A signal processing method for a single-sided laser vehicle detector of a highway toll station

Abstract

This invention discloses a signal processing method for a single-sided laser vehicle detector at a highway toll station, belonging to the field of signal processing technology. The method corrects inherent ranging biases by constructing a static background reference point cloud through point-by-point ranging error correction under a static, unloaded lane environment. Then, it performs directional cropping of the measured point cloud using the effective detection area of ​​each lane, eliminating redundant noise across lanes and invalid data from non-detection areas. Next, it removes suspended scattering noise under adverse weather conditions, addressing the problem of missed and false detections caused by weather interference while fully preserving the characteristics of the real vehicle point cloud. The discrete effective sampling points are then aggregated into point cloud clusters, and suspected vehicle targets are screened using displacement and directional angle deviations, narrowing down the target range for final accurate verification. Finally, verification based on three-dimensional contour specifications, road surface contact features, and lane space occupancy improves the accuracy and stability of vehicle detection, while also enhancing anti-interference capabilities.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, specifically to a signal processing method for a single-sided laser vehicle detector at a highway toll station. Background Technology

[0002] In the all-weather, all-condition application scenario of single-sided laser vehicle detection at highway toll stations, existing technologies have the following shortcomings: First, existing technologies do not construct a static background benchmark template adapted to the static scene of an empty toll station lane, making it easy to misjudge static background clutter as dynamic targets. Second, they struggle to distinguish the characteristic differences between high-reflectivity vehicle targets and dense weather patches, and cannot adaptively adapt to changes in interference characteristics under different weather conditions. Third, the lack of multi-dimensional target verification results in insufficient accuracy and stability in vehicle target detection, making it impossible to achieve stable identification of vehicle targets within the lane under complex conditions.

[0003] Therefore, there is an urgent need for a signal processing method for single-sided laser vehicle detectors at highway toll stations that can remove static background clutter, filter out multi-dimensional floating scattering clutter in adverse weather conditions, and perform target verification by combining the inherent motion and physical characteristics of vehicles, in order to solve the above-mentioned technical problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a signal processing method for a single-sided laser vehicle detector at highway toll stations. This method solves the problems of existing single-sided laser vehicle detection technologies at highway toll stations being susceptible to interference from static background clutter and suspended scattering noise caused by adverse weather conditions, resulting in high false detection and false negative rates for vehicle targets and poor anti-interference capabilities.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: a signal processing method for a single-sided laser vehicle detector at a highway toll station, comprising the following steps: correcting the ranging error point by point based on the original point cloud under a static environment with the lane empty, to obtain a static background reference point cloud.

[0006] The effective point cloud for each lane is obtained by cropping the measured point cloud based on the effective detection area corresponding to each lane.

[0007] The distance between the effective point cloud and the static background reference point cloud is calculated, and the static point cloud is removed to obtain the dynamic point cloud.

[0008] Meteorological interference points are screened from dynamic point clouds to obtain candidate valid point clouds.

[0009] The candidate valid point clouds are clustered to obtain point cloud clusters, and the displacement and angle deviation of each point cloud cluster are analyzed to determine the suspected vehicle targets.

[0010] The system performs three-dimensional contour analysis, road surface contact feature analysis, and lane occupancy analysis on suspected vehicle targets to identify the actual target vehicle.

[0011] Compared with existing technologies, this invention has the following advantages: First, it constructs a static background reference point cloud by correcting the point-by-point ranging error under a static environment with empty lanes, correcting the inherent ranging deviation of the equipment from the source and providing a stable and unified spatial reference. Then, it performs directional cropping of the measured point cloud through the effective detection area of ​​each lane, eliminating redundant noise points across lanes and invalid data in non-detection areas, reducing computational overhead. Next, it eliminates suspended scattering noise points under severe weather conditions, solving the problem of target missed detection and false detection caused by weather interference while completely preserving the characteristics of the real vehicle point cloud. Then, it aggregates discrete effective sampling points into point cloud clusters, and completes the screening of suspected vehicle targets by displacement and directional angle deviation, narrowing the target range for final accurate verification. Finally, through verification of three-dimensional contour specifications, road surface contact features, and lane space ratio, it improves the accuracy and stability of vehicle detection, while also enhancing anti-interference capabilities. Attached Figure Description

[0012] Figure 1 This is a flowchart of the signal processing method for a single-sided laser vehicle detector at a highway toll station according to the present invention;

[0013] Figure 2 This is a flowchart illustrating the process of obtaining candidate valid point clouds in the signal processing method of a single-sided laser vehicle detector at a highway toll station according to the present invention.

