinning workshop, the AGV controls its onboard LiDAR sensor to periodically scan the surrounding environment at a fixed scanning frequency (e.g., 10Hz). The environmental distance and contour information acquired in each scan operation constitutes a frame of LiDAR point cloud data. By collecting these multiple frames of point cloud data acquired sequentially over time, the aforementioned time-series continuous LiDAR point cloud data frame set is formed.

[0018] Step 200: Based on the laser point cloud data frame set, several dynamic targets are obtained, and dynamic feature parameters of each dynamic target are extracted, including at least the duration of existence and motion trajectory features of the dynamic target. Lightweight dynamic targets are identified from each dynamic target based on the dynamic feature parameters.

[0019] In this step, each frame of laser point cloud data is preprocessed (e.g., filtered and denoised); and a clustering algorithm, such as Euclidean clustering, is used to aggregate nearby points in space, forming several target point cloud clusters. Subsequently, a multi-target tracking algorithm, such as Kalman filtering, is used to correlate and match these point cloud clusters between consecutive frames, thereby generating a continuous motion trajectory for each target, thus yielding several dynamic targets.

[0020] Meanwhile, dynamic characteristic parameters are calculated for each dynamically tracked target, including at least: (1) duration of existence: the total time that the dynamic target has been stably tracked in consecutive frames from the first time it was detected to the current time; (2) motion trajectory characteristics: parameters that describe the target's motion pattern, such as instantaneous velocity, acceleration, rate of change of motion direction and curvature of trajectory.

[0021] The screening rules are set based on the above dynamic characteristic parameters. For example, if the duration of a dynamic target is short (e.g., less than 0.5 seconds) and its trajectory shows high randomness and discontinuity (e.g., irregular drifting), then the target is determined to be a lightweight dynamic target that is easily affected by airflow, mainly including flying fluff and accidentally drifting yarn in the workshop.

[0022] Step 300: Perform motion center analysis on the corresponding lightweight dynamic target based on the motion trajectory features, and obtain the motion center analysis results, including whether the lightweight dynamic target is in a motion state based on constraint centers, and the number of constraint centers.

[0023] In this step, a more in-depth physical motion analysis is conducted on the aforementioned lightweight dynamic targets that have been initially screened, in order to distinguish between harmless flying fluff and yarn that may constitute an obstacle.

[0024] First, a center of motion analysis is performed to determine whether the motion of a lightweight, dynamic target is constrained by one or more fixed points. Free-floating fluff moves randomly without a fixed center of constraint. In contrast, a piece of yarn that accidentally falls from equipment usually retains at least one end attached to the spinning machine or bobbin; its motion is actually a swinging or pulling motion around this or these fixed points.

[0025] By analyzing the historical motion trajectory characteristics of lightweight dynamic targets (such as the periodicity of the trajectory and the center of rotation), the motion center analysis results can be obtained: (1) Motion state judgment: judge whether the motion is based on the constraint center or the free motion state; (2) Number of constraint centers: further analyze the number of constraint centers, which are mainly divided into single constraint centers (such as one end of the yarn is fixed) and double constraint centers (such as both ends of the yarn are accidentally caught, forming a taut line segment).

[0026] Step 400: Based on the motion center analysis results, a number of obstacle targets are determined, and a driving trajectory is planned in the surrounding environment based on each obstacle target, and then executed.

[0027] In this step, based on the analysis results of step 300, the following decision rule is executed, for example: Lightweight dynamic targets, i.e., flying fluff, which are determined to be in a free-moving state, are filtered out and not considered obstacles because they do not pose a substantial obstacle to the AGV. Lightweight dynamic targets (i.e., yarn) that are determined to be in a motion state based on a constraint center, especially targets identified as having double constraint centers, mean that a piece of yarn may lie across the AGV's path, posing a risk of pulling on the equipment or the AGV itself when it passes over it, and are therefore identified as obstacle targets.

