Infrared anti-collision control system of maintenance working unit

Abstract

The invention discloses an infrared anti-collision control system of a maintenance working unit, and relates to the technical field of anti-collision control systems of the maintenance working unit, the infrared anti-collision control system obtains reflection infrared signals through an infrared sensor, classifies the reflection infrared signals through SNN-DBSCAN, finds out clusters corresponding to large obstacles, and provides an anti-collision control method for the maintenance working unit in order to more comprehensively obtain information of the large obstacles. The laser radar is used for collecting actual information of other sides of the large obstacle. The method comprises the following steps: projecting a large obstacle to a horizontal plane, describing a minimum convex polygon containing each point of the projection boundary of the large obstacle through a Graham scanning algorithm, calculating the area of the minimum convex polygon by utilizing a Gaussian area formula, comparing the area of the minimum convex polygon with the area occupied by a maintenance working unit, preliminarily judging whether a collision risk exists or not, and if not, judging whether the collision risk exists or not. And the road width is introduced for more detailed judgment, so that the problem of collision of the maintenance working unit caused by inaccurate judgment is remarkably reduced.

Description

Technical Field

[0001] The invention relates to the technical field of anti-collision control systems for maintenance work units, in particular to an infrared anti-collision control system for a maintenance work unit. Background Art

[0002] In the fields of road maintenance and construction, the working environment of maintenance crews is complex and changeable, and they often encounter various obstacles. Among them, large obstacles pose a serious threat to the safe operation of maintenance crews. When a large obstacle appears in front of the crew, it is crucial to accurately judge whether the crew can pass safely. Once the judgment is wrong, it is very likely to cause a collision accident.

[0003] When traditional infrared sensors are used to illuminate large obstacles, the collected reflected signals contain limited information due to the influence of the infrared sensor itself and the position of the large obstacle. When relying solely on the point cloud data of the lidar for analysis, it is difficult to accurately determine the boundaries and shape characteristics of large obstacles with irregular shapes, and thus it is impossible to accurately infer whether the maintenance work team can pass.

[0004] In addition, existing anti-collision systems often lack comprehensive consideration of key factors such as the area of ​​large obstacles and the relative position of maintenance work units and large obstacles. In actual operations, it is not enough to just know the existence of large obstacles. It is also necessary to deeply analyze the size of the area occupied by large obstacles and maintenance work units, whether the road width can accommodate the maintenance work unit to change lanes, etc., in order to make scientific and reasonable decisions and ensure the safe and efficient operation of the maintenance work unit. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an infrared anti-collision control system for a maintenance work unit, which solves the problem that the maintenance work unit cannot accurately judge whether it can pass through a large obstacle.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an infrared anti-collision control system for a maintenance work unit, comprising:

[0007] The information collection module installs an infrared sensor matrix and a laser radar with adjustable sensing angle in front of the maintenance work unit, which emits infrared rays and electromagnetic waves to large obstacles at different preset angles within a time T, and collects multiple groups of reflected infrared signals and target echoes;

[0008] Infrared signal processing module, the SNN-DBSCAN density clustering algorithm is used to cluster the reflected infrared signals, and the clusters corresponding to large obstacles are obtained according to the average intensity and angle standard deviation;

[0009] The target echo processing module constructs a probability density function using the kernel density estimation method according to the cluster corresponding to the large obstacle, substitutes the point cloud data into the probability density function, and determines whether it belongs to the large obstacle;

[0010] The information integration module projects the large obstacle onto the horizontal plane, converts its boundary points into rectangular coordinates, determines the minimum convex polygon containing all boundary points through the Graham scan algorithm, and calculates the area A1 of the minimum convex polygon using the Gaussian area formula;

[0011] The collision judgment module, if A1 < A2, initially judges that the maintenance work crew will not collide. Substitute the road width W. If W >= max{Lyang, Wyang} + 2S + L, it is judged that the maintenance work crew can safely pass the large obstacle; otherwise, it is judged that the maintenance work crew cannot safely pass the large obstacle, and a collision warning is triggered. Here, A2 represents the area of the maintenance work crew, L is the longest horizontal span of the minimum convex polygon, and Wyang and Lyang are the length and width of the maintenance work crew.