[0014] Figure 3 This is a flowchart illustrating the signal processing method for determining suspected vehicle targets in the single-sided laser vehicle detector at a highway toll station according to the present invention. Detailed Implementation

[0015] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. Please refer to the accompanying drawings. Figure 1 The present invention provides a technical solution: a signal processing method for a single-sided laser vehicle detector at a highway toll station, comprising the following steps: S1, performing distance measurement error correction point by point based on the original point cloud under a static environment of empty lanes to obtain a static background reference point cloud.

[0016] In order to establish a standardized three-dimensional spatial benchmark for highway toll station lanes under empty and interference-free conditions, the specific process of obtaining the static background benchmark point cloud is as follows: S101, calculate the difference between the measured distance value and the corresponding theoretical distance value of each original point cloud to obtain the distance deviation value.

[0017] S102. Perform reverse compensation correction on the original point cloud based on the ranging deviation value to obtain the corrected point cloud. That is, when the ranging deviation value is positive, subtract the corresponding ranging deviation value from the measured distance value of the sampling point; when the ranging deviation value is negative, add the absolute value of the corresponding ranging deviation value to the measured distance value of the sampling point.

[0018] S103. Perform mean filtering and fusion on the point clouds after correction of multiple frames to obtain a static background reference point cloud. The number of frames acquired shall not be less than 2.

[0019] The measured distance values ​​of the original point cloud were obtained by controlling a single-sided laser vehicle detector to collect data at the inherent scanning frame rate and inherent point frequency under static scenarios where the lane is empty, there is no dynamic interference, and there is no severe weather. Each frame of the original point cloud contains the three-dimensional coordinates, measured distance values, and echo intensity of each sampling point.

[0020] The theoretical distance value is obtained by selecting fixed reference points with known spatial coordinates within the lane (including the edge of the toll island, guardrail posts, and the center point of fixed road markings). The spatial coordinates of the fixed reference points are measured and calibrated by a laser rangefinder. Based on the TOF laser ranging principle, the theoretical distance value corresponding to each fixed reference point is calculated according to the known straight-line distance between the fixed reference point and the optical center of the detector. The calculation process is existing technology and will not be described in detail here.

[0021] It should be noted that the process of mean filtering and fusion of multi-frame corrected point clouds is as follows: All corrected point clouds are spatially divided using a 0.05m × 0.05m × 0.05m voxel grid, and the sampling points of all corrected point clouds are mapped to the corresponding voxel grids. For all sampling points within each voxel grid, the mean of the three-dimensional coordinates and the mean of the echo intensity are calculated as feature values ​​of the corresponding voxel grid, generating a single-frame standardized point cloud. Inter-frame weighted fusion is then performed on all standardized point clouds, with the weighting coefficients decreasing sequentially from the most recent frame to the oldest frame. The weighting coefficients can be determined using the exponential decay method, controlling the decay rate of historical frame weights with a fixed decay coefficient. Using the weight of the latest frame as the baseline, the weight is multiplied by the decay coefficient for each previous frame. Finally, all weights are normalized so that the sum of the normalized weighting coefficients is 1.

[0022] The static background reference point cloud obtained by multi-frame correction, voxelized mean filtering and time-weighted fusion can reduce the impact of device ranging error and environmental noise, forming a stable and uniform static background template. When compared with the measured effective point cloud, it can accurately remove static points, thus avoiding the static background from being misjudged as a moving target.

[0023] S2. Based on the effective detection area corresponding to each lane, the measured point cloud is cropped to obtain the effective point cloud corresponding to each lane.

[0024] The process of obtaining the effective detection area is as follows: based on the physical planning boundary of the lanes and combined with the width of the standard toll lane, the entire detection area is divided into equal-width and continuous intervals to obtain the effective detection area corresponding to each lane. For example, the width of the standard toll lane is 3.5m.

[0025] In this embodiment, the global measured point cloud collected by the laser vehicle detector is spatially cropped according to the effective detection area, retaining only the measured point cloud falling within the effective detection area of ​​each lane, thus obtaining an effective point cloud corresponding to each lane. This can eliminate irrelevant environmental points outside the lane, redundant interference points, and stray points outside the detection area, reducing the amount of data processed by the point cloud. At the same time, it avoids cross-lane interference from affecting the target detection results, ensuring that the point cloud data of each lane is pure, independent, and focused on the effective detection range.