[0028] Furthermore, the AGV takes into account all the final identified real obstacle targets (including traditional obstacles and new obstacles made of yarn) and plans a safe and efficient travel trajectory. The AGV then executes this planned trajectory, thereby avoiding misjudging flying fibers while effectively mitigating the potential risks posed by the yarn, ensuring the continuity and safety of the transfer process. Specifically, avoidance values ​​can be configured for obstacles with single or double constraint centers. These avoidance values ​​determine the distance between the travel trajectory and the obstacle target. Obviously, the distance between the travel trajectory and an obstacle target with a single constraint center is less than the distance to an obstacle target with a double constraint center (because obstacles with double constraint centers may be more dangerous for the AGV).

[0029] It is understandable that the planning of the driving trajectory can be handled by the AGV's own processor or by the server. The server can be a web server or an edge server, without specific limitations. When the server is responsible, the AGV will transmit the dynamic obstacle targets obtained above, as well as other static obstacle targets, to the server. The server will then create a map of the AGV's surrounding environment and use a suitable path planning algorithm to plan the driving trajectory.

[0030] This invention utilizes time-series analysis of laser point cloud data to extract the motion characteristics of dynamic targets, effectively distinguishing between lightweight flying fluff and fallen yarn in spinning workshops. Furthermore, by introducing motion center analysis, it accurately determines the physical motion constraint state of targets, filtering out unconstrained free-flying fluff and identifying single / double-constrained yarns that may pose an entanglement risk as real obstacles. This invention significantly reduces the false stop rate of AGVs caused by flying fluff interference, ensuring the continuity of material handling and production efficiency. It also effectively avoids the safety risks of AGVs pulling on fallen yarn, improves the reliability of perception and the level of intelligent operation in complex textile environments, and ultimately enhances the accuracy of trajectory planning.

[0031] As an example, after obtaining a time-series continuous set of laser point cloud data frames, the method further includes a data preprocessing step for the point cloud data frame set, specifically including: step 101, static background filtering of the point cloud data based on the prior structural information of the fixed equipment in the spinning workshop.

[0032] In this step, the internal layout of the spinning workshop is relatively fixed, including a large number of fixed devices with known locations and shapes, such as spinning machines, winding machines, support columns, and walls. When these fixed devices are scanned by lidar, they form a stable background point cloud. If not processed, they will continue to be misjudged as obstacles, severely increasing the subsequent computational burden and affecting the accuracy of perception.

[0033] Structural prior information refers to a digital map containing the precise 3D coordinates and geometric contours of all fixed equipment, obtained through pre-construction laser SLAM mapping or importing workshop design drawings (CAD models). In actual operation, the AGV obtains its pose in the current global coordinate system through its positioning system, and then registers each frame of laser point cloud data acquired in real time with this high-precision prior map. Through spatial coordinate transformation and comparison, point cloud data points falling within the known fixed equipment model range can be accurately identified. The identified point cloud data belonging to the static background is removed from the current frame. This operation significantly purifies the point cloud data, allowing subsequent processing to focus on truly dynamic or unknown objects, improving the real-time performance and accuracy of environmental perception.

[0034] Step 102: For the point cloud data after filtering out the static background, an outlier filtering algorithm based on statistical features is used to remove discrete noise points that do not satisfy the spatial distribution continuity caused by small-path suspended fluff in the air.

[0035] In this step, after static background filtering, the point cloud data still contains two main components: one is real dynamic targets (such as personnel, other AGVs, and falling yarn), and the other is instantaneous, discrete noise formed by the reflection of laser light from lightweight flying fluff such as cotton and chemical fibers diffused in the workshop. Refined filtering is needed for the latter, achieved using a statistical outlier filtering algorithm (e.g., a statistical outlier removal algorithm). The principle is as follows: the algorithm statistically analyzes the local spatial distribution characteristics of each point in the point cloud. Specifically, for any point P in the point cloud, its k nearest neighbors are searched, and the average distance from point P to these k neighbors is calculated. Subsequently, the average distance between all points in the entire point cloud or a local area and their neighbors is calculated, along with the mean (μ) and standard deviation (σ) of these distances. According to the characteristics of a normal (Gaussian) distribution, the average distance of the vast majority of points should fall within the interval [μ-ασ, μ+ασ] (where α is a scale parameter, typically taken as 1.0 to 3.0). Any point whose average distance is far beyond the upper limit of the interval (i.e., greater than μ+ασ) is considered an outlier that is statistically incompatible with the surrounding point cloud.