[0012] As a further solution of the present invention, the specific steps of clustering the reflected infrared signals using the SNN-DBSCAN density clustering algorithm are as follows:

[0013] Perform Min-Max normalization processing on the reflected infrared signal data, and calculate the Euclidean distance d between each data point;

[0014] According to the set nearest neighbor search radius r, find the set of nearest neighbors of each point;

[0015] For any two data points Pi and Pj, count the number num of their shared nearest neighbors;

[0016] If the shared nearest neighbor similarity num between two data points >= minPt, it is considered that they are density-related, and it is judged that they belong to the same cluster;

[0017] For all data points, gradually form different clusters by continuously merging points with sufficient shared nearest neighbor relationships.

[0018] As a further solution of the present invention, the specific rule for judging the cluster where the large obstacle is located is as follows:

[0019] Calculate the average intensity Iavg, average time delay tavg, and angular standard deviation θstd of all reflected infrared signals within each cluster;

[0020] If Iavg > Ith and θstd < θth, it is judged that the cluster is the cluster corresponding to the large obstacle; otherwise, it is excluded.

[0021] As a further solution of the present invention, the step of constructing a probability density function using a kernel density estimation method comprises:

[0022] Extract the distance and angle information in the cluster corresponding to the large obstacle, and combine the extended range of each distance and angle into a two-dimensional data area to form an extended data set Dexp;

[0023] Use the Gaussian kernel function to construct a probability density function on the expanded data set Dexp;

[0024] For a two-dimensional data point (d,θ), the kernel density estimate on the data set Dexp is:

[0025]

[0026] Among them, nexp is the number of points in the expanded data set Dexp, and σ is the bandwidth of the kernel function.

[0027] As a further solution of the present invention, the specific operation of extending the distance and angle is:

[0028] Determine a distance expansion value △d. If the error range of distance measurement is ±d, the distance value di in the cluster where the large obstacle is located determined by the infrared sensor is expanded to [di-△d, di+△d], and the expanded distance range is normalized to map it to the [0,1] interval.

[0029] Determine an angle expansion amount △θ, expand it to [θi-Δθ,θi+Δθ], and normalize the expanded angle range to the interval [0,2π];

[0030] The extended range of each distance and angle is combined into a two-dimensional data region to form an extended data set Dexp = {(di k, θi j)}, where k and j represent the subdivision point indexes within the extended range of distance and angle, respectively.

[0031] As a further solution of the present invention, for each point in the lidar point cloud data, its coordinates are converted into polar coordinates, and the distance and angle are expanded and normalized in the same way. The processed point cloud data is substituted into the constructed probability density function f, and the probability value Pj of the point belonging to a large obstacle is calculated. If Pj>=T, the point is judged as point cloud data belonging to a large obstacle, otherwise, it is excluded, where T is the probability threshold.

[0032] As a further solution of the present invention, the steps of determining the minimum convex polygon by Graham scanning algorithm are:

[0033] Select a reference point p0 from the point set, and calculate the polar angles of the remaining points relative to the reference point p0 (if the polar angles are the same, they are sorted from near to far from the reference point);

[0034] Assume that the sorted points are p1, p2, ..., pn-1, and the polar angle is calculated by the formula θ = arctan2(y-y0, x-x0), where n is the total number of points, (x0, y0) is the coordinate of the reference point, (x, y) is the coordinate of the point to be calculated, and the arctan2 function is an improved inverse tangent function;

[0035] Create a stack to store the points on the convex hull. First, push the reference point p0 and the first point p1 after sorting into the stack. Starting from the third point p2, traverse the sorted point set in sequence. For each point pi (i>=2): when there are at least two points in the stack, take out the two points at the top of the stack (set to ptop and ptop-1), and calculate the vector p top-1 p top and vector p top-1 p i The cross product of

[0036] If the cross product is greater than or equal to 0, pop the point at the top of the stack and continue to check the relationship between the two new top points of the stack and the current point until the cross product is less than 0 or there is only one point left in the stack;

[0037] When the cross product is less than 0, push the current point pi into the stack;

[0038] When all the sorted points are traversed, the remaining points in the stack are the points that constitute the convex hull. By connecting these points in order, we get the minimum convex polygon that contains all the given points.

