Front vehicle positioning method and device for intelligent freight formation

Abstract

The invention provides a front vehicle positioning method and device for an intelligent freight formation. The method comprises the following steps: respectively acquiring video images and data point clouds output by a camera and a laser radar mounted on a vehicle; inputting the video image into a 2D target detection model to obtain a target category and target detection frame data; determining a target detection frame of the front vehicle according to the target category, and screening the data point clouds based on target detection frame data; on the basis of the geometrical shape of the screened data point cloud, pose information of the front vehicle relative to the vehicle is calculated, and the pose information comprises vehicle center coordinates and the vehicle head direction. The image 2D target detection model is adopted to obtain the approximate position of the front vehicle, the data point cloud obtained by the laser radar is screened based on the target detection frame data, accurate positioning of the front vehicle can be achieved, and the method has the advantages of being high in calculation speed, high in positioning precision and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle positioning, and in particular relates to a method and device for positioning a leading vehicle in an intelligent freight formation. Background Art

[0002] In the field of intelligent truck transport platooning, in order to achieve stable driving of platoon vehicles in a set formation, it is usually necessary to obtain kinematic information such as the position, posture, and speed of other vehicles in the front platoon. Other kinematic information such as speed can usually be deduced from the posture information of the adjacent time periods. Therefore, it is very important to quickly and accurately obtain posture information (including position and driving direction or vehicle head direction).

[0003] Currently, there are two common approaches to acquiring position information: communication and sensor solutions. The communication approach uses navigation satellites to acquire information such as the vehicle's position and speed. This information is then shared through a network of vehicles within the platoon, transmitting this information to other vehicles. This approach presents two problems: First, truck platooning is primarily used for long-distance highway transport, which inevitably involves passing through long tunnels where satellite signals are blocked. This results in inadequate position information updates for the vehicles within the platoon, seriously impacting platoon safety. Second, in inclement weather such as thunderstorms, rain, and snow, not only is satellite communication quality susceptible to degradation, but inter-vehicle communication is also prone to instability and delays, making it difficult to accurately acquire information about other vehicles and increasing the risk of collisions. Sensor solutions often complement communication solutions, serving as an alternative when communication conditions are poor or nonexistent. Sensor solutions utilize lidar to acquire point cloud data. Deep learning-based 3D object detection algorithms (such as CenterPoint and PointPillars) generate a predicted bounding box of the preceding vehicle. Finally, the preceding vehicle's state information is derived based on the bounding box and adjacent temporal information. This solution has the following problems: First, the accuracy of pose estimation is not ideal. The direction of the predicted rectangular box given by the neural network often deviates from the actual direction of the vehicle, which is even visible to the naked eye. Although it meets general obstacle avoidance needs, it cannot meet high-precision formation maintenance needs; second, compared with the more mature target detection neural network based on 2D images, the target detection neural network based on 3D point cloud information has a higher false detection rate and missed detection rate, which is usually unacceptable for the planning and control algorithm of high-speed moving trucks; third, the target detection network of 3D point cloud usually produces higher computing delays due to the amount of data and the complexity of data representation, and the demand for hardware computing power will also be greater, which not only reduces the real-time nature of status information, but also increases the cost of computing hardware.

[0004] In summary, in order to achieve stable and accurate calculation of the posture information of the leading vehicle in a vehicle formation under harsh environments such as communication denial, a leading vehicle posture calculation method that does not rely on communication and can maintain stable detection and high-precision positioning is needed. Summary of the Invention

[0005] In order to solve the above-mentioned problems existing in the prior art, the present invention provides a method and device for locating the leading vehicle of an intelligent freight formation. The method adopts a sensor solution to obtain the approximate position of the leading vehicle through image 2D target detection, and accurately obtains the posture information of the leading vehicle by using the oblique line positioning corner point method, thereby solving the problems of insufficient precision and large computational complexity of radar point cloud 3D target detection.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions.

[0007] In a first aspect, the present invention provides a method for locating a leading vehicle in an intelligent freight platoon, comprising the following steps:

[0008] Obtain the video images and data point clouds output by the camera and lidar installed on the vehicle respectively;

[0009] Input the video image into a 2D object detection model to obtain object category and object detection frame data;

[0010] Determine a target detection frame of the preceding vehicle according to the target category, and filter the data point cloud based on the target detection frame data;

[0011] Based on the geometric shape of the filtered data point cloud, the position information of the preceding vehicle relative to the host vehicle is calculated. The position information includes the vehicle center coordinates and the vehicle head direction.