[0036] In a spinning workshop environment, small-diameter suspended fluff particles are small in size and sparsely distributed. The reflection points generated by the laser beam on them are often isolated in space, and their average distance to neighboring points is much greater than the typical distance between points within a continuous solid surface point cloud (such as a person's leg or the surface of a feed bin). Therefore, these fluff reflection points are efficiently and accurately identified as statistical outliers by the aforementioned algorithm and filtered out. Understandably, long chains or clumps of fluff will still be identified as lightweight dynamic targets and require further differentiation and identification in step 300.

[0037] This embodiment can effectively remove a large amount of invalid information caused by environmental interference, greatly reducing the probability of misjudging small-path flying catkins as obstacles in subsequent steps, and improving the reliability of dynamic target recognition and classification.

[0038] As an example, the method of identifying lightweight dynamic targets from each dynamic target based on dynamic feature parameters includes: step 201, calculating the duration of existence and the randomness index of the motion trajectory of each dynamic target; wherein, the randomness index of the motion trajectory is quantified by calculating the standard deviation of the change in the motion direction of the dynamic target in consecutive frames.

[0039] In this step, to achieve accurate and objective identification of lightweight dynamic targets and avoid subjective judgment, this invention uses the following two quantifiable key dynamic feature parameters as criteria: Existence duration: This characterizes the stability of the dynamic target within the perception field of view. For a tracked dynamic target, the total duration of its existence is the total time it is continuously tracked across consecutive frames from the moment it is first clustered and successfully associated in the point cloud data until the current frame. It can be understood that the existence duration can be directly calculated using the frame rate and the number of consecutive tracking frames.

[0040] The randomness index of motion trajectory: This index is used to accurately quantify the degree of disorder in the motion trajectory of a dynamic target. The specific calculation method is as follows: In a continuous frame sequence of the tracked dynamic target, its motion direction angle is calculated frame by frame, for example, the angle value in a polar coordinate system with the AGV as the origin. The standard deviation of the change in motion direction angle between these consecutive frames, i.e., the difference between the direction angles of two adjacent frames, is calculated. This standard deviation is defined as the randomness index. It is understandable that freely drifting catkins have frequent and irregular abrupt changes in their motion direction, resulting in a very large standard deviation of the change in direction; while the motion direction of intentional moving objects such as people and vehicles is usually continuous and smooth, and the standard deviation of the change in direction is relatively small.

[0041] Step 202: If the duration of the existence of the dynamic target is less than the first preset threshold and the randomness index of its motion trajectory is greater than the second preset threshold, then the dynamic target is determined to be a lightweight dynamic target.

[0042] In this step, a dynamic target is only identified as a lightweight dynamic target if it simultaneously meets the following two conditions: (1) The duration of its existence is less than the first preset threshold: This condition is used to filter targets that appear momentarily and briefly. Lightweight objects such as flying catkins may quickly drift across the laser scanning plane, and their duration of stable tracking is relatively short. The first preset threshold can be determined statistically or experimentally based on the actual scenario, for example, set to 0.3 to 0.8 seconds. It is worth noting that a yarn that has just detached from the device may also exhibit a brief and unstable motion state during the initial drifting stage, thus meeting this condition.

[0043] (2) Randomness Index > Second Preset Threshold: This condition is used to filter targets with irregular movement patterns. As mentioned above, a high randomness index is a typical characteristic of objects without active movement capabilities, such as flying catkins. The second preset threshold is also set based on actual scene measurement data to ensure that regular and irregular movements can be effectively distinguished. Similarly, a piece of yarn that falls freely, is fixed at one end, or even is fixed at both ends, will exhibit high randomness in its swinging or drifting trajectory during its initial movement phase.

[0044] Dynamic targets that simultaneously meet the above conditions have behavioral characteristics consistent with the physical characteristics of lightweight objects (including flying fluff and yarn in a specific state of motion) (short appearance time and chaotic movement trajectory), and are therefore initially identified as lightweight dynamic targets.

