CN115236628B - Method for detecting residual cargoes in carriage based on laser radar - Google Patents

Abstract

The present invention discloses a method for detecting residual cargo in a carriage based on a laser radar, which comprises extracting carriage data from a single-frame raw data; extracting a carriage target point cloud, and performing tilt correction on the collected single-frame point cloud contour; obtaining the current speed, converting a coordinate system with the radar as the origin into a coordinate system with the bottom of the carriage as the center, splicing the single-frame point cloud to form an overall point cloud map of the carriage; reducing noise and streamlining the carriage point cloud data; smoothing the spliced carriage point cloud data; slicing the carriage point cloud data; projecting the point cloud between two adjacent slices, and extracting a cross-sectional contour; obtaining a cross-sectional area for the extracted cross-sectional contour, and multiplying it by the slice spacing to obtain the volume of the segment to the point cloud; and accumulating and calculating the volume of the residual cargo in the carriage. The present invention can automatically detect the residual cargo in the train carriage without stopping the train, without the need for workers to climb into the carriage for observation, thereby reducing the labor intensity of workers and improving the safety of work, and the automatic detection efficiency is high and the accuracy is good.

Description

A method for detecting residual cargo in a carriage based on laser radar

Technical Field

The present invention relates to a method for detecting residual cargo in a carriage, and in particular to a method for detecting residual cargo in a carriage based on a laser radar.

Background technique

Cargo residue in carriages is a common phenomenon for railway freight vehicles. Due to the shape and characteristics of the objects loaded in the carriages or due to weather and other reasons, the items loaded in the carriages cannot be completely unloaded during unloading operations. For the residual cargo, workers often need to climb onto the carriages with the help of tools to observe the residual cargo in the box, and then arrange a corresponding number of workers to enter the carriages for cleaning according to the amount of residue to avoid the phenomenon of loss of tons. Such manual inspection methods not only have safety hazards and affect the normal operation of trains, but also increase the labor intensity of workers and low work efficiency.

Summary of the invention

In view of the problems existing in the above-mentioned prior art, the present invention provides a method for detecting residual cargo in a carriage based on laser radar, which does not affect the operation of the train, is convenient for detecting residual objects, does not require workers to board the train to check, and has high safety and efficiency.

To achieve the above object, the present invention provides the following technical solution: a method for detecting residual cargo in a carriage based on a laser radar, a method for detecting residual cargo in a carriage based on a laser radar, characterized in that it comprises the following steps:

S10, extracting carriage data from the single-frame original data using a cluster analysis method based on Euclidean distance, and determining the front and rear boundaries of the carriage using a three-dimensional coordinate relationship to obtain the first frame and the last frame data information;

S20, extracting the carriage target point cloud by using a conditional filtering method, and performing tilt correction on the collected single-frame point cloud contour;

S30, using a pre-set radar to obtain the current vehicle speed, converting the coordinate system with the radar as the origin into a coordinate system with the bottom of the car as the center, and using a displacement fusion algorithm to stitch the single-frame point cloud to form an overall point cloud map of the car;

S40, using statistical filtering and voxel gridding method to reduce noise and simplify the car point cloud data;

S50, smoothing the spliced carriage point cloud data using a moving least squares method;

S60, slicing the car body point cloud data along the vehicle running direction;

S70, the point cloud between two adjacent slices is projected, and the cross-sectional contour is extracted using the alpha algorithm and the ray 360-degree algorithm;

S80, for the extracted cross-sectional contour, the shoelace theorem is used to obtain the cross-sectional area, which is multiplied by the slice spacing to obtain the point cloud volume of the segment; the volume of the remaining cargo in the carriage is calculated by accumulation.

Furthermore, the conditional filtering and contour correction of the point cloud in step S20 includes: selecting the stationary point cloud of the frame scan according to the frame data obtained by scanning, finding the same column of point cloud data, analyzing its coordinate values, fitting and calculating the deflection angle, and performing rotation correction on the overall point cloud data.

