LiDAR 3D target detection method and system based on regional body column feature fusion

Abstract

The present application discloses a method and system for 3D target detection based on regional body column feature fusion laser radar, which relates to the field of laser radar perception technology. The method includes collecting a point cloud data set and its coordinates of the target area, determining the point cloud range and dividing it into multiple sub-areas. The maximum difference of the z coordinate in each sub-area is calculated, and a feature array is created. The maximum difference is assigned to the points in the sub-area to form a body column feature. The number of assignments is dynamically adjusted according to the target area to obtain the target feature. The present application enhances feature encoding through regional attention to improve the network's detection performance for small targets.

Description

Technical Field

[0001] The present application relates to the field of laser radar perception technology, and in particular to a laser radar 3D target detection method and system based on regional body column feature fusion. Background Art

[0002] Pillar-based LiDAR target detection algorithms (such as PointPillars) are a technology that has been widely used in the field of autonomous driving in recent years. The core idea of ​​this algorithm is to convert the 3D point cloud data generated by the LiDAR into a pseudo image so that it can be processed using a 2D convolutional neural network. The Pillar detection algorithm efficiently processes data by converting 3D point clouds into 2D images, reducing computational complexity. It uses a 2D convolutional network for feature extraction and detection, which improves accuracy and speed, and is particularly suitable for fast-response autonomous driving applications. This method is easy to integrate into existing systems and enhance environmental perception capabilities. Research such as PointPillars and VoxelNet has promoted technological progress and has been successful in practice.

[0003] Although the Pillar-based LiDAR target detection algorithm performs well in terms of efficiency and accuracy, it also has some disadvantages and limitations. 1) This method introduces quantization errors in the process of volume columnization and column division. Since the three-dimensional point cloud data is divided into fixed-size volume columns, the spatial resolution of the point cloud may not be sufficient to capture detailed features in some cases, especially for smaller or farther target objects, which may lead to a decrease in detection accuracy. 2) When processing point cloud data, the Pillar-based method simplifies the calculation by converting it into a pseudo-image, but this conversion also brings some information loss. During the conversion process, the spatial relationship and geometric details of the three-dimensional point cloud will be partially lost, affecting the final detection performance. Summary of the invention

[0004] The purpose of this application is to provide a 3D target detection method and system based on regional body column feature fusion LiDAR. By introducing regional attention to the feature encoding of body columns, a richer feature representation is brought about, enabling the network to better understand the details and contextual relationships of the target at different scales. The combination of the two effectively improves the detection performance of small targets.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for detecting 3D targets using a laser radar based on regional body column feature fusion, comprising:

[0007] Obtain the point cloud data set of the target scanning area and the coordinate information of each point cloud data in the point cloud data set.

[0008] According to the coordinate information of each point cloud data, the point cloud range of the target scanning area is determined; the point cloud range is represented by a body column feature encoding method.

[0009] Based on a preset allocation quantity, the point cloud range is discretized into a number of sub-areas; the preset allocation quantity includes the quantity required to be allocated on the x-axis and the quantity required to be allocated on the y-axis; each sub-area includes a number of point cloud data; each sub-area has a unique area identifier.

[0010] For each of the sub-regions, the maximum difference of the z coordinates of the points in the sub-region is calculated according to the coordinate information of each point cloud data in the sub-region.

[0011] Initialize a feature array containing only the z coordinate dimension to store the maximum z coordinate difference in each sub-region; for each sub-region, assign the maximum z difference of the corresponding region in the feature array to each point in the sub-region to obtain the body column feature after adding the regional body column feature.

[0012] According to the body column feature after adding the regional body column feature, based on the area occupied by the target to be measured in the target scanning area, the preset allocation quantity is dynamically adjusted to obtain the target feature of the target to be measured.

[0013] In a second aspect, the present application provides a laser radar 3D target detection system based on regional body column feature fusion, comprising:

[0014] The data acquisition module is used to obtain the point cloud data set of the target scanning area and the coordinate information of each point cloud data in the point cloud data set.

[0015] The range division module is used to determine the point cloud range of the target scanning area according to the coordinate information of each point cloud data; the point cloud range is represented by a body column feature encoding method.

[0016] The area discretization module is used to discretize the point cloud range into a number of sub-areas based on a preset allocation number; the preset allocation number includes the number to be allocated on the x-axis and the number to be allocated on the y-axis; each sub-area includes a number of point cloud data; each sub-area has a unique area identifier.