[0039] As a further solution of the present invention, the method for calculating the minimum convex polygon area using the Gaussian area formula is:

[0040] If the minimum convex polygon has n vertices, record the coordinates of these vertices in clockwise or counterclockwise order, set to P0(x0,y0), P1(x1,y1), ..., Pn-1(xn-1,yn-1);

[0041] For i from 0 to n-2, calculate (xi*yi+1)-(xi+1*yi) in sequence, and finally calculate (xn-2*y0-x0*yn-2);

[0042] The area of ​​the minimum convex polygon is calculated according to the formula A1=(1 / 2)*|sum((xi*yi+1)-(xi+1*yi))+(xn-2*y0-x0*yn-2)|.

[0043] The present invention provides an infrared anti-collision control system for a maintenance work unit, which has the following beneficial effects compared with the prior art:

[0044] (1) The present invention uses SNN-DBSCAN to cluster reflected infrared signals by calculating the shared nearest neighbor similarity between data points. For reflected infrared signals received at similar time and spatial positions, they can be clustered into a cluster based on their higher shared nearest neighbor similarity, paving the way for accurately determining the spatial position of large obstacles.

[0045] (2) The present invention mainly uses infrared sensors and supplemented by laser radar to obtain accurate location information of large obstacles. At the same time, the Graham scanning algorithm is used to obtain the minimum convex polygon of the horizontal plane projection of the large obstacle, and the Gaussian area formula is used to calculate the area of ​​the minimum convex polygon, which is compared with the area of ​​the maintenance work unit to preliminarily determine whether the maintenance work unit has a risk of collision. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a system block diagram of the present invention. DETAILED DESCRIPTION

[0047] 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.

[0048] like Figure 1 The present application provides an infrared anti-collision control system for a maintenance work unit, comprising:

[0049] Information collection module: An infrared sensor matrix with adjustable sensing angle is installed at a suitable position in front of the maintenance work unit. These sensors can emit infrared rays at different preset angles, receive reflected infrared signals, and collect multiple groups of reflected infrared signals within a preset time T. For example, the emission angle of the sensor is adjusted at certain angles (such as 10 degrees) from left to right in the horizontal direction, and n different emission angles are set in total, so as to achieve multi-angle coverage of the area in front of the maintenance work unit;

[0050] When the maintenance work unit is performing work, each infrared sensor emits infrared light at its own angle, and reflection will occur when encountering the object in front, and the reflected light will be received by the sensor. If at a certain moment, the intensity of the reflected light received by the i-th sensor is I i, and the time delay of the reflected light is ti (related to the distance, which can be calculated according to the formula d = (c*t) / 2, where c is the speed of infrared light and d is the distance to the large obstacle), by collecting these reflected light intensity and time delay information, a series of original infrared signal data reflecting the distribution of objects in front is obtained.