Furthermore, the coordinate system conversion and point cloud stitching in step S30 include: converting the coordinate system with the radar as the origin into the coordinate system with the car as the origin, determining the coordinate value conversion relationship according to the radar installation position relationship, and completing the coordinate system conversion; based on the point cloud stitching method of mobile displacement fusion, by assigning a new Z-axis coordinate value to the point cloud, all frame point cloud data are uniformly corrected to the same frame coordinate system to complete the point cloud stitching.

Furthermore, the method of statistical filtering and voxelization gridding in step S40 includes: the average spacing distribution relationship from the sampling point to the neighborhood point conforms to the Gaussian distribution function, and the outlier point cloud data is eliminated by setting a reasonable threshold; the point cloud data is downsampled using the voxelization gridding method, and the point cloud data falls into a grid of a specified size. For each grid, only the point cloud closest to the center of the grid is retained, and the remaining point clouds in the grid are deleted, thereby achieving the effect of point cloud simplification.

Furthermore, the slicing of the point cloud data in step S60 includes: intercepting the car body point cloud data with a cut plane parallel to the car body cross section at a certain interval along the vehicle running direction to form a segment of scattered point cloud data. The larger the interval, the more point cloud segments are cut, and vice versa.

Furthermore, the step S70 of extracting the point cloud contour includes: projecting the point clouds between adjacent slices, using the alpha algorithm to determine the approximate boundary contour; and then using the ray 360-degree scanning algorithm to perform secondary discrimination on the boundary contour, and eliminating boundary point cloud data that does not meet the conditions, thereby improving the accuracy of the boundary contour.

Furthermore, the calculation of the point cloud volume in step S80 includes: using the shoelace theorem to obtain the point cloud cross-sectional area of the irregular polygon, multiplying it by the slice interval to obtain the volume of the segment of point cloud data, and obtaining the overall point cloud volume of the carriage by accumulation; subtracting the point cloud volume of the empty carriage from the volume of the carriage when there is residual cargo, to obtain the volume of the residual cargo in the carriage.

Compared with the prior art, the present invention can automatically detect the residual cargo in the train compartment without stopping the train, without the need for workers to climb into the compartment to observe, thereby reducing the labor intensity of workers and improving work safety, and the automatic detection efficiency is high and the accuracy is good; the present invention can not only identify the area and volume of residual cargo in the railway compartment, but also can detect materials such as loose sand and loose stones loaded on open road trucks. According to the density of the loaded materials, the mass of the transferred objects can be calculated to determine the overloading situation of the truck.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flow chart of the present invention;

FIG2 is a diagram of carriage data according to an embodiment of the present invention;

FIG3 is a diagram showing a carriage boundary determination according to an embodiment of the present invention;

FIG4 is a diagram showing the causes of the tilt of the car point cloud contour according to an embodiment of the present invention;

FIG5 is a single-frame point cloud data correction diagram according to an embodiment of the present invention;

FIG6 is a schematic diagram of a coordinate system conversion principle according to an embodiment of the present invention;

FIG7 is a point cloud image of an original carriage spliced based on movement displacement according to an embodiment of the present invention;

FIG8 is a diagram showing the voxelized grid method and point cloud smoothing results according to an embodiment of the present invention;

FIG9 is a cross-sectional view of an empty carriage according to an embodiment of the present invention;

FIG10 is a cross-sectional view of the carriage profile when residues exist in the embodiment of the present invention;

FIG. 11 is a diagram showing the principle of calculating the area of an irregular polygon according to an embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

As shown in FIG1 , the present invention provides a technical solution, a method for detecting residual cargo in a carriage based on a laser radar:

S10, extracting carriage data from the single-frame original data using a cluster analysis method based on Euclidean distance, and determining the front and rear boundaries of the carriage using a three-dimensional coordinate relationship to obtain the first frame and the last frame data information;

The initial point cloud data is preprocessed using a clustering analysis algorithm based on Euclidean distance. The clustering conditions are: the search radius of the nearest neighbor search is 0.2 meters, the minimum number of cluster points is 1000, and the maximum number of cluster points is 5000. The specific implementation steps are as follows:

(1) Find a point P ₁₀ in space, find the n points closest to it in kdTree, and determine the distances from these n points to P _10. Put points P ₁₂ , P ₁₃ , P ₁₄ , etc. whose distances are less than the threshold r into class Q;

(2) Find a point P ₁₂ in Q(P ₁₀ ) and repeat (1);

(3) Find a point in Q (P ₁₀ , P ₁₂ ), repeat (1) to find P ₂₂ , P ₂₃ , P ₂₄ . . . and put them all into Q;

(4) When no new points can be added to Q, the search is completed.