[0012] Furthermore, the target categories include the leading vehicle in the platoon, other vehicles, pedestrians, traffic signs and traffic lights.

[0013] Furthermore, the method for screening the data point cloud includes:

[0014] Convert the coordinates (X, Y, Z) of the data point cloud in the laser radar coordinate system to the coordinates (x cam ,y cam ,z cam ), the conversion formula is:

[0015] P cam =M ex -1 P (1)

[0016] Where M ex is the camera's external parameter matrix, P = (X, Y, Z, 1) T , P cam =(x cam ,ycam ,z cam ,1) T ;

[0017] Convert the coordinates in the camera coordinate system to the coordinates in the image coordinate system (u, v). The conversion formula is:

[0018] P'=M in -1 P cam (2)

[0019] Where P' = (u, v, 1) T , M in is the intrinsic parameter matrix of the camera;

[0020] If u and v satisfy inequalities (3) and (4), then the point is retained; otherwise, the point is deleted. Inequalities (3) and (4) are:

[0021] u center -0.5w≤u≤u center +0.5w (3)

[0022] v center -0.5h≤v≤v center +0.5h (4)

[0023] Where, (u center ,v center ) is the coordinate of the center of the target detection frame in the image coordinate system, w and h are the width and height of the target detection frame.

[0024] Furthermore, the method for calculating the position information of the preceding vehicle relative to the vehicle includes:

[0025] The data point cloud is further screened based on the clustering operation and fitted into a rectangular data point cloud;

[0026] Based on the rectangular vertex features, the coordinates of the left and right corner points of the point cloud of the rear of the front vehicle are detected and calculated;

[0027] Based on the coordinates of the left corner point and the right corner point and the geometric dimensions of the preceding vehicle, the position information of the preceding vehicle relative to the present vehicle is calculated.

[0028] Furthermore, methods for further screening the data point cloud based on clustering operations include:

[0029] The clustering algorithm is used to perform clustering operations on the filtered data point cloud to obtain n data clusters. The distance between any two data points in each data cluster is less than or equal to the set distance threshold d. thresh , the distance between any two data points in different data clusters is greater than d thresh ;

[0030] Calculate the distance L between the geometric center of each data cluster and the location of the lidar i , and calculate the eigenvalue T of each data cluster i , the formula is:

[0031]

[0032] Where, T i is the eigenvalue of the i-th data cluster, N i is the number of data in the i-th data cluster, i = 1, 2, ..., n, a and b are artificially set normal numbers;

[0033] A data cluster with the largest eigenvalue is retained as the data point cloud after further screening.

[0034] Furthermore, the method for calculating the position information of the preceding vehicle relative to the host vehicle further includes:

[0035] Take the center of the vehicle as the origin o, the direction of the vehicle's front as the x-axis, and rotate 90 degrees counterclockwise about the x-axis. 0 The direction is the y-axis, and the rectangular coordinate system oxy of the vehicle body is established.

[0036] Furthermore, the method for calculating the coordinates of the left corner point and the right corner point relative to the vehicle includes:

[0037] The maximum deviation angle γ between the front vehicle's head direction at two adjacent moments, i.e., the shooting time of two adjacent frames of images, is determined according to the following inequality:

[0038]

[0039] Where, ω vmax is the maximum angular velocity of the vehicle turning, t dmax is the maximum interval between two adjacent moments;

[0040] Draw two straight lines l1 and l2 through the center of the preceding vehicle at the previous moment. The angles between l1 and l2 and the y-axis are α1 and α2 respectively. The equations of the lines l1 and l2 are:

[0041]

[0042] Where C1 and C2 are the signed straight-line distances from the straight line to the origin of the coordinate system, and the sign represents the position of the straight line relative to the normal direction. are the intercepts of the lines l1 and l2 on the x-axis, and θ is the angle between the front vehicle's head direction and the x-axis at the previous moment;

[0043] Substitute the coordinates (x, y) of each point in the data point cloud into the linear equation of l1 to obtain C1, calculate the (x, y) corresponding to the minimum value of C1, and obtain the coordinates (x, y) of the left corner of the point cloud of the rear of the front vehicle. l ,y l );

[0044] Substitute the coordinates (x, y) of each point in the data point cloud into the linear equation of l2 to obtain C2, calculate the (x, y) corresponding to the maximum value of C2, and obtain the coordinates (x, y) of the right corner point of the rear point cloud of the front vehicle. r ,y r ).