[0026] S3. Calculate the distance between the effective point cloud and the static background reference point cloud, remove the static point cloud, and obtain the dynamic point cloud.

[0027] Considering that the measured effective point cloud contains a large number of fixed static background points such as road surface, toll island, and guardrail, which will interfere with subsequent vehicle target recognition and easily cause false target detection, it is necessary to remove the fixed static background points. The specific process is as follows: S301, map both the effective point cloud and the static background reference point cloud to the above-mentioned 0.05m×0.05m×0.05m voxel grid.

[0028] S302. For each valid point cloud sampling point in the voxel grid, calculate its three-dimensional Euclidean distance to the static background reference point cloud in the corresponding voxel grid.

[0029] S303. If the three-dimensional Euclidean distance is not greater than the absolute value of the nominal ranging accuracy, it is determined to be a static point cloud and discarded to obtain a dynamic point cloud. The nominal ranging accuracy is the calibrated value, that is, the nominal ranging accuracy of the single-sided laser vehicle detector, for example, ±3cm.

[0030] This embodiment relies on the uniformly calibrated 0.05m×0.05m×0.05m voxel grid established during the initial construction of the static background reference point cloud to achieve a one-to-one mapping of the spatial positions of the effective point cloud and the static background reference point cloud, avoiding comparison errors caused by spatial misalignment. Furthermore, by calculating the three-dimensional Euclidean distance between the effective point cloud sampling points and the static background reference point cloud within the same voxel grid, and using the device's nominal ±3cm ranging accuracy as the threshold for determining dynamic and static points, the risk of misjudgment caused by fluctuations in the device's own ranging accuracy is avoided, and all static background interference points can be completely eliminated.

[0031] S4. Screen meteorological interference points from the dynamic point cloud to obtain candidate valid point clouds.

[0032] Considering that dynamic point clouds after static background removal may still contain meteorological floating scattering interference points such as rain, snow, and fog, which can easily lead to false target clustering and false or missed vehicle detections, therefore, if Figure 2 As shown, the specific process of obtaining the candidate valid point cloud is as follows: S401, based on the three-dimensional coordinates and laser echo intensity data of each sampling point in the dynamic point cloud corresponding to multiple consecutive frames, calculate the following feature parameters respectively: single-point echo intensity fluctuation.

[0033] Pearson correlation coefficient of the neighborhood voxel density corresponding to the sampling point.

[0034] Single-point coordinate inter-frame deviation.

[0035] S402. Calculate the arithmetic mean and standard deviation of the entire dynamic point cloud on the above three feature parameters respectively.

[0036] S403. For each sampling point in the dynamic point cloud, determine in sequence whether its single-point echo intensity fluctuation, neighborhood voxel density Pearson correlation coefficient, and single-point coordinate inter-frame deviation meet preset conditions; the preset conditions are that the arithmetic mean and standard deviation of each feature value relative to the corresponding features of the entire dynamic point cloud meet the set threshold range.

[0037] S404. If any feature parameter of a certain sampling point satisfies the preset condition, then the sampling point is determined to be a meteorological interference point and removed from the dynamic point cloud.

[0038] S405. Output the remaining point cloud data after removing meteorological interference points as candidate valid point clouds.

[0039] This embodiment anchors the differences between meteorological interference points and actual vehicle surface sampling points in terms of optical reflection characteristics, spatial distribution continuity, and motion consistency. It selects three non-overlapping and highly complementary dimensions: single-point echo intensity fluctuation, Pearson correlation coefficient of neighborhood voxel density, and single-point coordinate inter-frame deviation. Based on continuous multi-frame (exemplary, no less than 5 frames) point cloud data, it completes the quantitative calculation of features in each dimension, realizes the screening and elimination of meteorological interference points, and finally obtains candidate valid point clouds. It can eliminate interference points under meteorological conditions without damaging the features of the actual vehicle target point cloud.

[0040] The specific preset conditions are as follows: the inter-frame deviation of a single point coordinate exceeds the range of its corresponding arithmetic mean ± 3 standard deviations; the fluctuation of the single point echo intensity is greater than its corresponding arithmetic mean plus 3 standard deviations; and the Pearson correlation coefficient of the neighborhood voxel density is less than its corresponding lower threshold. The lower threshold is the maximum value between the arithmetic mean minus 3 standard deviations and -1. If a dynamic point cloud sampling point meets any of the above three conditions, it is determined to be a meteorological interference point and is removed.

