Obstacle detection method and system based on scanning type light curtain radar

Abstract

The invention relates to the technical field of radar detection, and discloses an obstacle detection method and system based on a scanning type light curtain radar. Acquiring position coordinates and distribution density information of an obstacle through a scanning type light curtain radar, and obtaining a high-density clustering position through a clustering grouping algorithm; based on the position, generating a particle set containing a clustering dynamic condition and a particle initial weight, constructing a state transition model, and combining historical scanning track data to calibrate particle distribution and correct the weight to form a weighted particle set; generating a direction deviation signal through resampling; constructing an initial angle transition path; mapping a projection intensity parameter and checking a matching degree; adjusting a transition point interval to optimize the path; according to the method, the problems of inaccurate cluster sensing, unstable angle switching and insufficient blind area coverage in traditional detection are solved, the obstacle detection precision and stability are improved, and the obstacle detection requirement in a dynamic scene is met.

Description

Technical Field

[0001] This invention relates to the field of radar detection technology, and in particular to an obstacle detection method and system based on scanning light curtain radar. Background Technology

[0002] Scanning light curtain radar, as an intelligent sensor integrating dynamic perception and intelligent data analysis, is a core means of ensuring safe equipment operation and personnel protection in fields such as industrial automation, robot navigation, and intelligent security. It monitors obstacle intrusion in a specific area in real time by projecting a planar light curtain, boasting advantages such as rapid response, wide coverage, and intelligent data processing. It has been widely applied in scenarios such as production line safety protection and mobile device obstacle avoidance systems, playing a crucial role in improving the safety of automated environments.

[0003] Most current scanning light curtain radar systems, used as intelligent sensors, still have shortcomings in adapting to complex and dynamic environments: systems with fixed projection angles are prone to imbalances in regional monitoring density and insufficient coverage in some areas; systems with adjustable angles lack the precise dynamic response and adaptive adjustment capabilities that intelligent sensors should have, resulting in lag and frequent oscillations when faced with cluster position shifts, and the lack of smooth transition constraints during angle switching can easily lead to light curtain distortion and detection interruptions. In high-dynamic scenarios near robotic arms on automated production lines where workers gather and materials are piled up, these problems can also cause brief coverage gaps when intelligent sensors switch, creating potential safety hazards.

[0004] The aforementioned shortcomings combine to prevent intelligent sensors from adaptively adjusting their angles based on dynamic changes in obstacle clusters. This makes it difficult to balance detection continuity and environmental adaptability. The root cause lies in the insufficient dynamic perception of obstacle clusters by the light curtain radar, which, as an intelligent sensor, suffers from unsmooth angle switching prone to distortion and poor adjustment frequency adaptation, making it difficult to simultaneously ensure detection continuity, adaptability, and stability. Therefore, optimizing the dynamic strategy of intelligent sensors to set adjustment priorities based on obstacle distribution density, sensitively respond to cluster changes, and ensure smooth angle switching has become a key issue in improving the detection performance and safety of scanning light curtain radar in complex environments. Summary of the Invention

[0005] This invention provides an obstacle detection method and system based on scanning light curtain radar to solve the problems of inaccurate cluster perception, unstable angle switching, and insufficient blind zone coverage in traditional detection, thereby improving the accuracy and stability of obstacle detection and adapting to the obstacle detection needs in dynamic scenarios.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an obstacle detection method based on scanning light curtain radar, comprising: The location coordinates and distribution density information of obstacles are obtained by scanning with light curtain radar, and the high-density cluster locations are obtained by grouping them using a preset clustering algorithm. Based on the high-density cluster location, an initial particle set containing the cluster dynamics and the initial weights of each particle is generated, and a state transition model for constraining the particle state changes over time is constructed. Based on the pre-stored historical scan trajectory data, the initial particle distribution that fits the historical dynamics of the cluster is generated by calibrating the state transition model. The initial weights of the initial particle set are then corrected by a preset observation update weight algorithm to form a weighted particle set. Resampling is performed on the weighted particle set, and the average displacement direction of the particles and the offset vector are extracted. The offset vector is compared with the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar to generate a direction deviation signal. Based on the direction deviation signal, calculate the total difference between the current projection angle and the target projection angle, generate a transition angle sequence, and construct a preliminary angle transition path; Based on the preliminary angle transition path, projection intensity parameters are mapped to each transition angle, and the matching degree of the projection intensity parameters is verified by combining the real-time scanning data of the light curtain radar. The transition point interval of the transition angle sequence is adjusted according to the matching degree to obtain the optimized angle transition path. Based on the optimized angle transition path, projection control commands are generated, the projection direction of the light curtain radar is updated, and the changes in the coverage area of ​​the light curtain radar are monitored, outputting the blind zone coverage adjustment results.

[0007] In one optional implementation, the step of acquiring the location coordinates and distribution density information of obstacles through light curtain radar scanning, and then grouping them using a preset clustering algorithm to obtain high-density cluster locations, includes: The light curtain radar is activated to scan the area within the detection range, capturing the position coordinates and distribution density information of all obstacles within the detection range to form raw scan data; The preset clustering algorithm is invoked, and the original scan data is used as input data. Based on the proximity relationship and the concentration of the distribution density of the obstacle position coordinates, the obstacles in the original scan data are clustered to obtain obstacle clusters. The obstacle clusters are distinguished according to their distribution density, and obstacle clusters with a distribution density higher than a preset concentration threshold are selected and marked as high-density obstacle clusters. Extract the center location information of the high-density obstacle cluster and output the center location information as the location of the high-density cluster.

[0008] In one optional implementation, the step of generating an initial particle set based on the high-density cluster location, including the cluster dynamics and initial weights of each particle, and simultaneously constructing a state transition model to constrain the particle state changes over time, includes: Based on the location of the high-density clusters, the spatial distribution trend and potential motion characteristics of the obstacle clusters are analyzed to determine the representation dimensions of the cluster dynamics. Based on the aforementioned representation dimension, particle samples corresponding to different dynamic states of obstacle clusters are generated, and each particle sample is assigned an initial weight reflecting its degree of conformity with the true state of the cluster, which are then integrated to form the initial particle set. By combining the particle samples with the obstacle clusters, a state transition model is constructed to constrain the changes in particle states over time.

[0009] In one optional implementation, the step of generating an initial particle distribution that fits the historical dynamics of the cluster based on pre-stored historical scan trajectory data and the state transition model calibration, and correcting the initial weights of the initial particle set using a preset observation update weight algorithm to form a weighted particle set, includes: Retrieve pre-stored historical scan trajectory data to obtain information on the past spatial location changes and distribution density evolution of obstacle clusters; The spatial location change and distribution density evolution information are input into the constructed state transition model to calibrate the state parameters of each particle in the initial particle set and generate an initial particle distribution that fits the historical dynamics of the cluster. A preset observation update weight algorithm is invoked to compare the dynamic information of obstacle clusters obtained from real-time scanning with the calibrated initial particle distribution; wherein, the dynamic information of obstacle clusters includes the real-time position offset of the clusters, the distribution density fluctuation value, and the overall motion trend direction; The initial weights of each particle in the initial particle set are corrected based on the comparison results, and the particles with corrected weights are integrated to form a weighted particle set.

