An infrared anti-collision control system for maintenance work units

Abstract

The present invention discloses an infrared anti-collision control system for a maintenance work unit. The present invention relates to the technical field of anti-collision control systems for maintenance work units. The present invention obtains reflected infrared signals through an infrared sensor, classifies them through SNN-DBSCAN, and finds clusters corresponding to large obstacles. To obtain more comprehensive information about large obstacles, a laser radar is used to collect actual information on the other sides of the large obstacle. The large obstacle is projected onto a horizontal plane, and the Graham scanning algorithm is used to describe the minimum convex polygon containing each point on the projected boundary of the large obstacle. The area of ​​the minimum convex polygon is calculated using the Gaussian area formula and compared with the area occupied by the maintenance work unit to preliminarily determine whether there is a collision risk. If not, the road width is introduced to make a more detailed judgment, significantly reducing the problem of maintenance work unit collisions caused by inaccurate judgments.

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 fields like road maintenance and construction, maintenance crews often encounter complex and ever-changing environments, often encountering various obstacles. Large obstacles, in particular, pose a serious threat to the safe operation of maintenance crews. When a large obstacle appears in front of a crew, accurately determining whether it can safely pass is crucial; a misjudgment can easily lead to a collision.

[0003] When traditional infrared sensors are used to illuminate large obstacles, the information contained in the reflected signals collected is limited 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 through.

[0004] Furthermore, existing collision avoidance systems often lack comprehensive consideration of key factors such as the area of ​​large obstacles and the relative position of the maintenance crew and the large obstacle. In actual operations, simply knowing the presence of a large obstacle is insufficient. In-depth analysis is required regarding the area occupied by the obstacle and the maintenance crew, as well as the road width and the ability of the maintenance crew to change lanes. Only then can scientific and rational decisions be made to ensure the safe and efficient operation of the maintenance crew. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, 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 is equipped with an infrared sensor matrix and a laser radar with adjustable sensing angles in front of the maintenance work unit. It emits infrared rays and electromagnetic waves at different preset angles at large obstacles within a time period T, and collects multiple sets of reflected infrared signals and target echoes.

[0008] The infrared signal processing module uses the SNN-DBSCAN density clustering algorithm to cluster the reflected infrared signals and obtain the clusters corresponding to large obstacles based on the average intensity and angle standard deviation;

[0009] The target echo processing module constructs a probability density function using the kernel density estimation method based on 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, then judge that the maintenance work crew can safely pass the large obstacle, otherwise, judge that the maintenance work crew cannot safely pass the large obstacle and trigger a collision warning. 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 judge 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, then judge that the cluster is the cluster corresponding to the large obstacle, otherwise, exclude it.

[0021] As a further embodiment of the present invention, the steps of constructing a probability density function using a kernel density estimation method include:

[0022] Extract the distance and angle information from the clusters corresponding to large obstacles, and combine the extended range of each distance and angle into a two-dimensional data region to form an expanded 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 expanding the distance and angle is:

[0028] Determine a distance expansion value △d. 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]. The expanded distance range is normalized to map it to the interval [0,1].

[0029] Determine an angle expansion value △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, θij)}, 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 similarly expanded and normalized. 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 determined to be 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 the Graham scanning algorithm are as follows:

[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 closest to farthest from the reference point);

[0034] Assume that the sorted points are p1, p2, ..., pn-1, and calculate the polar angle using 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, and (x, y) is the coordinate of the point to be calculated. 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 as 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 top point 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] After traversing all the sorted points, 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 area of ​​the minimum convex polygon 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, and set them as 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 the higher shared nearest neighbor similarity between them, 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. It is compared with the area of ​​the maintenance work unit to preliminarily determine whether the maintenance work unit has the 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 clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts 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] The information collection module installs an infrared sensor matrix with adjustable sensing angles at a suitable location in front of the maintenance unit. These sensors can emit infrared rays at different preset angles and receive reflected infrared signals, collecting multiple sets of reflected infrared signals within a preset time T. For example, the sensor's emission angle can be adjusted horizontally from left to right at a certain angle (e.g., 10 degrees), setting a total of n different emission angles to achieve multi-angle coverage of the area in front of the maintenance unit.

[0050] When the maintenance team is working, each infrared sensor emits infrared light at its own angle. Upon encountering an object ahead, the reflected light is reflected and received by the sensor. If, at a given moment, the intensity of the reflected light received by the i-th sensor is Ii, and the time delay of the reflected light is ti (which is related to distance and can be calculated using the formula d = (c*t) / 2, where c is the speed of infrared light and d is the distance to the large obstacle), then by collecting this reflected light intensity and time delay information, a series of raw infrared signal data reflecting the distribution of objects ahead is obtained.

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

[0052] The normalized multidimensional vector data is used as input and clustering is performed using the SNN-DBSCAN algorithm: 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, and this 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 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 representing a collection of reflected signals that may come from the same or a group of similar large obstacles. This is because in real-world 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 shared nearest neighbor clustering and combined with normalized data processing, can more effectively capture these inherent 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 transmission angles, the cluster where the large obstacle is located can be determined according to the following rule: the cluster with an average intensity exceeding a certain threshold Ith (i.e., Iavg>Ith) and an angular standard deviation less than a certain angular threshold θth (i.e., θstd<θth) is selected as the cluster where the large obstacle is located.