[0041] Furthermore, the process of calculating the single-point echo intensity fluctuation, the Pearson correlation coefficient of the neighborhood voxel density corresponding to the sampling point, and the single-point coordinate inter-frame deviation is as follows: for each sampling point, the three-dimensional coordinate Euclidean distance between adjacent frames is calculated, and the arithmetic mean of the three-dimensional coordinate Euclidean distance is used as the single-point coordinate inter-frame deviation.

[0042] Calculate the difference between the maximum and minimum values ​​of the laser echo intensity for all frames, and use this difference as the single-point echo intensity fluctuation for that sampling point.

[0043] The dynamic point cloud is mapped onto a voxel grid, and the target voxel to which each sampling point belongs is determined. The number of sampling points of the target voxel in each frame in consecutive frames is counted to obtain the neighborhood voxel density sequence. The frame number independent variable sequence is constructed. The correlation between the neighborhood voxel density sequence and the independent variable sequence is calculated based on the Pearson correlation coefficient calculation formula to obtain the Pearson correlation coefficient of the neighborhood voxel density corresponding to the sampling point.

[0044] Among them, the inter-frame deviation of single-point coordinates reflects the inter-frame stability of the sampling point's position in three-dimensional space and the continuity of its motion pattern. When the inter-frame deviation of single-point coordinates is stable within a reasonable range and the inter-frame fluctuation is minimal, it indicates that the motion of the sampling point is continuous and smooth, and conforms to the driving pattern of vehicles within the lane, thus increasing the probability that the target is a real vehicle.

[0045] The fluctuation in single-point echo intensity reflects the inter-frame stability of the reflection characteristics of the target corresponding to the sampling point to the laser beam, and is an optical characteristic that distinguishes rigid solid targets from suspended scatterers. The larger the value, the more drastic the fluctuation in the laser reflection characteristics of the sampling point, and the higher the probability that the corresponding target is a weather scatterer.

[0046] The Pearson correlation coefficient of neighborhood voxel density reflects the degree of linear correlation between the point cloud density of the target voxel where the sampling point is located and the change over consecutive frames. It characterizes the inter-frame continuity and spatial clustering stability of the point cloud distribution and is a spatial distribution feature that distinguishes between rigid targets with continuous overall motion and discrete interference points that arise and disappear randomly. Its theoretical value is [-1, 1]. In the toll station scene, the real vehicle is a rigid body with overall motion. The voxel region occupied by its body moves continuously with the vehicle's movement. The number of sampling points within the voxel will not change randomly. The voxel density is highly correlated between frames, so the value is always close to 1. The closer the value is to 0 or even negative, the more it indicates that there is no linear correlation between the point density of the voxel in two consecutive frames. The spatial point cloud distribution is random and disordered, and the higher the probability that the target is a meteorological interference point.

[0047] S5. Cluster the candidate valid point clouds to obtain point cloud clusters, and perform displacement and angle deviation analysis on each point cloud cluster to determine suspected vehicle targets.

[0048] like Figure 3As shown, since the candidate valid point cloud obtained after static background removal and meteorological interference screening is still discrete and independent single-point data, it cannot be mapped to specific entity targets and is difficult to support subsequent vehicle determination. Furthermore, the remaining scattered and isolated interference points are prone to false target misjudgment. Therefore, it is necessary to perform clustering processing on the candidate valid point cloud to obtain point cloud clusters. The specific process is as follows: S501, map the candidate valid point cloud to a 0.05m×0.05m×0.05m voxel grid and remove empty voxels that do not contain sampling points.

[0049] S502. Using a set multiple of the absolute value of the nominal ranging accuracy as the clustering neighborhood search radius, select any unclassified sampling point in the candidate valid point cloud as the initial clustering point. The set multiple can be manually adjusted according to the nominal ranging accuracy of the device and the actual scene, preferably 2 times, to balance the clustering integrity and the risk of over-segmentation.

[0050] S503. Search for all candidate valid point cloud sampling points within the search radius of the initial cluster point and assign them to the temporary cluster set corresponding to the current initial cluster point.

[0051] S504. Using each newly added sampling point in the temporary cluster set as an extension point, continue to search for unclassified candidate valid point cloud sampling points within the search radius of the corresponding cluster neighborhood, and iteratively add them to the temporary cluster set until no new sampling points are added to the temporary cluster set.

[0052] S505. After the iteration is completed, the temporary cluster set is determined as a point cloud cluster, and all sampling points in the point cloud cluster are marked as classified.