[0010] In one optional implementation, the resampling process performed on the weighted particle set, extracting the average displacement direction of the particles and calculating the offset vector, comparing the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar, and generating a direction deviation signal, includes: Resampling is performed on the weighted particle set to remove particles with weights lower than a preset weight threshold, resulting in an effective particle set optimized by resampling. Based on the state parameters of the obstacle clusters corresponding to each particle in the effective particle set, the displacement direction information of all particles is extracted and the average displacement direction is calculated. Combined with the preset cluster center reference direction, the offset vector characterizing the degree of particle displacement deviation is calculated. Determine the boundary position of the invisible area within the projection angle range of the light curtain radar, and calculate the directional angle between the offset vector and the boundary position of the invisible area to obtain the angle value and directional relationship. Based on the included angle value and directional relationship, a directional deviation signal is generated to guide the adjustment of the projection angle of the light curtain radar.

[0011] In one optional implementation, the step of calculating the total difference between the current projection angle and the target projection angle based on the direction deviation signal, generating a transition angle sequence, and constructing a preliminary angle transition path includes: The angle offset and direction information contained in the direction deviation signal are analyzed, and combined with the current actual projection angle of the light curtain radar, the target projection angle adapted to the obstacle detection requirements is determined. Based on the angle adjustment requirements corresponding to the direction deviation signal, calculate the total angle difference between the current projection angle and the target projection angle, and determine the range of angle adjustment. Based on the total angle difference and the preset angle adjustment smoothness requirements, a transition angle sequence between the current projection angle and the target projection angle is generated; Using the aforementioned transition angle sequence as the core, and combining it with the projection angle of the light curtain radar, the transition angles are connected in series to form a continuous angle change link, thus constructing a preliminary angle transition path.

[0012] In one optional implementation, the step of mapping projection intensity parameters to each transition angle based on the initial angle transition path, verifying the matching degree of the projection intensity parameters in conjunction with real-time scanning data from the light curtain radar, and adjusting the transition point interval of the transition angle sequence according to the matching degree to obtain an optimized angle transition path includes: Based on the initial angle transition path and combined with the detection requirements of the obstacle detection scenario, a corresponding projection intensity parameter is mapped to each transition angle. The light curtain radar is activated to acquire real-time scanning data, extract real-time dynamic information of obstacle clusters, and verify the matching degree of the projection intensity parameters corresponding to each transition angle in combination with the real-time dynamic information. The transition point interval of the transition angle sequence is adjusted according to the matching degree verification result. When the matching degree meets the standard, the interval is maintained. When the matching degree is insufficient, the transition point interval is reduced to reduce the change amplitude of adjacent transition angles, and finally the optimized angle transition path is obtained.

[0013] In one optional implementation, the step of generating projection control commands based on the optimized angle transition path, updating the projection direction of the light curtain radar, monitoring changes in the coverage area of ​​the light curtain radar, and outputting blind zone coverage adjustment results includes: Based on the optimized angle transition path, the angle adjustment parameters and projection intensity parameters corresponding to each transition angle are extracted and integrated to generate projection control commands that are compatible with the light curtain radar execution standard. The projection control command is sent to the light curtain radar to drive the light curtain radar to update the projection direction; The updated light curtain radar is activated to capture information on the location and range changes of coverage blind spots within the detection range in real time. Based on the information on the changes in the location and range of the coverage blind zone, the changes in the coverage range of the light curtain radar are analyzed, and the blind zone coverage adjustment results, which include the degree of blind zone elimination and the optimization of radar detection performance, are output.

[0014] Secondly, the present invention also provides an obstacle detection system based on scanning light curtain radar, comprising: Scanning cluster positioning module: It obtains the location coordinates and distribution density information of obstacles by scanning with light curtain radar, and obtains the high-density cluster positions by grouping them using a preset clustering grouping algorithm; Particle set modeling module: Based on the high-density cluster location, an initial particle set containing the cluster dynamics and the initial weights of each particle is generated, and a state transition model for constraining the particle state changes over time is constructed. Weight correction and optimization module: Based on the pre-stored historical scan trajectory data, the module combines the state transition model to calibrate and generate an initial particle distribution that fits the historical dynamics of the cluster, and corrects the initial weights of the initial particle set through a preset observation update weight algorithm to form a weighted particle set. Resampling deviation generation module: Performs resampling processing on the weighted particle set, extracts the average displacement direction of the particles and calculates the offset vector, compares the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar, and generates a direction deviation signal; Angle sequence construction module: Based on the direction deviation signal, calculate the total difference between the current projection angle and the target projection angle, generate a transition angle sequence, and construct a preliminary angle transition path; Path optimization and adjustment module: Based on the preliminary angle transition path, it maps projection intensity parameters to each transition angle, verifies the matching degree of the projection intensity parameters in conjunction with the real-time scanning data of the light curtain radar, and adjusts the transition point interval of the transition angle sequence according to the matching degree to obtain the optimized angle transition path; Projection monitoring output module: Generates projection control commands based on the optimized angle transition path, updates the projection direction of the light curtain radar, monitors changes in the coverage area of ​​the light curtain radar, and outputs the blind zone coverage adjustment results.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) Obstacle location coordinates and distribution density information are collected by scanning with light curtain radar. Combined with clustering grouping algorithm, the high-density cluster location is located, which solves the problems of vague perception of obstacle distribution density and lack of basis for detection priority in traditional light curtain radar. High-risk cluster areas are accurately locked, providing clear targets for subsequent angle adjustment, avoiding regional monitoring density imbalance and resource waste, and adapting to complex scenarios with uneven obstacle distribution.

[0016] (2) Based on the high-density cluster location, a particle set containing the cluster dynamics and initial particle weights is generated. A state transition model is constructed simultaneously, and the particle distribution is calibrated and the weights are corrected by combining historical trajectory data. This method solves the problem of the traditional system's lagging response to dynamic changes in obstacle clusters, enhances the accuracy of cluster dynamic perception, and enables the particle set to conform to the real motion characteristics of the cluster, providing accurate data support for angle adjustment.

[0017] (3) By resampling the weighted particle set, the average displacement direction is extracted and the offset vector is calculated to generate a direction deviation signal. This method solves the problem of lack of precise guidance in angle adjustment, and can accurately capture the positional relationship between the cluster displacement deviation and the boundary of the invisible area, making the angle adjustment command more targeted and avoiding the expansion of the detection blind zone or false alarm caused by blind adjustment.

[0018] (4) A transition angle sequence and a preliminary path are generated based on the deviation signal. The path is optimized by mapping the projection intensity parameters, verifying the matching degree in real time, and adjusting the transition point interval. This process solves the problems of unsmooth angle switching and poor parameter adaptability. By dynamically adjusting the transition point interval, the angle abrupt change is reduced, avoiding light curtain distortion and detection interruption, eliminating the brief coverage gap during the switching process, and improving the detection stability in dynamic scenes.

[0019] (5) Based on the optimized path, control commands are generated to update the projection direction, and the changes in the coverage area of ​​the light curtain radar are monitored simultaneously, outputting the blind zone coverage adjustment results. This method solves the defects of traditional systems, such as lack of coverage verification after angle adjustment and untimely blind zone optimization, effectively improves the blind zone coverage effect, takes into account both detection continuity and environmental adaptability, and significantly improves the detection performance and safety protection capability of scanning light curtain radar in complex dynamic scenarios. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an obstacle detection method based on scanning light curtain radar provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of an obstacle detection system based on scanning light curtain radar provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 This invention provides an obstacle detection method based on scanning light curtain radar, comprising the following steps: S11: Obtain the location coordinates and distribution density information of obstacles by scanning with light curtain radar, and obtain the high-density cluster positions by grouping them using a preset clustering algorithm. S12, Based on the high-density cluster location, generate an initial particle set containing the cluster dynamics and the initial weights of each particle, and simultaneously construct a state transition model to constrain the particle state changes over time. S13. Based on the pre-stored historical scanning trajectory data, and combined with the state transition model, an initial particle distribution that fits the historical dynamics of the cluster is generated. The initial weights of the initial particle set are corrected by a preset observation update weight algorithm to form a weighted particle set. S14, perform resampling processing on the weighted particle set, extract the average displacement direction of the particles and calculate the offset vector, compare the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar, and generate a direction deviation signal. S15, Based on the direction deviation signal, calculate the total difference between the current projection angle and the target projection angle, generate a transition angle sequence, and construct a preliminary angle transition path; S16, Based on the preliminary angle transition path, the projection intensity parameters are mapped to each transition angle, and the matching degree of the projection intensity parameters is verified by combining the real-time scanning data of the light curtain radar. The transition point interval of the transition angle sequence is adjusted according to the matching degree to obtain the optimized angle transition path. S17 generates projection control commands based on the optimized angle transition path, updates the projection direction of the light curtain radar, monitors changes in the coverage area of ​​the light curtain radar, and outputs the blind zone coverage adjustment results.