[0045] Furthermore, the angle θ between the front vehicle's head direction and the x-axis at the current moment is t for:

[0046]

[0047] In the formula, (x l ,y l )、(x r ,y r ) are the coordinates of the left and right corner points of the point cloud of the rear of the front vehicle.

[0048] Furthermore, the coordinates of the center of the preceding vehicle relative to the center of the current vehicle are:

[0049]

[0050] Where L is the vehicle body length.

[0051] In a second aspect, the present invention provides a leading vehicle positioning device for an intelligent freight platoon, comprising:

[0052] A data acquisition module is used to acquire video images and data point clouds output by the camera and lidar installed on the vehicle respectively;

[0053] An object detection module is used to input the video image into a 2D object detection model to obtain object category and object detection frame data;

[0054] A data screening module is used to determine the target detection frame of the preceding vehicle according to the target category, and to screen the data point cloud based on the target detection frame data;

[0055] The preceding vehicle positioning module is used to calculate the position information of the preceding vehicle relative to the present vehicle based on the geometric shape of the filtered data point cloud. The position information includes the vehicle center coordinates and the vehicle head direction.

[0056] Compared with the prior art, the present invention has the following beneficial effects.

[0057] The present invention obtains video images and data point clouds output by the camera and lidar installed on the vehicle, respectively, inputs the video images into a 2D target detection model, obtains target categories and target detection frame data, determines the target detection frame of the preceding vehicle based on the target category, and filters the data point cloud based on the target detection frame data. Based on the geometric shape of the filtered data point cloud, the position information of the preceding vehicle relative to the present vehicle is calculated. The position information includes the vehicle center coordinates and the direction of the vehicle head, thereby realizing the positioning of the preceding vehicle in an intelligent freight formation. The present invention obtains the approximate position of the preceding vehicle by using an image 2D target detection model (instead of radar point cloud 3D detection) and filters the data point cloud obtained by the lidar based on the target detection frame data, which can achieve accurate positioning of the preceding vehicle, with the advantages of fast calculation speed and high positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of a method for locating a leading vehicle in an intelligent freight platoon according to an embodiment of the present invention.

[0059] Figure 2 Schematic diagram of camera plane projection.

[0060] Figure 3 Schematic diagram of the leading vehicle positioning principle.

[0061] Figure 4 This is a block diagram of the composition of a leading vehicle positioning device for an intelligent freight formation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions, and advantages of the present invention more clear and understandable, the present invention is further described below with reference to the accompanying drawings and specific embodiments. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. 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.

[0063] Figure 1 This is a flow chart of a method for locating a leading vehicle in an intelligent freight platoon according to an embodiment of the present invention, comprising the following steps:

[0064] Step 101, respectively obtaining video images and data point clouds output by a camera and a lidar installed on the vehicle;

[0065] Step 102: Input the video image into a 2D object detection model to obtain object category and object detection frame data;

[0066] Step 103: determining a target detection frame of the preceding vehicle according to the target category, and filtering the data point cloud based on the target detection frame data;

[0067] Step 104 , based on the geometric shape of the filtered data point cloud, calculate the position information of the preceding vehicle relative to the present vehicle, the position information including the center coordinates of the vehicle and the direction of the vehicle head.

[0068] In this embodiment, step 101 is mainly used to obtain data on the vehicle in front of the vehicle. As mentioned above, commonly used posture information acquisition solutions include communication solutions and sensor solutions. This embodiment adopts a sensor solution, and two sensors, a camera and a lidar, are installed on each vehicle. The installation position can be in front of the vehicle, in front of the left, in front of the right, etc., where the vehicle in front can be observed when it goes straight, turns left, or turns right. The camera is used to capture video images of the target in front of the vehicle (vehicles, traffic signs, etc.); the lidar generally adopts a multi-line lidar, which emits a laser signal and receives the laser echo signal reflected by the target in front to obtain the point cloud data of the target in front (the echo signal of a target forms a point cloud data composed of many discrete points on the display). The multi-line lidar is convenient for obtaining distance and direction data at the same time, and the three-dimensional coordinates of each data point in the radar coordinate system are obtained through transformation.