[0055] The target echo processing module has already determined the partial location of the large obstacle through the relevant information of the infrared signal. However, it has not been able to obtain relevant information about the side and obstructing surface of the large obstacle. Therefore, the lidar 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 value △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 interval [0,1]. Similarly, considering the error or possible deviation of the angle measurement, an angle expansion value △θ is determined and expanded to [θi-Δθ,θi+Δθ]. The expanded angle range is normalized to the interval [0,2π]. After expansion and normalization, the expansion range of each distance and angle is combined into a two-dimensional data region to form an expanded data set Dexp={(dik,θij)}, where k and j represent the subdivision point indexes within the distance and angle expansion ranges, respectively.

[0056] Use the kernel density estimation method to construct the 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 to polar coordinates (dj, θj), and the distance and angle are expanded and normalized in the same way as when constructing the probability density function. 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 a 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 based on the infrared sensor and lidar, where v covers the serial numbers of all relevant boundary points. 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 the large obstacle 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 according to the distance from the reference point), and let the sorted points 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 and (x,y) are the coordinates of the point to be calculated. The arctan2 function is an improved inverse tangent function that can correctly handle the situations in each quadrant. 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. 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 on the top of the stack (set as 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 (considering it 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 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 obtain the minimum convex polygon containing all given points.

[0062] After obtaining the coordinates of each vertex of the minimum convex polygon through the Graham scan 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. 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);

[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 team as Wyang and the length as Lyang, and the minimum single-sided 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 crew), this is the initial condition for determining whether the maintenance crew can pass through the large obstacle. When we calculate that the projected area of ​​the large obstacle on the ground is larger than the minimum safe passage area for the maintenance crew, from the perspective of plane space occupation, this means that the space occupied by the large obstacle in the horizontal direction is larger than the space range that the maintenance crew can safely pass through;

[0069] In actual maintenance environments, safe passage does not only mean that the main part of the maintenance work unit can pass through, but also needs to consider the auxiliary equipment, working parts and sufficient safety distance of the maintenance work unit. Taking underground pipeline vehicles as an example, the vehicle itself has a certain body width, and it may also carry equipment such as crane arms and maintenance tools. These equipment also need to occupy a certain amount of space during the vehicle's driving process. Since the projected 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 passage process. Therefore, when the projected area of ​​a large obstacle is larger than the minimum safe passing area, it is determined that it cannot be safely passed based on comprehensive safety considerations;

[0070] Next, consider the road width. When the road is wide enough, not only must the maintenance crew be able to pass through in the width direction, but also the operating space when bypassing large obstacles must be considered. If the following two conditions can be met, the maintenance crew can be judged to be able to safely pass 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 team to pass through and maintain a safety margin;

[0072] Condition 2: W>=Lyang+2S+L. This condition takes into account that when the maintenance team passes a large obstacle along the road, whether driving in a straight line or preparing to go around a large obstacle, the width of the road can accommodate the length of the maintenance team and the length of the large obstacle, as well as 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 manager that a collision will occur if they move forward.

[0074] Some of the data in the above formulas are dimensionless and numerically calculated. 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 and are not intended to limit the present invention. 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 with adjustable induction angle and a lidar in the front of the maintenance work unit, and emits infrared rays and electromagnetic waves at preset different angles towards large obstacles within time T, and collects multiple groups of reflected infrared signals and target echoes; An infrared signal processing module uses the SNN-DBSCAN density clustering algorithm to cluster the obtained reflected infrared signals, and obtains the clusters corresponding to large obstacles according to the average intensity and 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 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. Among them, A2 represents the area of the maintenance work unit, L is the longest horizontal span of the smallest convex polygon, Wyang and Lyang are the width and length of the maintenance work unit, and S is the minimum space margin required for safe passage on one side; 2. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: The specific steps of using the SNN-DBSCAN density clustering algorithm to cluster the reflected infrared signals: 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 for 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, it is considered that they are density-related and judge that they belong to the same cluster. Among them, minPt is the set minimum shared nearest neighbor number threshold; 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, average time delay tavg, and angle standard deviation of all reflected infrared signals in each cluster ; If Iavg>Ith and , then the cluster is judged to be the cluster corresponding to the large obstacle, otherwise, it is excluded, where lth is the average intensity threshold, is the angle threshold.

4. The infrared anti-collision control system for a maintenance work unit according to claim 1, characterized in that: The steps of using the kernel density estimation method to construct a probability density function 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 expanded dataset Dexp, K() is the Gaussian kernel function, and k and j represent the subdivision point indices within the distance and angle expansion ranges, respectively.

5. The infrared anti-collision control system for a maintenance work unit according to claim 4, characterized in that: The specific operation of expanding the distance and angle is: Determine a distance expansion value △d. 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]. The expanded distance range is normalized to map it to the interval [0,1]. Determine an angle expansion △θ and expand it to , 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 distance and angle extended ranges, 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 determined to be the point cloud data belonging to a large obstacle, otherwise, it is excluded. The specific operation of the same expansion and normalization process is as follows: in terms of distance, a distance expansion amount △d is determined. For the distance value di in the cluster where the large obstacle is located determined by the infrared sensor, it is expanded to [di-△d,di+△d]. According to the normalization formula Normalize the expanded distance value di' to [0,1]; in terms of angle, determine an angle expansion amount , which is expanded to , according to the normalization formula The expanded angle value Normalized to , 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, calculate the polar angles of the remaining points relative to the reference point p0, and if the polar angles are the same, sort them from closest to farthest from the reference point; Assume that the sorted points are p1, p2, ..., pn-1, and calculate the polar angle using 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, and (x, y) is the coordinate of the point to be calculated. The arctan2 function is an improved inverse tangent function. Create a stack to store 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, 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 top point 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; After traversing all the sorted points, 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 area of ​​the minimum convex polygon 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