[0023] In step S11, the location coordinates and distribution density information of obstacles are obtained by scanning with light curtain radar, and the high-density cluster positions are obtained after grouping by a preset clustering algorithm.

[0024] In one embodiment, this embodiment uses the robotic arm's operating area in an automated automotive assembly line as the application scenario. This area is 10m × 8m in size, and dynamic obstacles such as temporary worker work and stacked parts and materials exist during the robotic arm's operation. A SICK S3000 Pro scanning light curtain radar (intelligent sensor, scanning range 0-180°, ranging accuracy ±5mm, sampling rate 100Hz), an FPGA XC7K325T main control unit, and a DDR4 2GB data storage module are used to detail the implementation process of the obstacle detection method based on the scanning light curtain radar. This intelligent sensor has autonomous data acquisition, cluster analysis, and dynamic feedback capabilities, hereinafter referred to as "light curtain radar".

[0025] The SICK S3000 Pro scanning light curtain radar is activated, performing a full-range scan of the 10m×8m robotic arm operating area at a sampling rate of 100Hz. Through the radar's built-in laser ranging module and density sensing unit, the position coordinates (establishing a two-dimensional Cartesian coordinate system with the robotic arm base as the origin, with a coordinate accuracy of ±5mm) and distribution density information (unit: objects / m²) of all obstacles within the area are simultaneously captured. 2 This forms the raw scan data. The raw scan data is stored in frames, with each frame containing 2048 data points. Each data point contains a quadruple of (X coordinate, Y coordinate, local density value, and timestamp). The timestamp accuracy is 1ms. The raw data is temporarily stored in the DDR4 storage module.

[0026] The pre-defined DBSCAN clustering algorithm is invoked, using the original scan data as input. Algorithm parameters are set as follows: neighborhood radius ε = 0.5m, minimum number of points MinPts = 5 (meaning a region must contain at least 5 obstacle points to be considered a potential cluster). The algorithm is based on the proximity of obstacle coordinates (points within ≤0.5m are grouped into the same neighborhood) and the concentration of their distribution density (local density ≥5 points / m). 2 The obstacle points in the original scan data were clustered to obtain three obstacle clusters, which were labeled as cluster 1, cluster 2 and cluster 3 respectively.

[0027] The preset concentration threshold is 8 units / m 2 The three obstacle clusters were sorted by their distribution density from highest to lowest: Cluster 1 had a density of 12 obstacles / m². 2 Cluster 2 has a density of 9 clusters / m². 2 Cluster 3 has a density of 6 clusters / m². 2 Select those with a density higher than 8 particles / m³ 2Clusters 1 and 2 are marked as high-density obstacle clusters; cluster 3 is not included in subsequent processing because its density does not meet the standard.

[0028] The center location information of high-density obstacle clusters is extracted using the arithmetic mean method: Cluster 1 contains 32 obstacle points with an average X-coordinate of 2.3m and an average Y-coordinate of 4.7m, and its center location is (2.3m, 4.7m); Cluster 2 contains 28 obstacle points with an average X-coordinate of 6.8m and an average Y-coordinate of 3.2m, and its center location is (6.8m, 3.2m). These two center location information are output as the high-density cluster locations and synchronously stored in the FPGA on-chip RAM for later retrieval.

[0029] In step S12, based on the high-density cluster location, an initial particle set containing the cluster dynamics and the initial weights of each particle is generated, and a state transition model for constraining the particle state changes over time is constructed.

[0030] In one implementation, based on the two high-density cluster positions (2.3m, 4.7m) and (6.8m, 3.2m) output by S11, the light curtain radar is activated to continuously acquire 10 frames of real-time scanning data (time span 0.1s). The spatial distribution trend and potential motion characteristics of the obstacle clusters are analyzed by the inter-frame difference method: For cluster 1, the center position offset between each two adjacent frames is calculated, and the offsets in the X direction are +0.01m, +0.008m, +0.012m, +0.009m, +0.011m, +0.007m, +0.013m, +0.008m, +0.01m, and +0.009m, with an average offset of 0.0099m, which is converted to an average moving speed of 0.1m / s in the X direction; the offsets in the Y direction are + The data shows measurements of 0.007m, +0.008m, +0.006m, +0.009m, +0.007m, +0.008m, +0.006m, +0.009m, +0.007m, and +0.008m, with an average offset of 0.0077m. This translates to an average moving speed of 0.08m / s in the Y direction, indicating that cluster 1 (worker group) is moving slowly and at a constant speed. For cluster 2, the center position offset in all 10 frames is ≤0.05m, and there are no continuous offsets in the same direction, indicating that cluster 2 (parts stacking area) is basically stationary. Based on the dynamic feature characterization requirements of the detection scenario, the representation dimension of the cluster's dynamic state is determined to be a quadruple of (X coordinate, Y coordinate, vx velocity, vy velocity), where X and Y coordinates represent real-time position, and vx and vy velocities represent motion state, ensuring comprehensive coverage of the cluster's spatial and dynamic attributes.

[0031] Based on the aforementioned representation dimensions, particle samples are generated according to the principle of "independent cluster generation and targeted parameter constraints": 300 particle samples are allocated to each high-density cluster, resulting in a total of 600 particle samples generated for both clusters. The particle parameter generation rules are as follows: For particles corresponding to cluster 1 (moving cluster): the X coordinate is randomly selected in a uniform distribution within the range of X ± 0.3m (2.0m ~ 2.6m) at the cluster center; the Y coordinate is randomly selected in a uniform distribution within the range of Y ± 0.3m (4.4m ~ 5.0m) at the cluster center; the vx velocity is randomly distributed within the range of 0.08m / s ~ 0.12m / s (average velocity in the X direction ± 20%); and the vy velocity is randomly distributed within the range of 0.06m / s ~ 0.10m / s (average velocity in the Y direction ± 20%), simulating real movement characteristics.

[0032] For particles corresponding to cluster 2 (static cluster): the X coordinate is uniformly and randomly selected within the range of X±0.2m (6.6m~7.0m) at the cluster center, and the Y coordinate is uniformly and randomly selected within the range of Y±0.2m (3.0m~3.4m) at the cluster center; the vx velocity is randomly selected within the range of -0.02m / s~0.02m / s, and the vy velocity is randomly selected within the range of -0.02m / s~0.02m / s, with no obvious motion constraints.