[0017] The coordinate calculation module is used to calculate the maximum difference of the z coordinates of each point in each sub-region according to the coordinate information of each point cloud data in the sub-region.

[0018] The feature fusion module is used to initialize a feature array containing only the z coordinate dimension, and store the maximum z coordinate difference in each sub-region; for each sub-region, the maximum z difference of the corresponding region in the feature array is assigned to each point in the sub-region to obtain the body column feature after adding the regional body column feature.

[0019] The target detection module is used to obtain the target features of the target to be detected after dynamically adjusting the preset allocation quantity based on the body column features after adding the regional body column features and the area occupied by the target to be detected in the target scanning area.

[0020] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0021] The present application provides a method and system for 3D target detection based on regional body column feature fusion laser radar. The method first obtains a point cloud data set of a target scanning area and extracts the coordinate information of each point cloud data in the data set. Subsequently, the point cloud range of the target scanning area is determined based on the coordinate information of the point cloud data, and the point cloud range is represented by the body column feature encoding technology. Based on a preset allocation number, the point cloud range is divided into a number of sub-areas, wherein the preset allocation number includes the number required to be allocated on the x-axis and the y-axis. Each sub-area contains a number of point cloud data and has a unique area identifier. For each sub-area, the maximum difference of the z coordinates of each point in the sub-area is calculated according to the coordinate information of each point cloud data in the sub-area. Then, a feature array with the same shape as the z coordinate is initialized, and the maximum z difference of the corresponding area in the feature array is assigned to each point in the sub-area, thereby obtaining the body column feature after adding the regional body column feature. According to the body column feature after adding the regional body column feature, and based on the area occupied by the target to be measured in the target scanning area, the preset allocation number is dynamically adjusted to finally obtain the target feature of the target to be measured. This application obtains the point cloud data set of the target scanning area, extracts the coordinate information, determines the point cloud range, and uses the body column feature encoding technology to represent it. After dividing the point cloud range into sub-areas, the maximum difference of the z coordinate of the point cloud data in each sub-area is calculated, and the feature array is initialized, and the maximum z difference is assigned to each point in the sub-area, thereby obtaining the body column feature with the regional body column feature added. The number of allocations is dynamically adjusted according to the area occupied by the target to be measured, and the target features of the target to be measured are finally accurately obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A flowchart of a method for detecting 3D targets using a lidar based on regional body-column feature fusion is provided in accordance with an embodiment of the present application.

[0024] Figure 2The basis provided for an embodiment of the present application and Schematic diagram of dividing the point cloud range.

[0025] Figure 3 A schematic diagram of the internal feature fusion of each region provided in an embodiment of the present application.

[0026] Figure 4 A network diagram provided for an embodiment of the present application.

[0027] Figure 5 This is a schematic diagram of the structure of a laser radar 3D target detection system based on regional body column feature fusion in one embodiment of the present application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0030] Embodiment 1

[0031] like Figure 1 As shown, this embodiment provides a method for detecting 3D targets using a laser radar based on regional body column feature fusion, including:

[0032] Step 101: Obtain a point cloud data set of a target scanning area and coordinate information of each point cloud data in the point cloud data set.

[0033] Step 102: Determine the point cloud range of the target scanning area according to the coordinate information of each point cloud data; the point cloud range is represented by a body column feature encoding method.

[0034] Step 103: Based on a preset allocation quantity, the point cloud range is discretized into a number of sub-areas; the preset allocation quantity includes the quantity to be allocated on the x-axis and the quantity to be allocated on the y-axis; each sub-area includes a number of point cloud data; each sub-area has a unique area identifier.

[0035] Step 104: For each of the sub-regions, the maximum difference of the z coordinates of the points in the sub-region is calculated according to the coordinate information of each point cloud data in the sub-region.

[0036] Step 105: Initialize a feature array containing only the z coordinate dimension to store the maximum z coordinate difference in each sub-region; for each sub-region, assign the maximum z difference of the corresponding region in the feature array to each point in the sub-region to obtain the body column feature after adding the regional body column feature.

[0037] Step 106: According to the body column feature after adding the regional body column feature, based on the area occupied by the target to be measured in the target scanning area, the preset allocation quantity is dynamically adjusted to obtain the target feature of the target to be measured.