[0053] S506. Repeat the steps of selecting unclassified sampling points, constructing a temporary cluster set, iteratively expanding, and determining point cloud clusters until all sampling points in the candidate valid point cloud are classified, resulting in all point cloud clusters. If an unclassified sampling point has no other sampling points within the cluster neighborhood search radius, it is marked as an isolated point and removed.

[0054] In this embodiment, the candidate valid point cloud is first spatially divided and empty voxels are removed using a voxel mesh. This reduces the amount of invalid computation while ensuring the consistency of the spatial coordinate system throughout the process. Then, the nominal ranging accuracy setting multiple that adapts to the ranging error of the device itself is used as the search radius of the clustering neighborhood to avoid the incorrect splitting of associated sampling points of the same vehicle due to the inherent ranging fluctuations of the device. Finally, through the iterative region growing clustering method, discrete sampling points that are close to each other in the three-dimensional space and belong to the same rigid vehicle target are aggregated into point cloud clusters. This provides a complete target carrier for subsequent centroid displacement calculation, three-dimensional contour judgment, and road contact feature judgment based on point cloud clusters, and also removes residual scattered interference points that cannot form effective clusters.

[0055] Since compliant vehicles, as rigid entities, must be able to travel smoothly and at a constant speed along the lane direction, they differ from non-vehicle interference clusters such as residual splash debris after clustering, incompletely removed weather clumps, and cross-lane stray points. Therefore, suspected vehicle targets can be identified by displacement and angular deviation. The specific process is as follows: Calculate the difference in centroid displacement between adjacent frames and the angular deviation between the cluster displacement direction and the lane direction based on the three-dimensional coordinates of all sampling points in consecutive frames of each point cloud cluster.

[0056] If a cloud cluster at a certain point satisfies the condition that the difference in centroid displacement between adjacent frame clusters is within the nominal ranging accuracy range and the included angle deviation is an acute angle, it is determined to be a suspected vehicle target and retained.

[0057] In this embodiment, the overall motion pattern of the entire target is first quantified into calculable values ​​by calculating the centroid coordinates of the point cloud clusters frame by frame. Then, the stability of the target motion is quantified by the difference in the centroid displacement of adjacent frames. The directional compliance of the target motion is quantified by the angular deviation between the cluster displacement direction and the lane direction. Finally, the nominal ranging accuracy adapted to the measurement characteristics of the device itself is used as the threshold for judging motion stability, and the acute angle that conforms to the toll station passage rules is used as the boundary for judging directional compliance. From all the clustered point cloud clusters, targets with stable motion and compliant driving along the lane are selected as suspected vehicle targets. Interference clusters that do not conform to the vehicle motion characteristics are removed simultaneously. This can narrow the processing scope of the subsequent real vehicle judgment stage, reduce invalid calculations, and improve real-time processing efficiency.

[0058] Furthermore, the process of calculating the difference in centroid displacement and the included angle deviation between adjacent frame clusters is as follows: calculate the arithmetic mean of the coordinates of all sampling points within the current frame of the point cloud cluster, and use it as the centroid coordinates of that frame cluster.

[0059] Calculate the three-dimensional Euclidean distance between the centroid coordinates of adjacent cloud clusters in two separate frames to obtain the three-dimensional displacement of the centroid of each adjacent cluster. Subtract the three-dimensional displacement of the centroid of the previous adjacent cluster from the displacement of the centroid of the next adjacent cluster to obtain the difference in the centroid displacement of the adjacent clusters.

[0060] Based on the lane direction marked in the effective detection area of ​​each lane, determine the three-dimensional unit vector of the driving direction of each lane.

[0061] For the same cloud cluster, calculate the difference vector of cluster centroid coordinates between each two adjacent frames, and normalize each difference vector to obtain the unit vector of cluster displacement direction for the corresponding adjacent frames.

[0062] Based on the vector dot product formula, the angle between the unit vector of each cluster displacement direction and the corresponding lane direction unit vector is calculated, and the arithmetic mean of the angles of all adjacent frames is taken as the angle deviation.

[0063] The difference in centroid displacement between adjacent frame clusters reflects the stability and continuity of the target's motion state. It is a motion characteristic that distinguishes between rigid, uniformly moving vehicle targets and non-vehicle interference clusters with irregular, variable-speed motion. The smaller the absolute value, the more stable the target's speed and state, and the more consistent it is with the characteristics of a vehicle traveling in a toll station lane. When the absolute value is within the nominal ranging accuracy range of the equipment, it can be determined as a compliant target with stable motion.