As shown in Figures 2 and 3, if the radar XOY plane is completely perpendicular to the train running direction, assuming that the width of a carriage is 3.3 meters and the height is 2.8 meters, in the carriage boundary judgment, point M is the highest point of the carriage, and the coordinates of point M are (x _m , y _max , z _m ), and point N is the lowest point of the carriage, and the coordinates of point N are (X _n , y _min , z _n ). According to the three-dimensional coordinate relationship, it can be known that:

y _max -y _min = 2.8,

Among them, y _max is the y coordinate of the highest point M, y _min is the y coordinate of the lowest point N, the height of the car is 2.8 meters, the car point cloud data is processed, and the left and right side walls of the car are hidden by 0.5 meters each. If the y _max and y _min of the hidden point cloud data change slightly, that is, the difference between y _max and y _min is greater than a certain threshold t, then the current car data is judged as car boundary data, otherwise it is judged as non-car boundary data. When the car boundary information appears for the first time, start recording data information; when the car data appears for the last time, stop recording data information to obtain a complete car data information.

S20, extracting the target point cloud of the carriage by using a conditional filtering method, and performing a tilt correction on the contour of the collected single-frame point cloud data;

If the radar XOY plane is completely perpendicular to the running direction of the train, then the laser points scanned in the same column should be on the same horizontal line, as shown in FIG4(a); if the radar XOY plane is not perpendicular to the running direction of the train and there is an inclined angle, then the points in the same column of data will also be inclined, as shown in FIG4(b); the method based on static point cloud correction in the present invention has the following steps:

(1) Randomly select a frame of point cloud data from the measurement scan, select a stationary object around the scanned object as a reference system (the wall next to the track is selected in the present invention), and extract the overall point cloud data of the reference system;

(2) Perform inverse calculation based on the three-dimensional coordinates of the point cloud data to obtain the column number of the point cloud in the radar scan;

(3) Extract the data Q = {P ₀ , P ₁ ..P ₁₅ } in the same column. The X and Y coordinate values of these data are approximately the same, and their Z axis data gradually changes according to a certain trend;

(4) Fitting a straight line according to the scattered point image of the Z-axis coordinate, calculating the equation of the fitted line, and using atan to obtain the angle θ _z relative to the Z-axis;

(5) Determine the rotation matrix and rotation direction through the angle θ, and implement the rotation transformation on the frame point cloud.

As shown in FIG5 , the white point cloud is the data after correction, and the gray point cloud is the data before correction.

S30, using a pre-set radar to obtain the current vehicle speed, converting the coordinate system with the radar as the origin into a coordinate system with the bottom of the car as the center, and using a displacement fusion algorithm to stitch the single-frame point cloud to form an overall point cloud map of the car;

As shown in Figure 6, point A is a scanning point in the car, h is the vertical distance between the laser radar and the bottom of the car. In the radar coordinate system, the coordinates of point A are (x, y, z) after the internal processing of the laser radar. After the coordinate system conversion, _O1 is the origin of the coordinates, and _O1 is on the same straight line. The coordinates of point A are ( _x1 , _y1 , _z1 ). According to the position geometry relationship,

As shown in Figure 7, this application assigns a new Z value to the Z-axis coordinate of the scanned point cloud based on the method of mobile displacement fusion compensation. According to the total number of frames of the scanned point cloud, each frame of point cloud data is translated in the direction of the train's advance in accordance with the scanning order to obtain a complete point cloud image based on the car coordinate system. Assuming that the train moves at a constant speed of V, the operating frequency of the laser radar is 10Hz, and the distance the train travels along the Z-axis direction in 0.1 seconds is V/10, the point cloud data is compensated for displacement:

_Zij ′＝ _Zij ±(ni)*v/10

Among them, _Zij ′ represents the Z coordinate value of the jth point in the i-th frame point cloud data after displacement compensation, _Zi represents the Z coordinate value of the jth point in the i-th frame point cloud data in the original coordinate system, n is the number of recorded point cloud frames, and i is the current frame number. When the Z-axis direction of the laser radar is the same as the running direction of the train, the above formula takes the "+" sign, otherwise, it takes the "-" sign.