[0045] This embodiment, based on the dual features of the existence duration and randomness index obtained by quantification, can efficiently and accurately screen out all targets with floating characteristics (including flying fluff and yarn) from complex dynamic environments, so as to ensure the quality and relevance of the input data in the subsequent motion center analysis stage.

[0046] As an example, the step of performing motion center analysis on the corresponding lightweight dynamic target based on the motion trajectory features to obtain the motion center analysis results includes: step 301, for each identified lightweight dynamic target, extracting its historical motion trajectory point set within the time window, and fitting the historical motion trajectory point set into a circle or ellipse.

[0047] In this step, to accurately analyze the motion pattern of lightweight dynamic targets, it is necessary to examine their overall motion trend over a period of time. Specifically, for each lightweight dynamic target identified in step 200, all consecutive position points within a preset time window (e.g., the most recent 1-2 seconds) are extracted from its tracked trajectory sequence to form the historical motion trajectory point set of the lightweight dynamic target. It is understood that the length of the aforementioned time window needs to be set reasonably to ensure that it includes enough trajectory points to reflect the motion pattern, while not introducing excessive noise or pattern changes due to an excessively long time period.

[0048] In a spinning workshop setting, the motion of a constrained yarn typically manifests geometrically as oscillation (approximately an arc or fan shape) around a fixed point (the constraint center) or vibration between two fixed points (approximately a line segment, whose trajectory may project onto a plane as an ellipse). Therefore, this embodiment employs a circular fitting algorithm (such as least squares circle fitting) and an elliptical fitting algorithm to perform curve fitting on the historical motion trajectory point set. It is understood that the fitting process aims to find an optimal circle or ellipse such that the sum of the distances from all points in the trajectory point set to the geometric boundary of that circle or ellipse is minimized.

[0049] Step 302: Based on the goodness of fit, determine whether the motion of the lightweight dynamic target revolves around a stable geometric center. Specifically: if the goodness of fit is higher than a preset threshold, determine that the lightweight dynamic target is in a motion state based on the constraint center, and determine the fitted circle center or ellipse center as the constraint center, and then determine the number of constraint centers based on the fitted shape.

[0050] In this step, the goodness of fit is a statistical measure used to assess the degree of matching between the set of historical motion trajectory points and the fitted circle or ellipse. For example, the coefficient of determination R² can be used. If the goodness of fit is higher than a preset threshold (which can be calibrated experimentally, for example, set to 0.7), it indicates that the actual motion trajectory of the lightweight dynamic target is highly consistent with the theoretical motion model constrained by the geometric center.

[0051] When the goodness of fit exceeds a preset threshold, the lightweight dynamic target is determined to be in a motion state based on constraint centers. At this point, the coordinates of the center of the fitted circle or the center of the ellipse are determined as the spatial location of the geometric center (i.e., the constraint center) around which the motion constraint is located. Further, based on the fitted model used, the number of constraint centers can be inferred. Specifically, if the trajectory point set highly fits a circular model, it usually indicates that the target mainly moves around a fixed constraint center point (single constraint center), such as the oscillation of a yarn fixed at one end; if the trajectory point set highly fits an elliptical model, it indicates that its motion is influenced by two main constraint points (double constraint centers), such as the vibration of a taut yarn with both ends attached. The final motion center analysis result includes the determination of whether the lightweight dynamic target is in a constrained state, as well as the inference of the location and number of constraint centers.

[0052] This embodiment transforms the abstract judgment of physical motion state into a precise mathematical model evaluation by fitting the motion trajectory to a geometric model, thereby effectively distinguishing between freely drifting fluff and yarn with constrained motion. Simultaneously, the objectivity and accuracy of the state determination are ensured through a goodness-of-fit threshold. Furthermore, the number of constraint centers inferred from the fitted shape can be used to accurately identify high-risk taut yarns (double constraint centers).