[0069] In this embodiment, step 102 is mainly used to perform target detection based on video images. In this embodiment, target detection is achieved by inputting the video image captured by the camera into a trained 2D neural network target detection model. The target detection algorithm can adopt mainstream 2D ​​target detection algorithms such as YOLO, FCOS, and RT-DETR. The target detection model outputs the target detection result including the target category and the target detection frame (the minimum circumscribed rectangle of the target image) data. The target detection frame data includes the center coordinates of the detection frame and the detection frame size (length and width). In this embodiment, the approximate position of the preceding vehicle is obtained by using image 2D target detection instead of radar point cloud 3D detection, which can significantly reduce the amount of calculation and help improve the data processing speed.

[0070] In this embodiment, step 103 is primarily used to filter the data point cloud obtained from the LiDAR. To reduce data processing overhead, this embodiment performs a first filter on the data point cloud based on the target detection frame of the preceding vehicle (later embodiments will also perform a second filter using clustering). First, the target detection frame of the preceding vehicle is determined based on the target category. Then, the data point cloud is filtered based on the center position, length, and width of the preceding vehicle's target detection frame. This means that points outside the target detection frame are deleted, retaining only those within the target detection frame.

[0071] In this embodiment, step 104 is primarily used to calculate the positional information of the preceding vehicle relative to the host vehicle. This embodiment calculates the positional information of the preceding vehicle relative to the host vehicle based on the geometric shape of the preceding vehicle's point cloud data distribution, namely, the coordinates of the preceding vehicle's center offset from the host vehicle's center, and the angle at which the preceding vehicle's frontal direction deviates from the host vehicle's frontal direction. When a platoon of trucks is traveling, the rear end of the preceding vehicle faces the host vehicle, so the point cloud shape of the truck can be fitted into a rectangular shape. Since the preceding vehicle's point cloud data is all relative to the host vehicle, combined with the geometric shape characteristics of the preceding vehicle's point cloud (rectangular), the positional information of the preceding vehicle relative to the host vehicle can be calculated.

[0072] This embodiment uses an image 2D target detection model instead of radar point cloud 3D detection to obtain the approximate position of the preceding vehicle, and locates the preceding vehicle in combination with the point cloud data of the preceding vehicle obtained by the lidar, which can improve the calculation speed without affecting the positioning accuracy; by filtering the data point cloud based on the target detection frame data, the amount of calculation can be further reduced, thereby achieving fast real-time positioning of the preceding vehicle.

[0073] As an optional embodiment, the target categories include the leading vehicle in the platoon, other vehicles, pedestrians, traffic signs, and traffic lights.

[0074] This example shows the target categories output by the 2D object detection model. Since this example is applied to intelligent truck platooning, the primary target category is the vehicle at the front of the platoon. Other target categories include other vehicles on the road, traffic signs, pedestrians on the sidewalk, and traffic lights at intersections.

[0075] As an optional embodiment, the method for screening the data point cloud includes:

[0076] Convert the coordinates (X, Y, Z) of the data point cloud in the laser radar coordinate system to the coordinates (x cam ,y cam ,z cam ), the conversion formula is:

[0077] P cam =M ex -1 P (1)

[0078] Where M ex is the camera's external parameter matrix, P = (X, Y, Z, 1) T , P cam =(x cam ,y cam ,z cam ,1) T ;

[0079] Convert the coordinates in the camera coordinate system to the coordinates in the image coordinate system (u, v). The conversion formula is:

[0080] P'=M in -1 P cam (2)

[0081] Where P' = (u, v, 1) T , M in is the intrinsic parameter matrix of the camera;

[0082] If u and v satisfy inequalities (3) and (4), then the point is retained; otherwise, the point is deleted. Inequalities (3) and (4) are:

[0083] u center -0.5w≤u≤u center +0.5w (3)

[0084] v center -0.5h≤v≤v center +0.5h (4)

[0085] Where, (u center ,v center ) is the coordinate of the center of the target detection frame in the image coordinate system, w and h are the width and height of the target detection frame.