[0038] In some embodiments, when executing step 101, the specific steps may be as follows:

[0039] In order to achieve detailed analysis and processing of a specific scan area, it is necessary to first obtain the point cloud dataset of the area. This dataset contains detailed information of all point cloud data in the area, including the specific coordinate position of each point cloud.

[0040] In some embodiments, when executing step 102, the specific steps may be as follows:

[0041] In the body column representation, Indicates that a non-empty column contains Lidar points, where represents a 7-dimensional real vector, They are The xyz coordinates of the points, is the received reflectivity. , , is the offset of each point relative to the center of the body column. is a seven-dimensional vector containing the above information. Contains The x, y, and z coordinates of a point are .

[0042] First, the x and y coordinates of the body column are subtracted from the starting value of the point cloud range to obtain the true coordinates of each body column in the point cloud plane.

[0043] (1).

[0044] (2).

[0045] In the formula, Represents the difference between the body column x coordinate and the point cloud x coordinate, Indicates the difference between the body column y coordinate and the point cloud y coordinate. represents the x-coordinate of each body column, The minimum x-coordinate representing the extent of the point cloud. represents the y coordinate of each body column, The minimum y coordinate representing the extent of the point cloud.

[0046] The width and height of the effective point cloud range in the x and y directions are then calculated using the maximum and minimum values ​​of the x and y coordinates of the body column. The body column is a three-dimensional structure used to represent the three-dimensional position of each point in the point cloud data:

[0047] (3).

[0048] (4).

[0049] (5).

[0050] (6).

[0051] (7).

[0052] (8).

[0053] In the formula, represents the x-coordinate range of the body column, Indicates the y-coordinate range of the body column. It means taking the maximum value in Tensor. Indicates taking the minimum value in Tensor. express The maximum value in express The maximum value of express The minimum value of express The minimum value of .

[0054] Wherein, in some embodiments, Figure 2 As shown, when executing step 103, the specific steps may be as follows:

[0055] Step 301: Based on the number to be allocated on the x-axis and the number to be allocated on the y-axis, the x-coordinate range and the y-coordinate range in the point cloud range are divided to obtain the width and height of each sub-area.

[0056] Step 302: Calculate the x-axis index and y-axis index of the sub-region index according to the difference between the body column x coordinate and the minimum x coordinate of the point cloud range, the difference between the body column y coordinate and the minimum y coordinate of the point cloud range, and the width and height of each sub-region.

[0057] Step 303: Determine the region number of each subregion according to the x-axis index of the subregion index, the y-axis index of the subregion index and the number to be allocated on the y-axis, and obtain each subregion of the point cloud range.

[0058] Specifically, first preset a variable and The range of the x- and y-axis directions is divided to discretize the point cloud range into several sub-areas.

[0059] Then divide the width and height of the point cloud by the preset variables and , determine the width and height of each sub-region.

[0060] (9).

[0061] (10).

[0062] In the formula, Indicates the number of items that need to be allocated on the x-axis. Indicates the amount that needs to be allocated on the y-axis, Indicates the width of each sub-area on the x-axis. Indicates the width of each sub-area on the y-axis. represents the x-coordinate range of the body column, Indicates the y-coordinate range of the body column.

[0063] Then divide both the x and y coordinates by the width and height of the subregion to ensure that the index is in the valid range.

[0064] (11).

[0065] (12).

[0066] In the formula, The x-axis index representing the sub-region index, The y-axis index representing the sub-region index, Indicates that the maximum value of the restriction variable does not exceed The function of Indicates that the maximum value of the restriction variable does not exceed function.

[0067] Then multiply the y coordinate of the sub-area by Add the x-coordinate of the sub-region to generate the sub-region number corresponding to each body column, so that each sub-region has a unique identifier. Finally, extract the z-axis coordinate of the body column and get all the unique sub-region numbers.

[0068] (13).

[0069] In the formula, Indicates a unique sub-region number.

[0070] In some embodiments, when executing steps 104-105, the specific steps may be as follows:

[0071] First, we need to calculate the maximum difference in the z coordinates of the points in each sub-region. This is done by traversing all unique sub-region IDs, and for each sub-region ID, extracting the z coordinates in the sub-region. This is done by traversing all unique sub-region IDs, obtaining the range of each sub-region, saving the coordinates of all points that fall in this sub-region, and then calculating the maximum difference in the z coordinates of all points in the sub-region. Store in a list.