[0064] The angular deviation between the cluster displacement direction and the lane direction quantifies the degree of deviation between the actual movement direction of the target corresponding to the point cloud cluster and the compliant driving direction stipulated by the lane. It is a directional feature that distinguishes vehicle targets traveling in the forward direction along the lane from non-vehicle interference clusters that cross, reverse, or move in an irregular direction. When the angular deviation is acute, it means that the target's movement direction is consistent with the lane's compliant driving direction. The smaller the value, the higher the degree of fit between the target's movement direction and the lane's driving direction, and the more closely it matches the characteristic of a vehicle traveling in a straight line along the lane.

[0065] S6. Perform three-dimensional contour judgment, road surface contact feature judgment, and lane ratio judgment on suspected vehicle targets to determine the real target vehicle.

[0066] For suspected vehicle targets obtained through previous screening using motion and orientation features, there may still be interfering targets that are not compliant vehicles, such as pedestrians, motorcycles, road debris, and suspended meteorological clumps that have not been completely eliminated. Therefore, it is necessary to further determine the vehicle based on its unique attributes. The specific process is as follows: S601, for the three-dimensional coordinates of all sampling points in the point cloud cluster corresponding to each suspected vehicle target, calculate the difference between the maximum and minimum values ​​of the X-axis, Y-axis, and Z-axis coordinates to obtain the length, width, and height of the point cloud cluster.

[0067] S602. A point cloud cluster whose length, width, and height all meet the proportional relationship of the outline of a vehicle traveling on a highway is deemed to have a qualified outline. For example, the proportional relationship of the outline of a vehicle traveling on a highway is: length 5m-18m, width 2m-2.5m, height 2m-4m, and the length-to-width ratio is not less than 2:1 and the height-to-width ratio is not greater than 2:1.

[0068] S603. Calculate the difference between the minimum Z-axis coordinate of the sampling point within the point cloud cluster of the suspected vehicle target and the Z-axis coordinate of the road reference. Point cloud clusters whose difference is within the nominal ranging accuracy range are judged as having qualified road contact.

[0069] S604. Calculate the ratio of the horizontal projection area of ​​the point cloud clusters of suspected vehicle targets onto the effective detection area to the horizontal area of ​​the effective detection area. Point cloud clusters whose ratio falls within the preset effective lane ratio range are considered to have a qualified lane ratio. The preset effective lane ratio range is 10%-90%.

[0070] S605. Suspected vehicles that simultaneously meet the requirements of conformity of outline, conformity of road surface contact, and conformity of lane ratio are identified as real target vehicles.

[0071] The length, width, and height of the point cloud cluster refer to the differences between the maximum and minimum coordinates of all sampling points within the point cloud cluster corresponding to the suspected vehicle target, along the X-axis (direction of lane travel), the Y-axis (perpendicular to lane direction), and the Z-axis (perpendicular to road surface). These three differences correspond to the target's longitudinal, lateral, and vertical dimensions in three-dimensional space, respectively, and are physical characteristics that distinguish compliant vehicles from other small or irregularly shaped moving targets. In the highway toll station detection scenario, only when all three differences are simultaneously within the range of standard highway vehicle outline dimensions stipulated by the state, and the length, width, and height ratios conform to the outline proportions of conventional passenger cars and freight cars, does it mean that the target's physical dimensions match those of a compliant vehicle, and the outline is deemed acceptable.

[0072] The difference between the minimum Z-axis coordinate and the road reference Z-axis coordinate quantifies the degree of contact between the lowest point of the target and the road surface, and is a spatial characteristic that distinguishes between suspended interference targets and grounded vehicles. When the absolute value of the difference is within the nominal ranging accuracy range of the equipment, it means that the lowest point of the target is in contact with the calibrated road surface, which conforms to the inherent physical characteristics of direct contact between vehicle tires and the road surface, and is judged as qualified road contact.

[0073] It should be noted that the calibration method for the road surface reference Z-axis coordinate is as follows: Under static conditions where the lane is unloaded, there is no dynamic interference, and there is no severe weather, at least 20 road surface sampling points are evenly selected within the effective detection area of ​​each lane. The Z-axis coordinates of each sampling point are collected by a single-sided laser vehicle detector, and their arithmetic mean is calculated as the road surface reference Z-axis coordinate of that lane. After calibration, this value is stored in the detector system as a fixed parameter.