[0053] As an example, the step of fitting the historical motion trajectory point set into a circle or ellipse includes: step 3011, after completing the preliminary fitting of the circle or ellipse, extracting multidimensional features in the fitting process, including at least the distribution density of trajectory points, the uniformity of the distribution of fitting residuals, and the continuity of the distribution of projection points of the trajectory point set on the fitting curve.

[0054] In this step, in existing technologies, when performing motion center analysis on the motion trajectory, it is usually based solely on the goodness of fit obtained from the initial fitting (such as R0). 2The single indicator (value) is used as the sole criterion for judgment. However, in the complex environment of a real spinning workshop, this method of judgment with a single indicator has obvious defects: for example, a short, sparse, and noisy free-flying yarn trajectory may, by chance, produce a high initial goodness of fit with a certain geometric model, leading to misjudgment as constrained motion; conversely, a real constrained yarn trajectory may have its initial goodness of fit underestimated due to uneven distribution of sampling points or the presence of a few outliers, resulting in missed detection.

[0055] To address the aforementioned issues, this embodiment further extracts multidimensional features that can more deeply reflect the quality of the fit after initial fitting. These features include at least the following: distribution density of trajectory points: used to evaluate the uniformity of the coverage of trajectory points on the entire fitted curve, avoiding misjudgments caused by local fitting. The calculation method is as follows: (1) Arc segment division: the fitted complete circular or elliptical curve is divided into N arc segments at equal angles (for example, N=36, i.e., one arc segment every 10 degrees); (2) Projection and counting: each point in the historical motion trajectory point set is vertically projected onto the fitted curve closest to it, and the arc segment number to which the projected point belongs is determined; (3) Density statistics: the number of projected points contained in each arc segment is counted; (4) Uniformity calculation: the standard deviation or coefficient of variation (standard deviation / mean) of the number of projected points in N arc segments is calculated. The standard deviation or coefficient of variation is the quantitative value of the distribution density feature. The smaller the value, the more uniform the distribution; the larger the value, the more concentrated the distribution, and the risk of local fitting exists.

[0056] Uniformity of the distribution of fitting residuals: used to detect whether there is systematic bias, thereby determining whether the fitting model itself is applicable. The calculation method is as follows: (1) Obtaining residuals: record the vertical distance from each trajectory point to the fitting curve as the fitting residual of that point, and at the same time record the azimuth angle of each trajectory point relative to the center of the fitting circle or the center of the ellipse; (2) Azimuth partitioning: divide the azimuth space of 0-360 degrees into M sectors (for example, M=8, that is, one sector every 45 degrees); (3) Sector residual statistics: calculate the average value of the fitting residuals of all trajectory points in each azimuth sector; (4) Uniformity calculation: calculate the standard deviation or range (maximum value minus minimum value) of the average residuals of these M sectors. The standard deviation or range is the quantitative value of the uniformity of the residual distribution. The smaller the value, the more uniform the distribution of residuals in different directions, with no obvious systematic bias; the larger the value, the more uniform the residuals in some directions are, indicating that there is systematic bias and the model may not be applicable.

[0057] The continuity of the distribution of projection points of the trajectory point set on the fitted curve: used to evaluate the temporal coherence of the motion and identify unreliable trajectories caused by tracking jitter or mismatch. The calculation method is as follows: (1) Projection in time order: according to the time order in which the trajectory points are collected, each point is projected onto the fitted curve to obtain a series of projection points arranged in time; (2) Calculation of the arc length distance between adjacent projection points: on the fitted curve, the shortest arc length distance along the curve between the i-th projection point and the i+1-th projection point is calculated; (3) Continuity evaluation: the mean μ and standard deviation σ of the arc length distance between all consecutive projection point pairs are statistically analyzed, and the proportion of these arc length distances exceeding the range of (μ+k*σ) (k is a constant, usually taken as 2 or 3) is calculated, or the coefficient of variation of these arc length distances is directly calculated. The proportion or coefficient of variation that exceeds the range is the quantitative value of the continuity characteristic of the projection point distribution. The smaller the value, the more coherent and stable the motion; the larger the value, the more abnormal jumps or regressions there are in the motion of the trajectory on the curve, the poor coherence, and the low reliability.