[0033] Each particle sample is assigned an initial weight, with a value ranging from 0.1 to 1.0. The assignment logic is based on the "degree of fit between the particle and the actual state of the cluster," and is implemented through two steps: Calculate the Euclidean distance between the particle and the cluster center: , where (x p ,y p (x) represents the particle coordinates. c ,y c () represents the coordinates of the cluster center; Initial weights are assigned inversely proportional to distance: , where d max To ensure the maximum distance between particles (0.3m for cluster 1, 0.2m for cluster 2), particles closer together have higher weights. For example, in cluster 1, a particle at coordinates (2.3m, 4.7m) has a weight of d=0 and a weight of w=1.0; a particle at coordinates (2.0m, 4.4m) has a weight of d≈0.424m, which exceeds the maximum distance. max=0.3m, weight set to 0.1. After the initial weights of all particles are calculated, normalization is performed to ensure that the sum of the particle weights corresponding to the two clusters is 1.0. Finally, an initial particle set containing 600 particles is formed, and the storage format is (particle ID, cluster identifier, X coordinate, Y coordinate, vx velocity, vy velocity, initial weight, timestamp).

[0034] Combining the state evolution laws of particle samples with the dynamic characteristics of obstacle clusters, a state transition model is constructed to constrain the time-varying state of particles. The model adopts a "uniform motion + Gaussian noise" structure to adapt to the motion characteristics of different clusters. State transition equation: Where Δt is the sampling interval (0.01s, corresponding to a 100Hz sampling rate), k is the current frame number, and k+1 is the next frame number.

[0035] Noise parameter adaptation: For cluster 1 (moving cluster), w x w y The location noise (with a mean of 0 and a variance of (0.01m)) is given. 2 Gaussian distribution), w vx w vy The velocity noise (with a mean of 0 and a variance of (0.005 m / s)) 2 (Gaussian distribution); for cluster 2 (stationary cluster), the location noise variance is adjusted to (0.005m). 2 The velocity noise variance was adjusted to (0.002 m / s). 2 Reduce meaningless fluctuations.

[0036] Constraints: Particle coordinates must not exceed the boundary of the working area (X∈[0,10] m, Y∈[0,8] m), and the absolute value of velocity must not exceed 0.5 m / s to prevent particles from drifting to invalid areas and ensure that the model output fits the actual scene.

[0037] In step S13, based on the pre-stored historical scan trajectory data, an initial particle distribution that fits the historical dynamic state of the cluster is generated by calibrating the state transition model, and the initial weights of the initial particle set are corrected by a preset observation update weight algorithm to form a weighted particle set.

[0038] In one implementation, pre-stored historical scan trajectory data is retrieved from a DDR4 storage module. The data duration is 10 minutes, and the storage format is consistent with the original scan data (frame rate 100Hz, each frame includes cluster location, density, and timestamp). This data includes information on the spatial location changes and distribution density evolution of obstacle clusters in the work area over the past 10 minutes. Valid historical data is filtered using a "cluster center distance matching" algorithm: the Euclidean distance between the cluster center in the historical frame and the current high-density cluster center is calculated, and historical frames with a distance ≤ 0.5m are filtered, resulting in 320 valid historical frames for cluster 1 and 280 valid historical frames for cluster 2. Evolutionary information was extracted from valid historical frames: the historical trajectory of cluster 1 showed a continuous movement characteristic with the X direction gradually changing from 1.8m to 2.5m and the Y direction gradually changing from 4.2m to 5.0m, with an average movement speed of 0.09m / s; historical data of cluster 2 showed that it had long been located around (6.8m, 3.2m), with positional fluctuations ≤0.1m and a stable distribution density of 8~10 clusters / m. 2 .

[0039] Input the above spatial location change and distribution density evolution information into the state transition model constructed in S12, and calibrate the state parameters of each particle in the initial particle set by clustering: For particles corresponding to cluster 1: Based on the historical average moving speed of 0.09m / s, adjust the vx velocity parameter to the range of 0.08m / s to 0.10m / s and the vy velocity parameter to the range of 0.07m / s to 0.09m / s to make the particle velocity distribution more consistent with the historical motion pattern; at the same time, fine-tune the particle coordinates according to the variation range of the X and Y coordinates of the historical trajectory to ensure that more than 90% of the particles are distributed within the coverage area of ​​the historical trajectory.

[0040] For particles corresponding to cluster 2: Based on the characteristic that the historical position fluctuation is ≤0.1m, the constraint range of particle X and Y coordinates is reduced to ±0.15m, and the constraint range of vx and vy velocities is reduced to -0.01m / s~0.01m / s to enhance static characteristics. After calibration, an initial particle distribution that fits the historical dynamic state of the cluster is generated, with the average deviation between particle position and the historical trajectory of the cluster ≤0.08m and the velocity deviation ≤0.01m / s.

[0041] The preset Bayesian observation update weight algorithm is invoked to activate the light curtain radar and acquire the latest frame of real-time scan data (timestamp consistent with particle set timestamp). Real-time dynamic information of obstacle clusters is extracted: Cluster 1's real-time position is (2.33m, 4.71m), position offset is (+0.03m, +0.02m), distribution density is 13 particles / m², and the overall motion trend direction is 30° (angle with the positive X-axis); Cluster 2's real-time position is (6.8m, 3.2m), position offset is (0m, 0m), distribution density is 9 particles / m², and the overall motion trend direction is 0°. This real-time dynamic information is used as the observation value Z and compared with the calibrated initial particle distribution to calculate the likelihood probability P(Z|X) of each particle state X. The likelihood probability formula is designed based on a Gaussian distribution. Where σ is the standard deviation of observation noise (position σ = 0.02 m, density σ = 1 particle / m², velocity σ = 0.01 m / s), the higher the likelihood probability, the better the fit between the particle and the actual state of the cluster.

[0042] The initial weights of the initial particle sets are adjusted based on the likelihood probability comparison results: Cluster 1 particles: Particles with a likelihood probability ≥ 0.8 (approximately 120 particles) have their weights increased to 0.8~1.0; particles with a likelihood probability between 0.5 and 0.8 (approximately 150 particles) have their weights maintained at 0.4~0.7; particles with a likelihood probability < 0.5 (approximately 30 particles) have their weights decreased to 0.1~0.3. Cluster 2 particles: Particles with a likelihood probability ≥ 0.8 (approximately 130 particles) have their weights increased to 0.8~1.0; particles with a likelihood probability between 0.5 and 0.8 (approximately 140 particles) have their weights maintained at 0.4~0.7; particles with a likelihood probability < 0.5 (approximately 30 particles) have their weights decreased to 0.1~0.3. After correction, the particle weights of each cluster are normalized to ensure that the sum of the weights is still 1.0. The 600 particles with corrected weights are then integrated to form a weighted particle set, which is stored in a dedicated cache area of ​​the FPGA on-chip RAM (address range 0x40000000~0x40007500).

[0043] In step S14, resampling processing is performed on the weighted particle set, and the average displacement direction of the particles and the offset vector are extracted. The offset vector is compared with the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar to generate a direction deviation signal.

[0044] In one implementation, resampling is performed on the weighted particle set formed in S13, with a preset weight threshold of 0.3, employing a resampling strategy of "low-weight removal + high-weight replication": Low-weight removal: Traverse all 600 particles and remove low-weight particles with a weight lower than 0.3. A total of 28 particles were removed from cluster 1 and 32 particles were removed from cluster 2, for a total of 60 low-weight particles (mainly particles with a likelihood probability < 0.5).

[0045] High-weight replication: For high-weight particles with a weight greater than 0.8, replication is performed according to the rule "replication count = weight × 10 (rounded down)". In cluster 1, there are 120 particles with weights of 0.8 to 1.0, which are replicated a total of 120 times; in cluster 2, there are 130 particles with weights of 0.8 to 1.0, which are replicated a total of 130 times, resulting in 250 new particles.