[0074] The ratio of the projected area of ​​the horizontal plane to the effective detection area of ​​the horizontal plane quantifies the size and proportion of the space occupied by the target in the horizontal plane of the lane. It is a characteristic that distinguishes between scattered small targets in the lane and vehicles that are normally passing through the lane. When the ratio is within the preset effective lane proportion range, it means that the size of the space occupied by the target in the lane is in line with the lane occupancy characteristics of regular passing vehicles, and the lane proportion is judged to be qualified.

[0075] This embodiment determines whether the physical dimensions of the target conform to the national regulations for highway vehicle outline specifications by using three-dimensional contours, whether the target conforms to the landing and driving characteristics of wheeled vehicles in direct contact with the road surface by using road surface contact features, and whether the target conforms to the space occupancy rules of vehicles in a single lane by using lane occupancy. It does not require complex model training and global statistical calculations, and can complete the determination based on existing point cloud cluster three-dimensional coordinate data. It eliminates interference targets that conform to motion characteristics but do not conform to vehicle physical attributes, ensuring the accuracy and reliability of the detection results.

[0076] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0077] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0078] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0079] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0080] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A signal processing method for a single-sided laser vehicle detector at a highway toll station, characterized in that, Includes the following steps: The ranging error is corrected point by point based on the original point cloud under static conditions of lane vacancy to obtain static background reference point cloud. The effective point cloud for each lane is obtained by cropping the measured point cloud based on the effective detection area corresponding to each lane. The distance between the effective point cloud and the static background reference point cloud is calculated, and the static point cloud is removed to obtain the dynamic point cloud. Meteorological interference points are screened from dynamic point clouds to obtain candidate valid point clouds; The candidate valid point clouds are clustered to obtain point cloud clusters, and the displacement and angle deviation of each point cloud cluster are analyzed to determine the suspected vehicle targets. The system performs three-dimensional contour analysis, road surface contact feature analysis, and lane occupancy analysis on suspected vehicle targets to identify the actual target vehicle.

2. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of obtaining the static background reference point cloud is as follows: The difference between the measured distance value and the corresponding theoretical distance value of each original point cloud is calculated to obtain the ranging deviation value; The original point cloud is corrected by reverse compensation based on the ranging deviation value to obtain the corrected point cloud. Mean filtering and fusion are performed on the point clouds after correction of multiple frames to obtain the static background reference point cloud.

3. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of obtaining dynamic point clouds is as follows: Both the effective point cloud and the static background reference point cloud are mapped to a voxel mesh; For each valid point cloud sampling point within a voxel grid, calculate its three-dimensional Euclidean distance to the static background reference point cloud within the corresponding voxel grid. If the three-dimensional Euclidean distance is not greater than the absolute value of the nominal ranging accuracy, it is determined to be a static point cloud and discarded to obtain a dynamic point cloud.

4. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of obtaining candidate valid point clouds is as follows: Based on the three-dimensional coordinates and laser echo intensity data corresponding to each sampling point in the dynamic point cloud across multiple consecutive frames, the following feature parameters are calculated: Single-point echo intensity fluctuation; Pearson correlation coefficient of the neighborhood voxel density corresponding to the sampling point; Inter-frame deviation of single-point coordinates; Calculate the arithmetic mean and standard deviation of the entire dynamic point cloud on the above three feature parameters respectively; For each sampling point in the dynamic point cloud, the single-point echo intensity fluctuation, the Pearson correlation coefficient of the neighborhood voxel density, and the single-point coordinate inter-frame deviation are sequentially determined to meet the preset conditions. The preset conditions are that the arithmetic mean and standard deviation of each feature value relative to the corresponding features of the entire dynamic point cloud meet the set threshold range. If any feature parameter of a sampling point satisfies the preset condition, the sampling point is identified as a meteorological interference point and removed from the dynamic point cloud. The remaining point cloud data after removing meteorological interference points will be output as candidate valid point clouds.

5. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 4, characterized in that, The process of calculating the single-point echo intensity fluctuation, the Pearson correlation coefficient of the neighborhood voxel density corresponding to the sampling point, and the inter-frame deviation of the single-point coordinates is as follows: For each sampling point, the three-dimensional Euclidean distance between adjacent frames is calculated, and the arithmetic mean of the three-dimensional Euclidean distance is used as the inter-frame deviation of the single-point coordinates. Calculate the difference between the maximum and minimum values ​​of the laser echo intensity values ​​for all frames, and use this as the single-point echo intensity fluctuation for that sampling point; The dynamic point cloud is mapped onto a voxel grid, and the target voxel to which each sampling point belongs is determined. The number of sampling points of the target voxel in each frame in consecutive frames is counted to obtain the neighborhood voxel density sequence. The frame number independent variable sequence is constructed. The correlation between the neighborhood voxel density sequence and the independent variable sequence is calculated based on the Pearson correlation coefficient calculation formula to obtain the Pearson correlation coefficient of the neighborhood voxel density corresponding to the sampling point.

6. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of clustering candidate valid point clouds to obtain point cloud clusters is as follows: Using a set multiple of the absolute value of the nominal ranging accuracy as the search radius of the clustering neighborhood, any unclassified sampling point in the candidate valid point cloud is selected as the initial cluster point; Search all candidate valid point cloud sampling points within the search radius of the initial cluster point and assign them to the temporary cluster set corresponding to the current initial cluster point; Using each newly added sampling point in the temporary cluster set as an expansion point, continue to search for unclassified candidate valid point cloud sampling points within the search radius of the corresponding cluster neighborhood, and iteratively add them to the temporary cluster set until no new sampling points are added to the temporary cluster set. The temporary cluster set after iteration is determined as a point cloud cluster, and all sampling points in the point cloud cluster are marked as classified. Repeat the steps of selecting unclassified sampling points, constructing temporary cluster sets, iteratively expanding and determining point cloud clusters until all sampling points in the candidate valid point cloud are classified and all point cloud clusters are obtained.

7. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of identifying suspected vehicle targets based on displacement and angle deviation analysis of each point cloud cluster is as follows: The difference in centroid displacement between adjacent frames and the angular deviation between the cluster displacement direction and the lane direction are calculated based on the three-dimensional coordinates of all sampling points in consecutive frames of each point cloud cluster. If a cloud cluster at a certain point satisfies the condition that the difference in centroid displacement between adjacent frame clusters is within the nominal ranging accuracy range and the included angle deviation is an acute angle, it is determined to be a suspected vehicle target and retained.

8. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 7, characterized in that, The process of calculating the difference in centroid displacement and the included angle deviation between adjacent frame clusters is as follows: Calculate the arithmetic mean of the coordinates of all sampled points in the current frame for each point cloud cluster, and use it as the centroid coordinates of the corresponding frame cluster; Calculate the three-dimensional Euclidean distance between the centroid coordinates of two adjacent cloud clusters at the same point in each frame to obtain the three-dimensional displacement of the centroid of each adjacent cluster. Subtract the three-dimensional displacement of the centroid of the previous adjacent cluster from the three-dimensional displacement of the centroid of the next adjacent cluster to obtain the difference in the centroid displacement of the adjacent clusters. Based on the lane direction marked in the effective detection area of ​​each lane, determine the three-dimensional unit vector of the driving direction of each lane; For the same cloud cluster, calculate the difference vector of cluster centroid coordinates between each two adjacent frames, and normalize each difference vector to obtain the unit vector of cluster displacement direction for the corresponding adjacent frames. Based on the vector dot product formula, the angle between the unit vector of each cluster displacement direction and the corresponding lane direction unit vector is calculated, and the arithmetic mean of the angles of all adjacent frames is taken as the angle deviation.

9. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of identifying the actual target vehicle is as follows: For the three-dimensional coordinates of all sampling points within the point cloud cluster corresponding to each suspected vehicle target, calculate the difference between the maximum and minimum values ​​of the X-axis, Y-axis, and Z-axis coordinates respectively to obtain the length, width, and height of the point cloud cluster; A point cloud cluster whose length, width and height all meet the proportional relationship of the outline of vehicles traveling on the highway is judged to be qualified in terms of outline. Calculate the difference between the minimum Z-axis coordinate of the sampling point within the point cloud cluster of the suspected vehicle target and the Z-axis coordinate of the road reference. Point cloud clusters whose difference is within the nominal ranging accuracy range are judged as having qualified road contact. Calculate the ratio of the horizontal projection area of ​​the point cloud clusters of suspected vehicle targets onto the effective detection area to the horizontal area of ​​the effective detection area, and determine the point cloud clusters whose ratio is within the preset effective lane ratio range as having a qualified lane ratio. Suspected vehicles that simultaneously meet the requirements of conformity in outline, road surface contact, and lane occupancy will be identified as actual target vehicles.

10. The signal processing method for a single-sided laser vehicle detector at a highway toll station according to claim 1, characterized in that, The process of obtaining the effective detection area is as follows: Based on the physical planning boundary of the lanes and combined with the standard toll lane width, the entire detection area is divided into equal-width and continuous intervals to obtain the effective detection area for each lane.

Images