[0051] The infrared information processing module takes all the collected reflection signal data as input. Before input, the multidimensional vector of each reflection signal data point needs to be normalized by Min-Max to ensure that the data of different dimensions are in a similar numerical range to avoid the clustering effect affected by dimensional differences. For example, each reflection signal data point is represented by a multidimensional vector (I, △t, θ), and all emission signal data points are traversed to count the maximum and minimum values ​​of each dimension data respectively. Normalization is performed according to the formula xnorm = (x-xmin) / (xmax-xmin), where x can represent data of dimensions such as I, △t, θ, I represents the intensity of the reflection signal, △t represents the time delay, and θ represents the corresponding emission angle;

[0052] The normalized multidimensional vector data is used as input, and the SNN-DBSCAN algorithm is used for clustering: for a given set of normalized data points, the Euclidean distance d between each data point is calculated, and the set of nearest neighbor points of each point is found according to the set nearest neighbor search radius r (the appropriate range is determined in the normalized numerical space); for any two data points Pi and Pj, the number of their shared nearest neighbor points is counted, which is used as a measure of the shared nearest neighbor similarity between them; if the shared nearest neighbor similarity between two data points is greater than or equal to minPt, they are considered to be related in density and are more likely to belong to the same cluster. Starting from all data points, different clusters are gradually formed by continuously merging points with sufficient shared nearest neighbor relationships;

[0053] After clustering, multiple clusters are obtained, each cluster represents a set of reflected signals that may come from the same or a group of similar large obstacles. In actual scenarios, reflected infrared signals from the same or similar large obstacles often have certain correlations in terms of intensity, time delay, and emission angle. The SNN-DBSCAN algorithm, based on the clustering method of shared nearest neighbors, combined with normalized data processing, can more effectively capture these intrinsic connections and aggregate the corresponding reflected signal data points into clusters, thereby preparing for further analysis and identification of large obstacles.

[0054] For each cluster, some internal statistics are calculated, such as the average intensity Iavg, average time delay tavg, and angular standard deviation θstd of all reflected signals in the cluster. Since large obstacles usually produce strong reflected signals with similar time delays at multiple adjacent emission angles, the cluster where the large obstacle is located can be determined according to the following rules: select the cluster whose average intensity exceeds a certain threshold Ith (i.e., Iavg>Ith) and whose angular standard deviation is less than a certain angular threshold θth (i.e., θstd<θth) as the cluster where the large obstacle is located.

[0055] The target echo processing module has determined the partial location of the large obstacle through the relevant information of the infrared signal, but has not obtained relevant information about the side and blocked surface of the large obstacle. Therefore, the laser radar is used to collect information on the other sides of the large obstacle to obtain a series of point cloud data. However, due to the accuracy of the infrared sensor and the uncertainty in the actual scene, a distance expansion △d is determined. If the error range of the distance measurement is ±d, the distance value di in the cluster where the large obstacle is located determined by the infrared sensor is expanded to [di-△d, di+△d], and the expanded distance range is normalized to map it to the [0,1] interval; similarly, considering the error or possible deviation of the angle measurement, an angle expansion △θ is determined, which is expanded to [θi-Δθ, θi+Δθ], and the expanded angle range is normalized to the [0,2π] interval; after expansion and normalization, each distance and angle expansion range is combined into a two-dimensional data area to form an expanded data set Dexp={(dik,θi j)}, where k and j represent the subdivision point indexes in the distance and angle expansion ranges, respectively;

[0056] Use the kernel density estimation method to construct a probability density function on the expanded data set Dexp, using the Gaussian kernel function For a two-dimensional data point (d,θ), the kernel density estimate on the data set Dexp is: Among them, nexp is the number of points in the expanded data set Dexp, and σ is the bandwidth of the kernel function;

[0057] For each point in the lidar point cloud data, its coordinates are converted into polar coordinates (dj, θj), and the distance and angle are expanded and normalized in the same way as when the probability density function is constructed. The processed point cloud data is substituted into the constructed probability density function f, and the probability value Pj = f(dj, θj) that the point belongs to a large obstacle is calculated;

[0058] A probability threshold T is set. If the probability value Pj of a point cloud data is greater than or equal to T, the point is determined to be point cloud data belonging to a large obstacle. Otherwise, it is excluded.

[0059] The information integration module obtains the coordinate set (dv, θv) of large obstacles from infrared sensors and lidar, where v covers all relevant boundary point numbers. This set includes the boundary conditions of large obstacles that can be detected from all angles.