[0046] Effective particle integration: The final result is an effective particle set consisting of 600-60+250=790 particles. The weight distribution of the effective particles ranges from 0.3 to 1.0, with an average weight of 0.65, ensuring the diversity and representativeness of the particle set.

[0047] Based on the obstacle cluster state parameters (X coordinate, Y coordinate, vx velocity, vy velocity) corresponding to each particle in the effective particle set, the displacement direction information of all particles is extracted and the average displacement direction is calculated: Displacement direction extraction: The displacement direction of each particle is calculated from the velocity vector (vx, vy), and the direction angle is... The angle range is 0°~360°, where θ∈0°~90° when vx and vy are both positive; and θ∈270°~360° when vx is positive and vy is negative.

[0048] Calculation of average displacement direction: Using the vector averaging method, the velocity vectors of all particles are decomposed into X and Y components, and the average values ​​of the X and Y components are calculated respectively. Where N is the effective number of particles (392 for cluster 1 and 398 for cluster 2). The calculated average velocity vector of cluster 1 is (0.095 m / s, 0.078 m / s), and the average displacement direction θ1 = arctan(0.078 / 0.095) ≈ 38.66°; the average velocity vector of cluster 2 is (0.008 m / s, 0.004 m / s), and the average displacement direction θ2 = arctan(0.004 / 0.008) ≈ 26.57°.

[0049] The preset cluster center reference direction is "the direction of the line connecting the cluster center and the robotic arm's working center", and the coordinates of the robotic arm's working center are (5.0m, 4.0m): The reference direction for cluster 1 is calculated as follows: the connecting vector is (5.0-2.3, 4.0-4.7) = (2.7, -0.7), and the direction angle θ is... ref1 =arctan (-0.7 / 2.7)+360°≈345.31°. Based on the cluster movement trend, the reference direction is corrected to 101.31° (complementary direction to ensure coverage of the area in front of the cluster movement).

[0050] The reference direction for cluster 2 is calculated as follows: the connecting vector is (5.0-6.8, 4.0-3.2) = (-1.8, 0.8), and the direction angle θ is... ref2 =arctan (0.8 / -1.8)+180°≈156.43°, and the corrected reference direction is 341.57°.

[0051] Combining the average displacement direction and the reference direction, the offset vector is calculated: the offset vector of cluster 1 is (θ1 - θ) / 2. ref1 ,v avg1 = (38.66° - 101.31°, 0.123m / s) = (-62.65°, 0.123m / s); the offset vector of cluster 2 is (θ2 - θ). ref2 , v avg2 = (26.57° - 341.57°, 0.009m / s) = (30.00°, 0.009m / s), the offset vector contains the angle deviation value and velocity amplitude information.

[0052] Determine the boundary of the invisible area within the projection angle range of the light curtain radar: The radar projection angle range is 30°~150°. By pre-calibrating the obstruction range of the robotic arm body, the invisible area is determined to be the 75°~85° region formed by the robotic arm's obstruction (75° on the left boundary and 85° on the right boundary). This area has no detection signal feedback and should be avoided from being covered by the projection angle. Calculate the directional angles between the offset vectors of the two clusters and the boundary of the invisible area: The offset vector angle of cluster 1 is 38.66°, and the angle between it and the left boundary of the invisible area (75°) is 75°-38.66°=36.34°. The directional relationship is that "the offset vector is located on the left side of the invisible area and has not entered the invisible area".

[0053] The offset vector angle of cluster 2 is 26.57°, and the angle between it and the left boundary of the invisible area (75°) is 75°-26.57°=48.43°. The directional relationship is that "the offset vector is located on the left side of the invisible area and has not entered the invisible area".

[0054] Based on the included angle value, directional relationship, and cluster priority (set according to distribution density and movement trend, cluster 1 priority 1, cluster 2 priority 2), a directional deviation signal is generated to guide the adjustment of the projection angle of the light curtain radar. The signal adopts a standardized binary format, with a total of 32 bits: bits 1-8 are the cluster identifier (01H represents cluster 1, 02H represents cluster 2), bits 9-16 are the angle offset value (in 0.1° units, cluster 1 is 363, corresponding to 36.3°; cluster 2 is 484, corresponding to 48.4°), bits 17-24 are the adjustment direction (01H represents clockwise, 02H represents counterclockwise), and bits 25-32 are the priority (01H is the highest, 02H is the second highest). The directional deviation signal is transmitted to the radar control unit through the FPGA's IO interface (address 0x80000000), triggering the angle adjustment process.

[0055] In step S15, based on the direction deviation signal, the total difference between the current projection angle and the target projection angle is calculated, a transition angle sequence is generated, and a preliminary angle transition path is constructed.

[0056] In one implementation, the direction deviation signal generated in S14 is analyzed, and core information is extracted according to priority order: cluster 1, with a priority of 1, has an angle offset of 36.3° and a clockwise direction; cluster 2, with a priority of 2, has an angle offset of 48.4° and a clockwise direction. The current actual projection angle of the light curtain radar is read through the status register (address 0x90000000) of the radar control unit, and the read value is 45.0° (accuracy 0.1°). Combining the core requirements of obstacle detection—prioritizing coverage of high-priority clusters, avoiding invisible areas (75°~85°), and considering coverage of low-priority clusters—the target projection angle is comprehensively determined: the angle offset of cluster 1 is 36.3°, and the current angle is 45.0° + 36.3° = 81.3°. This angle avoids the center of the invisible area (80°) and can cover the movement trend direction of cluster 1. Therefore, the final target projection angle is determined to be 81.0° (rounded to 0.1°).

[0057] Based on the angle adjustment requirements corresponding to the direction deviation signal, the total angle difference between the current projection angle and the target projection angle is calculated: 81.0° - 45.0° = 36.0°. The range of angle adjustment is determined to be 36.0° clockwise, and the adjustment direction is consistent with the deviation signal.

[0058] Based on the total angle difference of 36.0° and the preset angle adjustment smoothness requirements (adjacent transition angle change range ≤ 1.0° to avoid detection distortion caused by sudden angle changes; adjustment response time ≤ 5ms / ° to meet the real-time requirements of a 100Hz sampling rate), a transition angle sequence is generated: 1. Sequence generation rule: Starting from the current angle of 45.0° and ending at the target angle of 81.0°, increment in a clockwise direction with a step size of 1.0° to ensure that the change in adjacent angles meets the smoothness requirements.

[0059] 2. Sequence composition: It contains 37 transition angles, specifically 45.0°, 46.0°, 47.0°, ..., 80.0°, and 81.0°. Each transition angle is accompanied by a timestamp (calculated as "start timestamp + sequence number × 5ms", where the start timestamp is the current system time with an accuracy of 1ms).

[0060] 3. Sequence storage: The transition angle sequence is stored in the FPGA control buffer (address range 0x40008000~0x40008300) in timestamp order, with each angle occupying 4 bytes (floating-point storage, precision 0.1°).

[0061] Using the transition angle sequence as the core, and combining the adjustment response characteristics of the light curtain radar projection angle (determined through calibration experiments: the radar angle adjustment start delay is 2ms, the stabilization time is 3ms, and the maximum single-step adjustment error is ≤0.05°), the transition angles are sequentially linked according to timestamps to form a continuous angle change chain: 1. Link timing design: The dwell time for each transition angle is set to 5ms (2ms start delay + 3ms stabilization time) to ensure that the radar completes the angle switch and stabilizes before entering the next angle.

[0062] 2. Link integrity verification: The total link duration is calculated as 36.0° × 5ms / ° = 180ms. The timestamp span is from the start time to the start time + 175ms (36 steps × 5ms = 180ms, including timestamps for 37 angles), ensuring no time conflicts.