[0058] Step 3012: Input the goodness of fit obtained from the initial fitting and the multidimensional features into the pre-trained prediction model. The prediction model performs a comprehensive analysis on the input features and outputs the corrected goodness of fit.

[0059] In this step, the prediction model is pre-built and trained in this embodiment. This model is trained using a large number of labeled trajectory samples (whose true motion state is known), enabling it to learn the complex, non-linear combination relationship between the initial goodness of fit and various multi-dimensional features. In practical applications, the initial goodness of fit and the extracted multi-dimensional features are input into the prediction model for comprehensive analysis. For example, it can identify unreliable fits with high initial goodness of fit but uneven residual distribution, and output a lower corrected goodness of fit; it can also identify reliable fits with moderate initial goodness of fit but good performance across all features, and output a higher corrected goodness of fit.

[0060] Ultimately, the corrected goodness of fit output by the prediction model incorporates more dimensions of quality information, which more accurately reflects the degree of fit between the trajectory and the constrained motion model than a single preliminary goodness of fit. This can significantly improve the accuracy and robustness of motion center state determination and effectively reduce the misjudgment rate in complex environments.

[0061] Understandably, the prediction model can be constructed using suitable existing algorithms, such as gradient boosting decision tree models like XGBoost or LightGBM, which consist of multiple decision trees. During the training phase, the prediction model iteratively generates new decision trees, each dedicated to correcting the residuals predicted by the previous tree. Ultimately, the output of the prediction model is a weighted sum of the predictions from multiple trees. In this scheme, the training objective of the prediction model is to make the weighted sum as close as possible to the true correction goodness of fit (which can be labeled by experts based on the actual physical state of the trajectory).

[0062] The LightGBM example is as follows: Input layer: Input feature vector: [Initial goodness of fit (R²), distribution density feature value (such as coefficient of variation), residual distribution uniformity feature value (such as standard deviation), projection point continuity feature value (such as coefficient of variation)].

[0063] Core model parameters: n_estimators (number of trees): 100. max_depth (maximum depth of a single tree): 6 (used to limit tree depth, prevent overfitting, and ensure model generalization ability). learning_rate (learning rate): 0.1. subsample (the proportion of data subsampled when training each tree): 0.8 (enhances robustness).

[0064] Output layer: Outputs a continuous value between 0 and 1, which is the final goodness of fit after intelligent model correction.

[0065] The specific processing flow is roughly as follows: The model internally uses the comprehensive judgment of its 100 decision trees to output a corrected goodness-of-fit value. For example, even if the initial R² is 0.75, if the distribution density is uneven and the residuals have systematic bias, the model outputs a corrected value of only 0.4; conversely, if the initial R² is 0.7 and all auxiliary features perform well, the model outputs a corrected value as high as 0.85.

[0066] like Figure 2 As shown, this embodiment of the invention also provides an AGV environmental perception system 200 based on a laser beam, including: an on-board lidar 201, configured to: continuously scan the surrounding environment of the AGV in the spinning workshop to obtain a time-series continuous set of laser point cloud data frames.

[0067] The target identification unit 202 is configured to: derive several dynamic targets based on the laser point cloud data frame set, and extract dynamic feature parameters of each dynamic target, including at least the duration of existence and motion trajectory features of the dynamic target, and identify lightweight dynamic targets from each dynamic target based on the dynamic feature parameters.

[0068] The target differentiation unit 203 is configured to perform motion center analysis on the corresponding lightweight dynamic target based on the motion trajectory features, and obtain the motion center analysis results, including whether the lightweight dynamic target is in a motion state based on the constraint center, and the number of constraint centers.

[0069] The trajectory planning unit 204 is configured to: determine a number of obstacle targets based on the motion center analysis results, plan a driving trajectory in the surrounding environment based on each obstacle target, and execute the plan.

[0070] As an example, such as Figure 3 As shown, the system also includes a preprocessing unit 205, which is configured to: perform static background filtering on point cloud data based on prior structural information of fixed equipment in the spinning workshop; and use a statistical feature-based outlier filtering algorithm on the point cloud data after static background filtering to remove discrete noise points that do not meet the spatial distribution continuity caused by small-diameter suspended fluff in the air.