[0060] For each integrated boundary point (dv, θv), use the formula to convert it into rectangular coordinates (xk, yk). The conversion formula is: xk = dkcosθk, yk = dksinθk. Through this conversion, all boundary points of large obstacles can be expressed in a rectangular coordinate system with the maintenance work unit as the origin.

[0061] According to the converted rectangular coordinate points, the Graham scanning algorithm is used to determine the minimum convex polygon containing all boundary points: select a reference point from the point set, usually the point with the smallest ordinate (if there are multiple points with the smallest ordinate, the point with the smallest abscissa is selected) as the starting point p0, this point must be a point on the convex hull and can be used as a reference point for subsequent scanning; calculate the polar angles of the remaining points relative to the reference point p0 (if the polar angles are the same, they are sorted from near to far from the reference point), and the sorted points are set to be p1, p2, ..., pn-1, where n is the total number of points in the point set (excluding the reference point). Point p0), calculate the polar angle by formula θ=arctan2(y-y0,x-x0), where (x0,y0) are the coordinates of the reference point, (x,y) are the coordinates of the point to be calculated, arctan2 function is an improved inverse tangent function, it can correctly handle the situation of each quadrant, and the return value range is (-π,π]; create a stack to store the points on the convex hull, first push the reference point p0 and the first point p1 after sorting into the stack, start from the third point p2, traverse the sorted point set in sequence, for each point pi (i>=2): when there are at least two points in the stack, take out the two points at the top of the stack (set to p top and p top-1 ), calculate the vector p top-1 p top and vector p top-1 p iIf the cross product is greater than or equal to 0, it means that the current point pi is on the right side of the straight line formed by the two points at the top of the current stack (considered in a counterclockwise direction), which means that the point at the top of the current stack is not a point on the convex hull. Pop the point at the top of the stack and continue to check the relationship between the two new points at the top of the stack and the current point until the cross product is less than 0 or there is only one point left in the stack. When the cross product is less than 0, it means that the current point pi makes the point at the top of the current stack form a part of the convex hull, and the current point pi is pushed into the stack; after traversing all the sorted points, the remaining points in the stack are the points that constitute the convex hull. Connect these points in order to get the minimum convex polygon containing all given points;

[0062] After obtaining the coordinates of each vertex of the minimum convex polygon through the Graham scanning algorithm, the area of ​​the minimum convex polygon is calculated using the Gaussian area formula:

[0063] Assume that the minimum convex polygon has n vertices, and record the coordinates of these vertices in clockwise or counterclockwise order, set as P0(x0,y0), P1(x1,y1), ..., Pn-1(xn-1,yn-1);

[0064] For i from 0 to n-2, calculate the value of (xi*yi+1)-(xi+1*yi) in turn, and finally calculate the value of (xn-2*y0-x0*yn-2);

[0065] Add the previously calculated terms (xi*yi+1)-(xi+1*yi)(i∈[0,n-2]) and (xn-2y0-x0yn-2) to get a sum;

[0066] Finally, multiply this sum by 1 / 2 and take its absolute value. The result is the area of ​​the convex polygon, i.e. A1 = (1 / 2)*|sum((xi*yi+1)-(xi+1*yi))+(xn-2*y0-x0*yn-2)|;

[0067] The collision judgment module measures the road width as W, the longest horizontal span of the convex polygon projected by the large obstacle as L, the width of the maintenance work unit as Wyang and the length as Lyang, and the minimum single-side space margin required for safe passage as S;

[0068] First, from the perspective of area, if A1>=A2 (where A2 is the area occupied by the maintenance work unit), this is the initial condition for preliminarily judging whether the maintenance work unit can pass through the large obstacle. When we calculate that the projection area of ​​the large obstacle on the ground is larger than the minimum safe passing area of ​​the maintenance work unit, from the perspective of plane space occupation, this means that the space range occupied by the large obstacle in the horizontal direction is larger than the space range that the maintenance work unit can safely pass through;