[0086] This embodiment provides a technical solution for the first screening of the data point cloud. As mentioned above, the first screening is performed based on the target detection frame data of the preceding vehicle, that is, the data point cloud outside the target detection frame is deleted, and the data point cloud inside the target detection frame is retained. Since the target detection frame and the data point cloud belong to different coordinate systems, coordinate conversion is required. First, the coordinates (X, Y, Z) of the data point cloud in the laser radar coordinate system are converted into the coordinates (x cam ,y cam ,z cam ), the conversion formula is as follows (1). Then the coordinates in the camera coordinate system are converted to the coordinates (u, v) in the image coordinate system, and the conversion formula is as follows (2). The conversion matrix involved in the coordinate conversion is the internal and external parameter matrix M in 、M ex , are all pre-calibrated and are mature existing technologies. Specific expressions are not given here. Finally, based on whether the coordinates (u, v) satisfy inequalities (3) and (4), we determine whether the corresponding data point is within the target detection box, and then filter the data point cloud.

[0087] As an optional embodiment, the method for calculating the position information of the preceding vehicle relative to the host vehicle includes:

[0088] The data point cloud is further screened based on the clustering operation and fitted into a rectangular data point cloud;

[0089] Based on the rectangular vertex features, the coordinates of the left and right corner points of the point cloud of the rear of the front vehicle are detected and calculated;

[0090] Based on the coordinates of the left corner point and the right corner point and the geometric dimensions of the preceding vehicle, the position information of the preceding vehicle relative to the present vehicle is calculated.

[0091] This embodiment provides a technical solution for calculating the position information of the preceding vehicle relative to the host vehicle. First, the data point cloud is further filtered based on a clustering operation and then fitted into a rectangle. This process has two advantages: first, it further reduces the computational effort; second, it further removes interfering data points, facilitating the fitting of the data point cloud into a standardized geometric shape, such as a rectangle. Vehicle positioning based on standardized geometric shapes is facilitated. The four vertices are the most critical characteristic points of a rectangular shape. In fact, knowing the positions of two vertices at the ends of a diagonal line uniquely determines the position and size of the rectangle. While two vertices on the same side of a rectangle cannot uniquely define a rectangle, they can still be combined with the rectangle's length or width. Based on the vertex features of the rectangle, this embodiment identifies and calculates the coordinates of the left and right corners of the preceding vehicle's rear point cloud. The left and right corners of the preceding vehicle's rear point cloud are selected because the preceding vehicle's rear is close to the host vehicle's sensor, minimizing obstruction, resulting in high data quality and recognition accuracy. Finally, based on the coordinates of the left and right corner points and the vehicle dimensions (length and width), the position information of the preceding vehicle relative to the vehicle is calculated.

[0092] As an optional embodiment, a method for further screening the data point cloud based on the clustering operation includes:

[0093] The clustering algorithm is used to perform clustering operations on the filtered data point cloud to obtain n data clusters. The distance between any two data points in each data cluster is less than or equal to the set distance threshold d. thresh , the distance between any two data points in different data clusters is greater than d thresh ;

[0094] Calculate the distance L between the geometric center of each data cluster and the location of the lidar i , and calculate the eigenvalue T of each data cluster i , the formula is:

[0095]

[0096] Where, T i is the eigenvalue of the i-th data cluster, N i is the number of data in the i-th data cluster, i = 1, 2, ..., n, a and b are artificially set normal numbers;

[0097] A data cluster with the largest eigenvalue is retained as the data point cloud after further screening.

[0098] This embodiment provides a technical solution for performing a second screening of the data point cloud. The second screening is implemented based on a clustering operation, that is, the data points with a closer distance are divided into a data cluster, and then a data cluster with a closer distance to the laser radar and a larger number of data points is selected as the data point cloud after the second screening. In this way, the data point cloud that is most likely to be the preceding vehicle can be selected, and the interference data or noise data points that do not belong to the preceding vehicle can be deleted. The clustering operation can be implemented by limiting the distance between any two data points in each data cluster to be less than or equal to a set distance threshold, and the distance between any two data points in different data clusters to be greater than a set distance threshold. By calculating the eigenvalue of each data cluster according to formula (5), the data cluster with the largest eigenvalue is retained to achieve data cluster screening. The eigenvalue can be regarded as the distance L between the center of the data cluster and the laser radar. i and the number of data clusters N i According to formula (5), L i The smaller, N i The larger the value, the higher the comprehensive score. In formula (5), a and b are constants greater than 0 and can be set manually based on experience. For example, a = 10 and b = 2 can be set.