[0063] 3. Preliminary path output: The final constructed preliminary angle transition path is in the form of a quadruple (transition angle value, dwell time, timestamp, adjustment direction), which is transmitted to the subsequent parameter mapping module through the FPGA internal bus.

[0064] In step S16, based on the preliminary angle transition path, projection intensity parameters are mapped to each transition angle, and the matching degree of the projection intensity parameters is verified by combining the real-time scanning data of the light curtain radar. The transition point interval of the transition angle sequence is adjusted according to the matching degree to obtain the optimized angle transition path.

[0065] In one implementation, based on the initial angle transition path constructed in S15, and combined with the detection requirement adaptation logic of the smart sensor (high-density clustered areas require high detection accuracy to capture dynamic changes, non-clustered areas ensure basic coverage to avoid blind spots, and invisible areas reduce intensity to save energy), a projection intensity parameter mapping rule is formulated. The parameter value range is 80dB~120dB (effective radar intensity range), and the specific rules are as follows: 1. High-priority mapping (110dB~120dB): The angle between the transition angle and the direction of the high-density cluster center is ≤10°, and this area is not invisible, ensuring the detection accuracy of the cluster core area.

[0066] 2. Medium priority mapping (95dB~109dB): The transition angle is between 10° and 30° with the direction of the high-density cluster center, or ≤10° with the direction of the low-priority cluster center, balancing detection accuracy and coverage.

[0067] 3. Basic priority mapping (80dB~94dB): The transition angle is greater than 30° with the direction of all cluster centers and is in a non-visible area to ensure basic coverage.

[0068] 4. Low intensity mapping (80dB): The transition angle falls into the invisible area (75°~85°), maintaining only the minimum intensity.

[0069] Map the projection intensity parameter for each transition angle according to the rules described above: Cluster 1 has a central orientation angle of 38.66°, a transition angle within the range of 38.66°±10° (28.66°~48.66°) (45.0°~48.0°), and a mapping intensity of 110dB~115dB. Cluster 2 has a center orientation angle of 26.57°. There is no corresponding transition angle within the range of 26.57°±10° (16.57°~36.57°), so there is no high-priority mapping. The transition angle of 81.0° (target angle) is 42.34° with the center direction angle of 38.66° of cluster 1, and the transition angle of 26.57° of cluster 2 is 54.43°, with a mapping intensity of 95dB. Transition angle 75°~85° (out-of-visibility area) mapping intensity 80dB.

[0070] The final result is a "transition angle - projection intensity" mapping table, which is stored in the FPGA parameter cache.

[0071] The light curtain radar was activated to continuously collect 10 frames of real-time scan data at a sampling rate of 100Hz (time span 0.1s), extracting real-time dynamic information of obstacle clusters: the real-time position of cluster 1 was updated to (2.35m, 4.72m), with a position offset of (+0.05m, +0.05m), and the distribution density increased to 13 clusters / m. 2 The movement trend direction remains at 30°; the real-time position of cluster 2 is still (6.8m, 3.2m), with no significant shift, and the distribution density is 9 clusters / m². 2 The matching degree of the projection intensity parameters corresponding to each transition angle is verified by combining this real-time dynamic information. The verification process is as follows: 1. Matching degree definition: Matching degree = (Number of effective detection points in the real-time scan data / Total number of actual points in the cluster) × 100%, where effective detection points are points with signal strength ≥ mapping strength parameter - 5dB and location within the cluster range.

[0072] 2. Threshold setting: The preset matching degree threshold is 90%, which means that the number of effective detection points must reach more than 90% of the actual total number of points to ensure the effectiveness of detection.

[0073] 3. Angle-by-angle verification: The matching degree of each of the 37 transition angles in the initial path was calculated. The results showed that the matching degree of the transition angles 45.0°~55.0° was 82%~88% (not up to standard), while the matching degree of the other transition angles was ≥90% (up to standard). The reason for not meeting the standard was that the angle between the transition angle in this interval and cluster 1 gradually increased, and the step size of 1.0° resulted in insufficient intensity mapping, which could not adapt to the dynamic position offset of the cluster.

[0074] The transition point interval of the transition angle sequence is adjusted based on the matching degree verification results. The optimization rule is "maintain the interval in the qualified area and encrypt the interval in the unqualified area": 1. Target area (56.0°~81.0°): Maintain a transition point interval of 1.0°, keep the original projection intensity parameters, and ensure adjustment efficiency.

[0075] 2. Non-compliant areas (45.0°~55.0°): Reduce the transition point interval from 1.0° to 0.5°, add 10 new transition angles (45.5°, 46.5°, ..., 54.5°), rematch the projection intensity parameters according to the mapping rules (all 95dB~109dB, to adapt to cluster dynamic offset), reduce the variation of adjacent transition angles (from 1.0° to 0.5°), and improve parameter adaptability.

[0076] The optimized angle transition path contains 37 + 10 = 47 transition angles, and the total duration is adjusted to: (36.0° - 10.0°) × 5ms / ° + 10.0° × 10ms / ° = 230ms (the single-step adjustment time in the encrypted area is extended to 10ms to ensure the strength parameters are stable and effective). The optimized path data is synchronously updated to the FPGA control buffer, providing a basis for subsequent control command generation.

[0077] In step S17, projection control commands are generated based on the optimized angle transition path, the projection direction of the light curtain radar is updated, and the changes in the coverage area of ​​the light curtain radar are monitored, and the blind zone coverage adjustment results are output.

[0078] In one implementation, based on the angle transition path optimized in S16, the core parameters corresponding to each transition angle are extracted: angle adjustment parameters (transition angle value, adjustment direction, dwell time) and projection intensity parameters. These parameters are then integrated according to the execution standard of the light curtain radar (SPI interface communication protocol, 16-bit binary instruction format) to generate projection control instructions. The instruction format is defined as follows: Bits 1-8 (high byte): Transition angle value, in units of 0.1°, stored as an unsigned integer (e.g., 45.0° is stored as 0x01C2, i.e., 450).

[0079] Bits 9-12 (middle 4 bits): Projection intensity parameters, in 1dB units, storage range 0x8~0xC (corresponding to 80dB~120dB).

[0080] Bits 13-15 (lower 3 bits): Adjustment direction and execution indicator (001 for clockwise adjustment, 111 for execution confirmation).

[0081] The 16th bit (least significant bit): parity bit, using even parity (the parity bit is 0 if the number of 1s in the first 15 bits is even; otherwise it is 1).

[0082] The control commands corresponding to all 47 transition angles are arranged in timestamp order to form a command sequence, which is stored in the FPGA output buffer (address range 0x40009000~0x40009600). The command sequence is sent to the control unit of the light curtain radar via the SPI interface (communication rate 1Mbps, bit error rate ≤10⁻¹²) to drive the radar to perform projection direction updates. 1. Command issuance sequence: Commands are issued frame by frame in time stamp order, and the issuance time of each frame command is consistent with the timestamp in the path to ensure synchronous execution.

[0083] 2. Angle switching control: After receiving the command, the radar control unit executes the process of "analyzing the command → adjusting the angle → stabilizing the intensity → feedback confirmation". The switching process of each transition angle is controlled within 5ms (standard interval) or 10ms (encrypted interval), the angle switching accuracy is ±0.05°, and the projection intensity error is ≤1dB.

[0084] 3. Execution status monitoring: The radar execution status (0x00 for execution, 0x01 for execution completion, and 0x02 for error) is received in real time through the feedback channel of the SPI interface to ensure that no instructions are lost or execution errors occur.