[0069] In the actual maintenance environment, safe passage does not only mean that the main part of the maintenance work unit can pass, but also needs to consider the auxiliary equipment and working parts of the maintenance work unit and ensure sufficient safety distance. Taking the underground pipeline vehicle as an example, the vehicle itself has a certain body width, and it may also be equipped with crane arms, maintenance tools and other equipment. These equipment also need to occupy a certain space during the vehicle's driving. Since the projection area of ​​large obstacles is too large, even if the main body of the vehicle barely passes, the auxiliary equipment may collide with the large obstacle and cause equipment damage. Therefore, maintaining a certain safety distance is to avoid collisions due to minor errors during the passing process. Therefore, when the projection area of ​​a large obstacle is larger than the minimum safe passing area, it is determined that it cannot be passed safely based on comprehensive safety considerations;

[0070] Then, consider the road width factor. When the road width is sufficient, it is necessary not only to ensure that the maintenance work unit can pass through in the width direction, but also to consider the operating space when bypassing large obstacles. If the following two conditions can be met, it is judged that the maintenance work unit can safely pass through large obstacles:

[0071] Condition 1: W>=Wyang+2S+L. This condition ensures that after removing the length occupied by large obstacles in the road width direction, the remaining space is sufficient for the maintenance work unit to pass through and maintain a safety margin;

[0072] Condition 2: W>=Lyang+2S+L. This condition takes into account that when the maintenance work unit passes a large obstacle along the road, whether it is driving in a straight line or preparing to bypass a large obstacle, the width of the road can accommodate the length of the maintenance work unit and the length of the large obstacle and the safety margin on both sides;

[0073] If any of the above three conditions is not met, the maintenance crew alarm mechanism will be triggered to remind the maintenance crew management personnel that a collision will occur if they move forward.

[0074] Some of the data in the above formulas are numerically calculated by removing their dimensions. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0075] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An infrared anti-collision control system for a maintenance work unit, characterized in that: It includes the following steps: An information collection module installs an infrared sensor matrix and a lidar with adjustable induction angle at the front of the maintenance work unit, and emits infrared rays and electromagnetic waves at preset different angles to large obstacles within time T, and collects multiple groups of reflected infrared signals and target echoes; An infrared signal processing module clusters the obtained reflected infrared signals by using the SNN-DBSCAN density clustering algorithm, and obtains the clusters corresponding to large obstacles according to the average intensity and the angle standard deviation; A target echo processing module uses the kernel density estimation method to construct a probability density function according to the clusters corresponding to large obstacles, substitutes the point cloud data into the probability density function, and judges whether it belongs to a large obstacle; An information integration module projects the large obstacle onto the horizontal plane, converts its boundary points into rectangular coordinates, determines the smallest convex polygon containing all boundary points through the Graham scan algorithm, and calculates the area A1 of the smallest convex polygon by using the Gaussian area formula; A collision judgment module, if A1 < A2, then preliminarily judges that the maintenance work unit will not collide. Substitute the road width W. If W >= max{Lyang, Wyang}+2S+L, then judge that the maintenance work unit can safely pass the large obstacle, otherwise, judge that the maintenance work unit cannot safely pass the large obstacle and trigger a collision warning, where A2 represents the area of the maintenance work unit, L is the longest horizontal span of the smallest convex polygon, and Wyang and Lyang are the length and width of the maintenance work unit.

2. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: The specific steps of clustering the reflected infrared signals by using the SNN-DBSCAN density clustering algorithm: Perform Min-Max normalization processing on the reflected infrared signal data, and calculate the Euclidean distance d between each data point; According to the set nearest neighbor search radius r, find the set of nearest neighbor points of each point; For any two data points Pi and Pj, count the number num of their shared nearest neighbor points; If the shared nearest neighbor similarity num between two data points >= minPt, then it is considered that they are density-related and judge that they belong to the same cluster; For all data points, gradually form different clusters by continuously merging points with sufficient shared nearest neighbor relationships.