[0099] As an optional embodiment, the method for calculating the position information of the preceding vehicle relative to the host vehicle further includes:

[0100] Take the center of the vehicle as the origin o, the direction of the vehicle's front as the x-axis, and rotate 90 degrees counterclockwise about the x-axis. 0 The direction is the y-axis, and the rectangular coordinate system oxy of the vehicle body is established.

[0101] This embodiment provides a plane rectangular coordinate system that needs to be established when calculating the posture information of the preceding vehicle relative to the vehicle. Since the posture information is the posture information of the preceding vehicle relative to the vehicle, the origin of the established coordinate system is the center of the vehicle, the x-axis direction is the direction of the vehicle's front, and the x-axis is rotated 90 degrees counterclockwise. 0 Get the y-axis. The coordinate system oxy is as follows Figure 3 The coordinates involved in the following embodiments are all coordinates in the coordinate system oxy.

[0102] As an optional embodiment, the method for calculating the coordinates of the left corner point and the right corner point relative to the vehicle includes:

[0103] The maximum deviation angle γ between the front vehicle's head direction at two adjacent moments, i.e., the shooting time of two adjacent frames of images, is determined according to the following inequality:

[0104]

[0105] Where, ω vmaxis the maximum angular velocity of the vehicle turning, t dmax is the maximum interval between two adjacent moments;

[0106] Draw two straight lines l1 and l2 through the center of the preceding vehicle at the previous moment. The angles between l1 and l2 and the y-axis are α1 and α2 respectively. The equations of the lines l1 and l2 are:

[0107]

[0108] From the geometric relationship, we can see that, where C1 and C2 are the signed straight-line distances from the straight line to the origin of the coordinate system, and the sign represents the position of the straight line relative to the normal direction. are the intercepts of the lines l1 and l2 on the x-axis, and θ is the angle between the front vehicle's head direction and the x-axis at the previous moment;

[0109] Substitute the coordinates (x, y) of each point in the data point cloud into the linear equation of l1 to obtain C1, calculate the (x, y) corresponding to the minimum value of C1, and obtain the coordinates (x, y) of the left corner of the point cloud of the rear of the front vehicle. l ,y l );

[0110] Substitute the coordinates (x, y) of each point in the data point cloud into the linear equation of l2 to obtain C2, calculate the (x, y) corresponding to the maximum value of C2, and obtain the coordinates (x, y) of the right corner point of the rear point cloud of the front vehicle. r ,y r ).

[0111] This embodiment provides a technical solution for calculating the coordinates of the left corner point and the right corner point. The left corner point and the right corner point are two vertices located on a rectangular point cloud. This embodiment identifies the corner points based on the characteristics of the rectangular vertices. The technical principle of corner point identification is: first draw two straight lines l1 and l2 through the same point (i.e., the center of the preceding vehicle at the previous moment), and by setting the angles α1 and α2 between them and the y-axis, make the data point cloud of the preceding vehicle at the current moment all fall within the fan-shaped area formed by rotating counterclockwise from l1 to l2, as shown in the figure. Figure 3As shown by the two dotted lines in ; then l1 and l2 are moved parallel to the direction close to the data point cloud. Obviously, during the translation process, the intercept C1 of l1 on the x-axis gradually increases and first passes through the left corner point; during the translation process, the intercept C2 of l2 on the x-axis gradually decreases and first passes through the right corner point. Therefore, when l1 passes through the data point cloud, the point corresponding to the minimum value of C1 is the left corner point, and when l2 passes through the data point cloud, the point corresponding to the maximum value of C2 is the right corner point. According to the technical principle, it is easy to obtain the technical solutions listed above. For example, by setting the maximum deflection angle γ, the direction of the straight lines l1 and l2, that is, the angles α1 and α2 with the y-axis, can be determined, thereby obtaining the equations of the straight lines l1 and l2, as shown in formula (7); and by substituting the coordinates (x, y) of each point in the data point cloud into the straight line equation, it is to keep the direction or slope of l1 and l2 unchanged (tanα1, tanα2 unchanged) and perform the translation operation, so that the x-axis intercepts C1 and C2 corresponding to different (x, y) can be obtained. Finally, by solving the (x, y) corresponding to the minimum and maximum values ​​of C1 and C2 respectively, the coordinates of the left and right corner points can be obtained.