[0071] As an example, the target identification unit 202 is configured to: calculate the duration of existence and the randomness index of the motion trajectory of each dynamic target; wherein the randomness index of the motion trajectory is quantified by calculating the standard deviation of the change in the motion direction of the dynamic target in consecutive frames; if the duration of existence of the dynamic target is less than a first preset threshold and the randomness index of its motion trajectory is greater than a second preset threshold, then the dynamic target is determined to be a lightweight dynamic target.

[0072] As an example, the target differentiation unit 203 is configured to: for each identified lightweight dynamic target, extract its historical motion trajectory point set within a time window, and fit the historical motion trajectory point set into a circle or ellipse; determine whether the motion of the lightweight dynamic target revolves around a stable geometric center based on the goodness of fit. Specifically, the target differentiation unit is configured to: if the goodness of fit is higher than a preset threshold, determine that the lightweight dynamic target is in a motion state based on a constraint center, and determine the fitted circle center or ellipse center as the constraint center, and then determine the number of constraint centers based on the fitted shape.

[0073] As an example, the target discrimination unit 203 is further configured to: after completing the initial fitting of a circle or ellipse, extract multi-dimensional features from the fitting process, including at least the distribution density of trajectory points, the uniformity of the fitting residual distribution, and the continuity of the distribution of projection points of the trajectory point set on the fitting curve; input the initial goodness of fit and the multi-dimensional features into a pre-trained prediction model, and perform a comprehensive analysis of the input features through the prediction model to output the corrected goodness of fit.

[0074] Although the invention has been specifically shown and described with reference to preferred embodiments, those skilled in the art will understand that various modifications in form and detail may be made without departing from the spirit and scope of the invention. Accordingly, the disclosed invention should be considered merely illustrative and limited only by the scope specified in the appended claims.

Claims

1. An AGV environmental perception method based on laser beams, characterized in that, The process includes the following steps: An AGV in the spinning workshop controls an onboard LiDAR to continuously scan the surrounding environment, obtaining a time-series continuous set of LiDAR point cloud data frames; based on the LiDAR point cloud data frames, several dynamic targets are identified, and dynamic feature parameters of each dynamic target are extracted, including at least the duration of the dynamic target's existence and its motion trajectory characteristics; lightweight dynamic targets are identified from among the dynamic targets based on the dynamic feature parameters; based on the motion trajectory characteristics, motion center analysis is performed on the corresponding lightweight dynamic targets to obtain motion center analysis results, including whether the lightweight dynamic target is in a motion state based on a constraint center, and the number of constraint centers; based on the motion center analysis results, several obstacle targets are identified, and a driving trajectory is planned in the surrounding environment based on each obstacle target and executed.

2. The AGV environmental perception method based on laser beams according to claim 1, characterized in that: After obtaining a time-series continuous set of laser point cloud data frames, the method further includes a data preprocessing step for the point cloud data frame set, specifically including: static background filtering of the point cloud data based on the prior structural information of the fixed equipment in the spinning workshop; and using a statistical feature-based outlier filtering algorithm on the point cloud data after static background filtering to remove discrete noise points that do not satisfy the continuity of spatial distribution caused by small-diameter suspended fluff in the air.

3. The AGV environmental perception method based on laser beams according to claim 1, characterized in that: Identifying lightweight dynamic targets from various dynamic targets based on dynamic feature parameters includes: calculating the duration of existence and the randomness index of the motion trajectory of each dynamic target; wherein, the randomness index of the motion trajectory is quantified by calculating the standard deviation of the change in the motion direction of the dynamic target in consecutive frames; if the duration of existence of the dynamic target is less than a first preset threshold and the randomness index of its motion trajectory is greater than a second preset threshold, then the dynamic target is determined to be a lightweight dynamic target.