[0072] (14).

[0073] In the formula, Indicates the maximum difference in the z coordinates of the point cloud in the sub-region, Represents the maximum z coordinate of the point cloud in the sub-region, Indicates the minimum z-coordinate of the point cloud within the subregion.

[0074] like Figure 4 As shown, initialize a Tensor with the same shape as the z coordinate, also called a feature array, named max_diffs_per_point. The feature array needs to be assigned to each point in the sub-region. The method is to traverse all unique sub-region IDs. For each sub-region ID, assign the maximum z difference of the corresponding sub-region to each point in the sub-region. Finally, expand max_diffs_per_point by one dimension to get max_diffs, and merge it into the body column feature to get , the schematic diagram of the internal features after fusion is as follows Figure 3 shown.

[0075] (15).

[0076] In the formula, represents the initial body column characteristics, It means to concatenate two tensors in the specified dimension. Represents the body column feature after adding the regional body column feature.

[0077] In some embodiments, when executing step 106, the specific steps may be as follows:

[0078] Since step 103 is used, this application will preset a variable and Divide the range in the x and y axis directions to discretize the point cloud range into several sub-regions. These two values ​​determine the size of the sub-region features extracted in this application. However, since the range of the effective point cloud in each scene is dynamic, if the point cloud sub-region is divided into a fixed size, the size of the sub-region for feature extraction in different scenes will vary greatly. Therefore, set and Some judgment is needed. The details are as follows:

[0079] First, the point cloud range of the kitti dataset is [0, -39.68, -3, 69.12, 39.68, 1]. The size of the body column set in this embodiment is 0.16m*0.16m, so there are a maximum of 432 body columns in the x-coordinate direction of the entire point cloud range, and a maximum of 496 body columns in the y-coordinate direction. It can be seen that and The value of cannot be arbitrarily set to the same fixed value.

[0080] Therefore, in this embodiment, the dynamic setting and First, the standing area of ​​pedestrians is about 0.3m*0.3m, so Set to 216, Setting it to 248 can ensure that the range of each segment is about 0.3m*0.3m, which can accommodate the information of a person. Second, since the point cloud does not cover the entire plane, a threshold is set. hour, ,when hour, . The same can be said hour, ,when hour, . Set it up like this and The value of can ensure that each sub-region can better extract the features of small targets such as pedestrians. Specifically, when W>216, Xs=216, when W≤216, Xs=108, when H>248, Ys=248, when H≤248, Ys=124.

[0081] This application also tests the proposed lidar 3D target detection method, in which all test data are based on the KITTI dataset, including 7481 training samples and 7518 test samples. In order to evaluate the performance of the model, the training samples are divided into 3712 training sets and 3769 validation sets. In all training processes, only the lidar point cloud data of the training set is used, and three categories of cars, pedestrians, and bicycles are trained and validated. The evaluation of each category is divided into three difficulty levels: easy, medium, and difficult according to the size, truncation, and occlusion status of the target. The average precision (AP_R40) is used to evaluate the target detection results of the three categories separately, and the average AP (mAP) of the medium-difficulty AP of the three categories is calculated as the comprehensive performance indicator of the model.

[0082] The test uses the PyTorch framework and is trained and evaluated on a hardware environment equipped with a 3090Ti GPU and an Intel i7-12700K CPU. The network is trained for 80 epochs using the Adam optimizer, and the mini-batch size is set to 16. The initial learning rate is 0.006, and the learning rate is dynamically adjusted using the OneCycle strategy. The maximum number of points per column is limited to 5.

[0083] Table 1 below shows the evaluation results of this embodiment in the bird's eye view (BEV) and compares it with other existing target detection algorithms. The results show that this embodiment has obvious advantages over the baseline model Pointpillar, and the mAP value is increased from 67.62% of the baseline model to 68.85%. It is particularly noteworthy that in terms of pedestrian detection, the pedestrian targets of different difficulties in this embodiment are improved by 3.97%, 4.58%, and 3.33% respectively.

[0084] Table 1 Results of KITTI test BEV detection benchmark

[0085]

[0086] Embodiment 2

[0087] like Figure 5 As shown, this embodiment provides a laser radar 3D target detection system based on regional body column feature fusion, including:

[0088] The data acquisition module 501 is used to acquire the point cloud data set of the target scanning area and the coordinate information of each point cloud data in the point cloud data set.