[0112] As an optional embodiment, the angle θ between the front vehicle's head direction and the x-axis at the current moment is t for:

[0113]

[0114] In the formula, (x l ,y l )、(x r ,y r ) are the coordinates of the left and right corner points of the point cloud of the rear of the front vehicle.

[0115] This embodiment provides a method for calculating the angle θ between the front vehicle's head direction and the x-axis at the current moment. t This embodiment is based on the obtained left and right corner coordinates (x l ,y l )、(x r ,y r )Calculate θ t Since the direction of the vehicle head is the direction of the symmetry axis of the vehicle body and also the direction of the long side of the rectangular point cloud, according to Figure 3 The geometric relationship between θ and t The calculation formula is as shown in formula (8).

[0116] As an optional embodiment, the coordinates of the center of the preceding vehicle relative to the center of the vehicle at the current moment are:

[0117]

[0118] Where L is the vehicle body length.

[0119] This embodiment provides a technical solution for calculating the center coordinates of the vehicle ahead at the current moment. This embodiment is based on the obtained left and right corner coordinates (x l ,y l )、(x r ,y r ), and the vehicle body length L calculate the center coordinates of the front vehicle. Figure 3 The geometric relationship between the front vehicle and the center coordinates can be obtained as shown in formula (9).

[0120] Figure 4 This is a schematic diagram of the composition of a leading vehicle positioning device for an intelligent freight platoon according to an embodiment of the present invention. The device includes:

[0121] The data acquisition module 11 is used to respectively acquire the video images and data point clouds output by the camera and lidar installed on the vehicle;

[0122] The target detection module 12 is used to input the video image into a 2D target detection model to obtain target category and target detection frame data;

[0123] A data screening module 13 is used to determine the target detection frame of the preceding vehicle according to the target category, and to screen the data point cloud based on the target detection frame data;

[0124] The preceding vehicle positioning module 14 is used to calculate the position information of the preceding vehicle relative to the present vehicle based on the geometric shape of the filtered data point cloud. The position information includes the vehicle center coordinates and the vehicle head direction.

[0125] The device of this embodiment can be used to perform Figure 1 The technical solution of the method embodiment shown has similar implementation principles and technical effects, which will not be repeated here.

[0126] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for positioning the leading vehicle of an intelligent freight platoon, characterized in that: The following steps are involved: Obtain the video images and data point clouds output by the camera and lidar installed on the vehicle respectively; Input the video image into a 2D object detection model to obtain object category and object detection frame data; Determine a target detection frame of the preceding vehicle according to the target category, and filter the data point cloud based on the target detection frame data; Based on the geometric shape of the filtered data point cloud, the position information of the preceding vehicle relative to the host vehicle is calculated. The position information includes the vehicle center coordinates and the vehicle head direction.

2. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 1, characterized in that: The target categories include the leading vehicle in the platoon, other vehicles, pedestrians, traffic signs and traffic lights.

3. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 1, characterized in that: The method for screening the data point cloud includes: Convert the coordinates (X, Y, Z) of the data point cloud in the laser radar coordinate system to the coordinates (x cam ,y cam ,z cam ), the conversion formula is: P cam =M ex -1 P (1) Where M ex is the camera's external parameter matrix, P = (X, Y, Z, 1) T , P cam =(x cam ,y cam ,z cam ,1) T ; Convert the coordinates in the camera coordinate system to the coordinates in the image coordinate system (u, v). The conversion formula is: P'=M in -1 P cam (2) Where P' = (u, v, 1) T , M in is the intrinsic parameter matrix of the camera; If u and v satisfy inequalities (3) and (4), then the point is retained; otherwise, the point is deleted. Inequalities (3) and (4) are: and center -0.5w≤u≤u center +0.5w (3) in center -0.5h≤v≤v center +0.5h (4) Where, (u center ,v center ) is the coordinate of the center of the target detection frame in the image coordinate system, w and h are the width and height of the target detection frame.

4. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 1, characterized in that: The method for calculating the position information of the preceding vehicle relative to the vehicle includes: The data point cloud is further screened based on the clustering operation and fitted into a rectangular data point cloud; Based on the rectangular vertex features, the coordinates of the left and right corner points of the point cloud of the rear of the front vehicle are detected and calculated; Based on the coordinates of the left corner point and the right corner point and the geometric dimensions of the preceding vehicle, the position information of the preceding vehicle relative to the present vehicle is calculated.

5. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 4, characterized in that: Methods for further screening data point clouds based on clustering operations include: The clustering algorithm is used to perform clustering operations on the filtered data point cloud to obtain n data clusters. The distance between any two data points in each data cluster is less than or equal to the set distance threshold d. thresh , the distance between any two data points in different data clusters is greater than d thresh ; Calculate the distance L between the geometric center of each data cluster and the location of the lidar i , and calculate the eigenvalue T of each data cluster i , the formula is: Where, T i is the eigenvalue of the i-th data cluster, N i is the number of data in the i-th data cluster, i = 1, 2, ..., n, a and b are artificially set normal numbers; A data cluster with the largest eigenvalue is retained as the data point cloud after further screening.

6. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 4, characterized in that: The method for calculating the position information of the preceding vehicle relative to the own vehicle further includes: Take the center of the vehicle as the origin o, the direction of the vehicle's front as the x-axis, and rotate 90 degrees counterclockwise about the x-axis. 0 The direction is the y-axis, and the rectangular coordinate system oxy of the vehicle body is established.

7. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 6, characterized in that: The method for calculating the coordinates of the left corner point and the right corner point relative to the vehicle includes: The maximum deviation angle γ between the front vehicle's head direction at two adjacent moments, i.e., the shooting time of two adjacent frames of images, is determined according to the following inequality: Where, ω vmax is the maximum angular velocity of the vehicle turning, t dmax is the maximum interval between two adjacent moments; Draw two straight lines l1 and l2 through the center of the preceding vehicle at the previous moment. The angles between l1 and l2 and the y-axis are α1 and α2 respectively. The equations of the lines l1 and l2 are: Where C1 and C2 are the signed straight-line distances from the line to the origin of the coordinate system, and the sign represents the position of the line relative to the normal direction. are the intercepts of the lines l1 and l2 on the x-axis, and θ is the angle between the front vehicle's head direction and the x-axis at the previous moment; Substitute the coordinates (x, y) of each point in the data point cloud into the linear equation of l1 to obtain C1, calculate the (x, y) corresponding to the minimum value of C1, and obtain the coordinates (x, y) of the left corner point of the rear point cloud of the front vehicle. l ,y l ); Substitute the coordinates (x, y) of each point in the data point cloud into the linear equation of l2 to obtain C2, calculate the (x, y) corresponding to the maximum value of C2, and obtain the coordinates (x, y) of the right corner point of the rear point cloud of the front vehicle. r ,y r ).

8. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 7, characterized in that: The angle θ between the front vehicle's head direction and the x-axis at the current moment t for: In the formula, (x l ,y l )、(x r ,y r ) are the coordinates of the left and right corner points of the point cloud of the rear of the front vehicle.

9. The method for positioning the leading vehicle of an intelligent freight platoon according to claim 8, characterized in that: The coordinates of the center of the preceding vehicle relative to the center of this vehicle at the current moment are: Where L is the vehicle body length.

10. A leading vehicle positioning device for an intelligent freight platoon, characterized in that: include: A data acquisition module is used to acquire video images and data point clouds output by the camera and lidar installed on the vehicle respectively; An object detection module is used to input the video image into a 2D object detection model to obtain object category and object detection frame data; A data screening module is used to determine the target detection frame of the preceding vehicle according to the target category, and to screen the data point cloud based on the target detection frame data; The preceding vehicle positioning module is used to calculate the position information of the preceding vehicle relative to the present vehicle based on the geometric shape of the filtered data point cloud. The position information includes the vehicle center coordinates and the vehicle head direction.

Images