3. The infrared anti-collision control system for a maintenance work unit according to claim 2, characterized in that: The specific rules for judging the cluster where the large obstacle is located are: Calculate the average intensity Iavg, the average time delay tavg, and the angle standard deviation θstd of all reflected infrared signals within each cluster; If Iavg > Ith and θstd < θth, then judge that this cluster is the cluster corresponding to the large obstacle, otherwise, exclude it.

4. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: The steps of constructing a probability density function by using the kernel density estimation method include: Extract the distance and angle information in the cluster corresponding to the large obstacle, and combine the extended ranges of each distance and angle into a two-dimensional data area to form an extended data set Dexp; Use the Gaussian kernel function to construct a probability density function on the extended data set Dexp; For the two-dimensional data point (d, θ), the kernel density estimation on the data set Dexp is: where nexp is the number of points in the extended data set Dexp, and σ is the bandwidth of the kernel function.

5. The infrared anti-collision control system for a maintenance work unit according to claim 4, characterized in that: The specific operations for extending distance and angle are: Determine a distance expansion value △d. If the error range of distance measurement is ±d, the distance value di in the cluster where the large obstacle is located determined by the infrared sensor is expanded to [di-△d, di+△d], and the expanded distance range is normalized to map it to the [0,1] interval. Determine an angle expansion amount △θ, expand it to [θi-Δθ,θi+Δθ], and normalize the expanded angle range to the interval [0,2π]; The extended range of each distance and angle is combined into a two-dimensional data region to form an extended data set Dexp = {(dik, θij)}, where k and j represent the subdivision point indexes within the extended range of distance and angle, respectively.

6. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: For each point in the lidar point cloud data, its coordinates are converted into polar coordinates, and the distance and angle are expanded and normalized in the same way. The processed point cloud data is substituted into the constructed probability density function f, and the probability value Pj of the point belonging to a large obstacle is calculated. If Pj>=T, the point is judged as point cloud data belonging to a large obstacle, otherwise, it is excluded, where T is the probability threshold.

7. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: The steps to determine the minimum convex polygon using the Graham scan algorithm are: Select a reference point p0 from the point set, and calculate the polar angles of the remaining points relative to the reference point p0 (if the polar angles are the same, they are sorted from near to far from the reference point); Assume that the sorted points are p1, p2, ..., pn-1, and the polar angle is calculated by the formula θ = arctan2(y-y0, x-x0), where n is the total number of points, (x0, y0) is the coordinate of the reference point, (x, y) is the coordinate of the point to be calculated, and the arctan2 function is an improved inverse tangent function; Create a stack to store the points on the convex hull. First, push the reference point p0 and the first point p1 after sorting into the stack. Starting from the third point p2, traverse the sorted point set in sequence. For each point pi (i>=2): When there are at least two points in the stack, take out the two points at the top of the stack (set to ptop and ptop-1), and calculate the vector and vector The cross product of If the cross product is greater than or equal to 0, pop the point at the top of the stack and continue to check the relationship between the two new top points of the stack and the current point until the cross product is less than 0 or there is only one point left in the stack; When the cross product is less than 0, push the current point pi into the stack; When all the sorted points are traversed, the remaining points in the stack are the points that constitute the convex hull. By connecting these points in order, we get the minimum convex polygon that contains all the given points.

8. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: The method for calculating the minimum convex polygon area using the Gaussian area formula is: If the minimum convex polygon has n vertices, record the coordinates of these vertices in clockwise or counterclockwise order, and set them as P0(x0,y0), P1(x1,y1), ..., Pn-1(xn-1,yn-1); For i from 0 to n-2, calculate (xi*yi+1)-(xi+1*yi) in sequence, and finally calculate (xn-2*y0-x0*yn-2); The area of ​​the minimum convex polygon is calculated according to the formula A1=(1 / 2)*|sum((xi*yi+1)-(xi+1*yi))+(xn-2*y0-x0*yn-2)|.

Images