S40, using statistical filtering and voxel gridding method to reduce noise and simplify the car point cloud data;

As shown in Figure 8, the statistical filtering steps are as follows:

(1) Establish the point cloud topological structure relationship for the target point cloud through the KD-tree tree;

(2) Find the neighborhood of the sampling point through the index and calculate the average Euclidean distance from each sampling point to the points in the neighborhood;

(3) Calculate the average distance from all points in the point cloud dataset to their neighbors;

(4) The obtained distance distribution conforms to the Gaussian function, so the mean μ and standard deviation σ are calculated;

(5) Set a threshold to remove noise data. From the distribution characteristics of the Gaussian function, we know that all points in the range of (μ-σ*std, μ+σ*std) belong to the valid point cloud, where std is called the standard deviation multiple, which is used to adjust the threshold range. If the value of a sampling point d _i is greater than μ+σ*std or less than μ-σ*std, it is determined to be a noise point and filtered.

The voxelized grid selects a grid size of 0.1m to simplify the point cloud data.

S50, smoothing the spliced carriage point cloud data using the moving least squares method; the basic steps of the moving least squares method are as follows:

(1) Input the fitting data point area and perform grid operation on the input area;

(2) Determine the range of the area affected by the grid point x and the number of nodes affected within the range; calculate the node value of the grid point according to the formula;

(4) Traverse all grid points and repeat steps (2) and (3);

(5) Connect the values calculated at the grid points to form a smooth curve.

S60, slicing the car body point cloud data along the vehicle running direction;

Since the train's forward direction is the positive half of the Z axis, the Z axis is selected as the cutting direction to divide the car model point cloud into several segments. The density of the point cloud is limited. It is difficult to form the shape outline of the point cloud simply by relying on a thin plane, so this plane needs to have a certain "thickness". The two adjacent slices just form an equidistant point cloud segment, so the point cloud between the two slice planes can be directly projected to obtain the outline of the segment:

Among them, i is the sequence number of the slice, _Zi is the position of the cutting plane, l is the interval distance of the slice, Z _min and Z _max represent the minimum and maximum values of the point cloud coordinates in the Z-axis direction, respectively, and n is the number of cutting planes.

S70, the point cloud between two adjacent slices is projected, and the cross-sectional contour is extracted using the alpha algorithm and the ray 360-degree algorithm;

The generated point cloud is projected on the XOY plane, and the alpha algorithm is used to extract the approximate point cloud outline. Then, the outline is screened by the ray 360-degree algorithm, as shown in Figures 9 and 10. The steps of the ray 360-degree algorithm are as follows:

(1) Calculate the center of gravity of the point cloud O

(2) Pick any point P ₀ (x ₀ , y ₀ ) as the starting scanning point, connect OP ₀ as the reference scanning ray, and scan the remaining data points in the counterclockwise direction along the scanning line;

(3) Calculate the angles θ _i between the remaining scanning points and the reference line;

(4) Sort the points according to the size of the required angle _θi , and then connect the point clouds in order to form the outline of the point cloud.

S80, for the extracted cross-sectional contour, the shoelace theorem is used to obtain the cross-sectional area, which is multiplied by the slice spacing to obtain the point cloud volume of the segment; by accumulation, the volume of the remaining cargo in the carriage is calculated.

As shown in Figure 11, the shoelace theorem is to calculate the area of a closed image through the determinant when the vertex coordinates of the polygon are known. It is stipulated that the irregular cross section is composed of n vertices P ₁ , P ₂ ...P _n . According to the ray scanning method, the sorted points are connected end to end in a counterclockwise order to form a closed polygon denoted as P _i . From the above derivation process, it can be seen that the calculated cross-sectional area A _i is:

Wherein, _Ai is the area of the i-th polygonal cross section, the coordinates of the polygon vertex _Pi are ( _xi , _yi ), _xn = _x1 , _yn = _y1 .