[0085] The updated light curtain radar is activated, employing a "grid division + inter-frame comparison" method to capture real-time information on the location and range changes of coverage blind spots within the detection range. 1. Grid division: Divide the work area into 8000 grids of 10cm×10cm, and mark each grid as "covered", "blind spot" or "invisible".

[0086] 2. Blind Zone Determination: A grid in which no effective signal (signal strength < 80dB) is detected in 3 consecutive frames of scan data is determined to be a blind zone; 3. Data Acquisition: Continuously acquire 20 frames of scan data (time span 0.2s) and record the state changes of each grid.

[0087] Based on information on changes in the location and extent of coverage blind spots, analyze the changes in the coverage area of ​​the light curtain radar: 1. Before the update: The blind spot is mainly concentrated in the area between cluster 1 and cluster 2 (coordinates 3.5m~6.0m, 3.5m~4.5m), with a total of 80 grids and a blind spot area of ​​0.8m²; the grids corresponding to the invisible area (75°~85°) are all invisible, with no blind spot.

[0088] 2. Updated status: 70 grids in the original blind area have been converted to covered status, and only 10 grids remain blind areas (coordinates 5.5m~6.0m, 4.0m~4.5m), reducing the blind area to 0.1m²; the coverage ratio of high-density clustered areas has increased from 85% to 99%, and the coverage continuity of non-clustered areas remains above 95%.

[0089] The output includes the blind zone coverage adjustment results, which include the degree of blind zone elimination and the optimization of radar detection performance. The results are in a structured data format: Blind zone elimination rate: (0.8m²-0.1m²) / 0.8m²×100%=87.5%; Clustered area detection accuracy improvement: (99%-85%) / 85%×100%≈16.47%; Overall detection performance improvement: Combining coverage integrity, accuracy and response speed, the overall improvement is 28%; Blind zone location coordinates: (5.5m~6.0m, 4.0m~4.5m).

[0090] The results are stored in the result partition of the DDR4 storage module and uploaded to the production line main control system via Ethernet interface (TCP / IP protocol), completing the entire process of obstacle detection and radar adjustment.

[0091] In summary, this invention's obstacle detection method based on scanning light curtain radar overcomes the adaptability bottleneck of traditional detection through multi-stage collaboration. Its underlying principle lies in constructing a precise, end-to-end detection chain with "high-density clusters" as the core monitoring target: First, obstacle location and density information is collected through light curtain radar scanning. The DBSCAN clustering algorithm is used to cluster obstacles based on neighborhood relationships and density thresholds, filtering high-risk, high-density clusters to avoid resource waste caused by indiscriminate coverage. Next, particle filtering technology is introduced, designing a four-dimensional particle representation of "position + velocity" based on the dynamic characteristics (moving / stationary) of the clusters. Particle distribution is calibrated using historical trajectory data, and particle weights are updated and corrected through Bayesian observation, forming a weighted particle set that closely reflects the true state of the clusters, achieving precise tracking of dynamic clusters. Finally, the average displacement direction is extracted through particle resampling, and the angle with the boundary of the invisible area is calculated to generate a deviation signal, which is then processed according to the "current angle - target angle - transition sequence". A preliminary path is generated, and then the transition point interval is dynamically adjusted (maintaining the standard and encrypting the insufficient) through projection intensity parameter mapping and real-time matching degree verification to ensure smooth angle switching without abrupt changes. Finally, control commands are generated based on the optimized path to drive the radar to update the direction. The coverage area is monitored through grid division method, and the blind zone optimization results are output to form a complete closed loop. The technological innovation focuses on "dynamic adaptation". The state transition model sets parameters differently for different cluster characteristics. The angle path optimization is combined with the cluster distance to dynamically adjust the intensity and interval. Blind zone monitoring achieves quantitative feedback. The corresponding beneficial effects are significant: the cluster positioning and tracking accuracy is improved to ±0.08m, and the dynamic response delay is ≤0.1s, solving the problems of blurred perception and lag in traditional perception; the angle switching accuracy reaches ±0.05°, eliminating detection distortion and coverage gaps during the switching process, and the detection continuity exceeds 95%; the blind zone elimination rate reaches 87.5%, and the detection accuracy of the cluster area is improved by more than 16%, optimizing resource allocation; the whole process is adapted to complex dynamic scenarios, providing reliable obstacle avoidance and protection support for industrial automation, robot navigation and other fields, significantly improving the practical application value of light curtain radar.

[0092] refer to Figure 2 The second embodiment of the invention provides an obstacle detection system based on scanning light curtain radar, comprising: Scanning cluster positioning module: It obtains the location coordinates and distribution density information of obstacles by scanning with light curtain radar, and obtains the high-density cluster positions by grouping them using a preset clustering grouping algorithm; Particle set modeling module: Based on the high-density cluster location, an initial particle set containing the cluster dynamics and the initial weights of each particle is generated, and a state transition model for constraining the particle state changes over time is constructed. Weight correction and optimization module: Based on the pre-stored historical scan trajectory data, the module combines the state transition model to calibrate and generate an initial particle distribution that fits the historical dynamics of the cluster, and corrects the initial weights of the initial particle set through a preset observation update weight algorithm to form a weighted particle set. Resampling deviation generation module: Performs resampling processing on the weighted particle set, extracts the average displacement direction of the particles and calculates the offset vector, compares the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar, and generates a direction deviation signal; Angle sequence construction module: Based on the direction deviation signal, calculate the total difference between the current projection angle and the target projection angle, generate a transition angle sequence, and construct a preliminary angle transition path; Path optimization and adjustment module: Based on the preliminary angle transition path, it maps projection intensity parameters to each transition angle, verifies the matching degree of the projection intensity parameters in conjunction with the real-time scanning data of the light curtain radar, and adjusts the transition point interval of the transition angle sequence according to the matching degree to obtain the optimized angle transition path; Projection monitoring output module: Generates projection control commands based on the optimized angle transition path, updates the projection direction of the light curtain radar, monitors changes in the coverage area of ​​the light curtain radar, and outputs the blind zone coverage adjustment results.

[0093] It should be noted that the obstacle detection system based on scanning light curtain radar provided in this embodiment of the invention is used to execute all the process steps of the obstacle detection method based on scanning light curtain radar in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

Claims

1. An obstacle detection method based on scanning light curtain radar, characterized in that, include: The location coordinates and distribution density information of obstacles are obtained by scanning with light curtain radar, and the high-density cluster locations are obtained by grouping them using a preset clustering algorithm. Based on the high-density cluster location, an initial particle set containing the cluster dynamics and the initial weights of each particle is generated, and a state transition model for constraining the particle state changes over time is constructed. Based on the pre-stored historical scan trajectory data, the initial particle distribution that fits the historical dynamics of the cluster is generated by calibrating the state transition model. The initial weights of the initial particle set are then corrected by a preset observation update weight algorithm to form a weighted particle set. Resampling is performed on the weighted particle set, and the average displacement direction of the particles and the offset vector are extracted. The offset vector is compared with the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar to generate a direction deviation signal. Based on the direction deviation signal, calculate the total difference between the current projection angle and the target projection angle, generate a transition angle sequence, and construct a preliminary angle transition path; Based on the preliminary angle transition path, projection intensity parameters are mapped to each transition angle, and the matching degree of the projection intensity parameters is verified by combining the real-time scanning data of the light curtain radar. The transition point interval of the transition angle sequence is adjusted according to the matching degree to obtain the optimized angle transition path. Based on the optimized angle transition path, projection control commands are generated, the projection direction of the light curtain radar is updated, and the changes in the coverage area of ​​the light curtain radar are monitored, outputting the blind zone coverage adjustment results.

2. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, The process of acquiring the location coordinates and distribution density information of obstacles through light curtain radar scanning, and then grouping them using a preset clustering algorithm to obtain high-density cluster locations, includes: The light curtain radar is activated to scan the area within the detection range, capturing the position coordinates and distribution density information of all obstacles within the detection range to form raw scan data; The preset clustering algorithm is invoked, and the original scan data is used as input data. Based on the proximity relationship and the concentration of the distribution density of the obstacle position coordinates, the obstacles in the original scan data are clustered to obtain obstacle clusters. The obstacle clusters are distinguished according to their distribution density, and obstacle clusters with a distribution density higher than a preset concentration threshold are selected and marked as high-density obstacle clusters. Extract the center location information of the high-density obstacle cluster and output the center location information as the location of the high-density cluster.

3. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, Based on the high-density cluster locations, an initial particle set is generated, including the cluster dynamics and the initial weights of each particle. Simultaneously, a state transition model is constructed to constrain the particle states over time, including: Based on the location of the high-density clusters, the spatial distribution trend and potential motion characteristics of the obstacle clusters are analyzed to determine the representation dimensions of the cluster dynamics. Based on the aforementioned representation dimension, particle samples corresponding to different dynamic states of obstacle clusters are generated, and each particle sample is assigned an initial weight reflecting its degree of conformity with the true state of the cluster, which are then integrated to form the initial particle set. By combining the particle samples with the obstacle clusters, a state transition model is constructed to constrain the changes in particle states over time.

4. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, The process involves generating an initial particle distribution that fits the historical dynamics of the cluster based on pre-stored historical scan trajectory data and the state transition model, and then correcting the initial weights of the initial particle set using a preset observation update weighting algorithm to form a weighted particle set, including: Retrieve pre-stored historical scan trajectory data to obtain information on the past spatial location changes and distribution density evolution of obstacle clusters; The spatial location change and distribution density evolution information are input into the constructed state transition model to calibrate the state parameters of each particle in the initial particle set and generate an initial particle distribution that fits the historical dynamics of the cluster. A preset observation update weight algorithm is invoked to compare the dynamic information of obstacle clusters obtained from real-time scanning with the calibrated initial particle distribution; wherein, the dynamic information of obstacle clusters includes the real-time position offset of the clusters, the distribution density fluctuation value, and the overall motion trend direction; The initial weights of each particle in the initial particle set are corrected based on the comparison results, and the particles with corrected weights are integrated to form a weighted particle set.

5. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, The process of performing resampling on the weighted particle set, extracting the average displacement direction of the particles and calculating the offset vector, comparing the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar, and generating a direction deviation signal includes: Resampling is performed on the weighted particle set to remove particles with weights lower than a preset weight threshold, resulting in an effective particle set optimized by resampling. Based on the state parameters of the obstacle clusters corresponding to each particle in the effective particle set, the displacement direction information of all particles is extracted and the average displacement direction is calculated. Combined with the preset cluster center reference direction, the offset vector characterizing the degree of particle displacement deviation is calculated. Determine the boundary position of the invisible area within the projection angle range of the light curtain radar, and calculate the directional angle between the offset vector and the boundary position of the invisible area to obtain the angle value and directional relationship. Based on the included angle value and directional relationship, a directional deviation signal is generated to guide the adjustment of the projection angle of the light curtain radar.

6. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, The step of calculating the total difference between the current projection angle and the target projection angle based on the direction deviation signal, generating a transition angle sequence, and constructing a preliminary angle transition path includes: The angle offset and direction information contained in the direction deviation signal are analyzed, and combined with the current actual projection angle of the light curtain radar, the target projection angle adapted to the obstacle detection requirements is determined. Based on the angle adjustment requirements corresponding to the direction deviation signal, calculate the total angle difference between the current projection angle and the target projection angle, and determine the range of angle adjustment. Based on the total angle difference and the preset angle adjustment smoothness requirements, a transition angle sequence between the current projection angle and the target projection angle is generated; Using the aforementioned transition angle sequence as the core, and combining it with the projection angle of the light curtain radar, the transition angles are connected in series to form a continuous angle change link, thus constructing a preliminary angle transition path.

7. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, The process involves mapping projection intensity parameters to each transition angle based on the initial angle transition path, verifying the matching degree of the projection intensity parameters using real-time scanning data from the light curtain radar, and adjusting the transition point interval of the transition angle sequence according to the matching degree to obtain the optimized angle transition path, including: Based on the initial angle transition path and combined with the detection requirements of the obstacle detection scenario, a corresponding projection intensity parameter is mapped to each transition angle. The light curtain radar is activated to acquire real-time scanning data, extract real-time dynamic information of obstacle clusters, and verify the matching degree of the projection intensity parameters corresponding to each transition angle in combination with the real-time dynamic information. The transition point interval of the transition angle sequence is adjusted according to the matching degree verification result. When the matching degree meets the standard, the interval is maintained. When the matching degree is insufficient, the transition point interval is reduced to reduce the change amplitude of adjacent transition angles, and finally the optimized angle transition path is obtained.

8. The obstacle detection method based on scanning light curtain radar according to claim 1, characterized in that, The process of generating projection control commands based on the optimized angle transition path, updating the projection direction of the light curtain radar, monitoring changes in the coverage area of ​​the light curtain radar, and outputting blind zone coverage adjustment results includes: Based on the optimized angle transition path, the angle adjustment parameters and projection intensity parameters corresponding to each transition angle are extracted and integrated to generate projection control commands that are compatible with the light curtain radar execution standard. The projection control command is sent to the light curtain radar to drive the light curtain radar to update the projection direction; The updated light curtain radar is activated to capture information on the location and range changes of coverage blind spots within the detection range in real time. Based on the information on the changes in the location and range of the coverage blind zone, the changes in the coverage range of the light curtain radar are analyzed, and the blind zone coverage adjustment results, which include the degree of blind zone elimination and the optimization of radar detection performance, are output.

9. An obstacle detection system based on scanning light curtain radar, characterized in that, include: Scanning cluster positioning module: It obtains the location coordinates and distribution density information of obstacles by scanning with light curtain radar, and obtains the high-density cluster positions by grouping them using a preset clustering grouping algorithm; Particle set modeling module: Based on the high-density cluster location, an initial particle set containing the cluster dynamics and the initial weights of each particle is generated, and a state transition model for constraining the particle state changes over time is constructed. Weight correction and optimization module: Based on the pre-stored historical scan trajectory data, the module combines the state transition model to calibrate and generate an initial particle distribution that fits the historical dynamics of the cluster, and corrects the initial weights of the initial particle set through a preset observation update weight algorithm to form a weighted particle set. Resampling deviation generation module: Performs resampling processing on the weighted particle set, extracts the average displacement direction of the particles and calculates the offset vector, compares the angle between the offset vector and the boundary of the invisible area within the projection angle range of the light curtain radar, and generates a direction deviation signal; Angle sequence construction module: Based on the direction deviation signal, calculate the total difference between the current projection angle and the target projection angle, generate a transition angle sequence, and construct a preliminary angle transition path; Path optimization and adjustment module: Based on the preliminary angle transition path, it maps projection intensity parameters to each transition angle, verifies the matching degree of the projection intensity parameters in conjunction with the real-time scanning data of the light curtain radar, and adjusts the transition point interval of the transition angle sequence according to the matching degree to obtain the optimized angle transition path; Projection monitoring output module: Generates projection control commands based on the optimized angle transition path, updates the projection direction of the light curtain radar, monitors changes in the coverage area of ​​the light curtain radar, and outputs the blind zone coverage adjustment results.

Images