4. The AGV environmental perception method based on laser beams according to claim 3, characterized in that: Based on the motion trajectory features, motion center analysis is performed on the corresponding lightweight dynamic targets to obtain motion center analysis results, including: for each identified lightweight dynamic target, extracting its historical motion trajectory point set within a time window, and fitting the historical motion trajectory point set into a circle or ellipse; judging whether the motion of the lightweight dynamic target revolves around a stable geometric center based on the goodness of fit, specifically: if the goodness of fit is higher than a preset threshold, it is determined that the lightweight dynamic target is in a motion state based on the constraint center, and the fitted circle center or ellipse center is determined as the constraint center, and then the number of constraint centers is obtained according to the fitted shape.

5. The AGV environmental perception method based on a laser beam according to claim 4, characterized in that: The process of fitting the historical trajectory point set into a circular or elliptical shape includes: after completing the initial circular or elliptical fitting, extracting multidimensional features from the fitting process, including at least the distribution density of trajectory points, the uniformity of the fitting residual distribution, and the continuity of the projection points of the trajectory point set on the fitting curve; inputting the goodness of fit obtained from the initial fitting and the multidimensional features into a pre-trained prediction model, and using the prediction model to perform a comprehensive analysis of the input features to output a corrected goodness of fit.

6. An AGV environmental perception system based on a laser beam, characterized in that, The system includes: a vehicle-mounted lidar configured to continuously scan the surrounding environment of an AGV in a spinning workshop to obtain a time-series continuous set of lidar point cloud data frames; a target initial recognition unit configured to: derive several dynamic targets based on the set of lidar point cloud data frames, and extract dynamic feature parameters of each dynamic target, including at least the duration of the dynamic target's existence and its motion trajectory characteristics, and identify lightweight dynamic targets from among the dynamic targets based on the dynamic feature parameters; a target differentiation unit configured to: perform motion center analysis on the corresponding lightweight dynamic targets based on the motion trajectory characteristics, and obtain motion center analysis results, including whether the lightweight dynamic target is in a motion state based on a constraint center, and the number of constraint centers; and a trajectory planning unit configured to: determine several obstacle targets based on the motion center analysis results, plan a driving trajectory in the surrounding environment based on each obstacle target, and execute the plan.

7. The AGV environmental perception system based on a laser beam according to claim 6, characterized in that: The system also includes a preprocessing unit configured to: perform static background filtering on point cloud data based on prior structural information of fixed equipment in the spinning workshop; and use a statistical feature-based outlier filtering algorithm on the point cloud data after static background filtering to remove discrete noise points that do not satisfy spatial distribution continuity caused by small-diameter suspended fluff in the air.

8. The AGV environmental perception system based on a laser beam according to claim 6, characterized in that: The target identification unit is configured to: calculate the duration of existence and the randomness index of the motion trajectory of each dynamic target; wherein the randomness index of the motion trajectory is quantified by calculating the standard deviation of the change in the motion direction of the dynamic target in consecutive frames; if the duration of existence of the dynamic target is less than a first preset threshold and the randomness index of its motion trajectory is greater than a second preset threshold, then the dynamic target is determined to be a lightweight dynamic target.

9. The AGV environmental perception system based on a laser beam according to claim 8, characterized in that: The target differentiation unit is configured to: for each identified lightweight dynamic target, extract its historical motion trajectory point set within a time window, and fit the historical motion trajectory point set into a circle or ellipse; determine whether the motion of the lightweight dynamic target revolves around a stable geometric center based on the goodness of fit. Specifically, the target differentiation unit is configured to: if the goodness of fit is higher than a preset threshold, determine that the lightweight dynamic target is in a motion state based on a constraint center, and determine the fitted circle center or ellipse center as the constraint center, and then determine the number of constraint centers based on the fitted shape.

10. The AGV environmental perception system based on a laser beam according to claim 9, characterized in that: The target discrimination unit is further configured to: after completing the preliminary fitting of a circle or ellipse, extract multi-dimensional features from the fitting process, including at least the distribution density of trajectory points, the uniformity of the fitting residual distribution, and the continuity of the projection points of the trajectory point set on the fitting curve; input the goodness of fit obtained from the preliminary fitting and the multi-dimensional features into a pre-trained prediction model, and perform a comprehensive analysis of the input features through the prediction model to output the corrected goodness of fit.

Images