It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any slight modification, equivalent substitution and improvement made to the above embodiment based on the technical essence of the present invention should be included in the protection scope of the technical solution of the present invention.

Claims (6)

1. A method for detecting residual cargo in a carriage based on laser radar, characterized in that it comprises the following steps: S10, extracting carriage data from the single-frame original data using a cluster analysis method based on Euclidean distance, and determining the front and rear boundaries of the carriage using a three-dimensional coordinate relationship to obtain the first frame and the last frame data information; S20, extracting the carriage target point cloud by using a conditional filtering method, and performing tilt correction on the collected single-frame point cloud contour; S30, using a pre-set radar to obtain the current vehicle speed, converting the coordinate system with the radar as the origin into a coordinate system with the bottom of the car as the center, and using a displacement fusion algorithm to stitch the single-frame point cloud to form an overall point cloud map of the car; S40, using statistical filtering and voxel gridding method to reduce noise and simplify the car point cloud data; S50, smoothing the spliced carriage point cloud data using a moving least squares method; S60, slicing the car body point cloud data along the vehicle running direction; S70, the point cloud between two adjacent slices is projected, and the cross-sectional contour is extracted using the alpha algorithm and the ray 360-degree algorithm; S80, for the extracted cross-sectional profile, the shoelace theorem is used to obtain the cross-sectional area, which is multiplied by the slice spacing to obtain the volume of the point cloud of the segment; the volume of the remaining cargo in the carriage is calculated by accumulation; The calculation of the point cloud volume in step S80 includes: using the shoelace theorem to obtain the point cloud cross-sectional area of the irregular polygon, multiplying it by the slice interval to obtain the volume of the segment of point cloud data, and obtaining the overall point cloud volume of the carriage by accumulation; subtracting the point cloud volume of the empty carriage from the volume of the carriage when there is residual cargo, to obtain the volume of the residual cargo in the carriage. 2. According to the method of detecting residual cargo in a carriage based on laser radar according to claim 1, it is characterized in that the conditional filtering and contour correction of the point cloud in step S20 includes: selecting the static point cloud scanned by the frame data obtained by scanning, finding the same column of point cloud data, analyzing its coordinate values, fitting and calculating the deflection angle, and performing rotation correction on the overall point cloud data. 3. According to the method of detecting residual cargo in a carriage based on laser radar according to claim 1, it is characterized in that the coordinate system conversion and point cloud splicing in step S30 include: based on the coordinate system with the radar as the origin, the coordinate system is converted into the coordinate system with the carriage as the origin, and the coordinate value conversion relationship is determined according to the radar installation position relationship to complete the coordinate system conversion; based on the point cloud splicing method of mobile displacement fusion, by assigning a new Z-axis coordinate value to the point cloud, all frame point cloud data are uniformly corrected to the same frame coordinate system to complete the point cloud splicing. 4. According to the method of claim 1, the method of detecting residual cargo in a carriage based on laser radar is characterized in that the statistical filtering and voxel grid method in step S40 includes: the average spacing distribution relationship from the sampling point to the neighborhood point conforms to the Gaussian distribution function, and the outlier point cloud data is eliminated by setting a threshold; the point cloud data is down-sampled using the voxel grid method, and the point cloud data falls into a grid of a specified size, and each grid only retains the point cloud closest to the grid center, and the remaining point clouds in the grid are deleted. 5. According to the method of claim 1, which is based on laser radar to detect residual cargo in a carriage, it is characterized in that the slicing of the point cloud data in step S60 includes: intercepting the point cloud data of the carriage at intervals along the direction of vehicle movement with a section parallel to the cross section of the carriage to form segments of scattered point cloud data. 6. According to the method of claim 1, the laser radar-based method for detecting residual cargo in a carriage is characterized in that extracting the point cloud contour in step S70 includes: projecting the point cloud between adjacent slices and determining the approximate boundary contour using an alpha algorithm; then using a ray 360-degree scanning algorithm to perform a secondary judgment on the boundary contour and eliminate boundary point cloud data that does not meet the conditions.