[0089] The range division module 502 is used to determine the point cloud range of the target scanning area according to the coordinate information of each point cloud data; the point cloud range is represented by a body column feature encoding method.

[0090] The area discretization module 503 is used to discretize the point cloud range into a number of sub-areas based on a preset allocation number; the preset allocation number includes the number to be allocated on the x-axis and the number to be allocated on the y-axis; each sub-area includes a number of point cloud data; each sub-area has a unique area identifier.

[0091] The coordinate calculation module 504 is used to calculate the maximum difference of the z coordinates of each point in each sub-region according to the coordinate information of each point cloud data in the sub-region.

[0092] The feature fusion module 505 is used to initialize a feature array containing only the z coordinate dimension, and store the maximum z coordinate difference in each sub-region; for each sub-region, the maximum z difference of the corresponding region in the feature array is assigned to each point in the sub-region to obtain the body column feature after adding the regional body column feature.

[0093] The target detection module 506 is used to dynamically adjust the preset allocation quantity based on the body column feature after adding the regional body column feature and the area occupied by the target to be detected in the target scanning area to obtain the target feature of the target to be detected.

[0094] Specifically, the regional discrete module includes:

[0095] The size calculation submodule is used to divide the x-coordinate range and the y-coordinate range in the point cloud range based on the number to be allocated on the x-axis and the number to be allocated on the y-axis to obtain the width and height of each sub-area.

[0096] The index calculation submodule is used to calculate the x-axis index and y-axis index of the sub-region index according to the difference between the x-coordinate of the body column and the minimum x-coordinate of the point cloud range, the difference between the y-coordinate of the body column and the minimum y-coordinate of the point cloud range, and the width and height of each sub-region.

[0097] The area division submodule is used to determine the area number of each sub-area according to the x-axis index of the sub-area index, the y-axis index of the sub-area index and the number to be allocated on the y-axis, so as to obtain each sub-area in the point cloud range.

[0098] In summary, this application has the following technical effects:

[0099] At present, 3D target detection algorithms usually use average body column encoding and dynamic average body column encoding methods to encode body column features. However, in practical applications, since the points of each body column are separated, the connection between points and between body columns is lost, which is very unfavorable to the network. At the same time, the receptive field of the network is also an important factor affecting network performance. It is difficult for traditional 3D target detection feature fusion networks to obtain a larger receptive field. This embodiment considers using global average body column encoding to add global features to each body column, and increases the network receptive field through the hole pyramid pooling module based on channel attention, which significantly improves the accuracy of small target detection without losing detection speed. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A 3D target detection method based on regional body column feature fusion laser radar, characterized in that: The method for detecting 3D targets by laser radar based on regional body column feature fusion includes: Obtaining the point cloud data set of the target scanning area and the coordinate information of each point cloud data in the point cloud data set; According to the coordinate information of each point cloud data, the point cloud range of the target scanning area is determined; the point cloud range is represented by a body column feature encoding method; Based on a preset allocation quantity, the point cloud range is discretized into a plurality of sub-areas; the preset allocation quantity includes a quantity to be allocated on the x-axis and a quantity to be allocated on the y-axis; each sub-area includes a plurality of point cloud data; each sub-area has a unique area identifier; For each of the sub-regions, calculating the maximum difference of the z coordinates of the points in the sub-region according to the coordinate information of each point cloud data in the sub-region; Initialize a feature array containing only the z coordinate dimension, and store the maximum difference of the z coordinate in each sub-region; for each sub-region, assign the maximum difference of the z coordinate of the corresponding region in the feature array to each point in the sub-region, and obtain the body column feature after adding the regional body column feature; According to the body column feature after adding the regional body column feature, based on the area occupied by the target to be measured in the target scanning area, the preset allocation quantity is dynamically adjusted to obtain the target feature of the target to be measured; Wherein, according to the coordinate information of each point cloud data, determining the point cloud range of the target scanning area specifically includes: According to the formula , determine the x-coordinate range of the body column; According to the formula , determine the y-coordinate range of the body column; in, , , , , express The maximum value in express The maximum value of express The minimum value of express The minimum value of Indicates the difference between the x-coordinate of the body column and the minimum x-coordinate of the point cloud range. Indicates the difference between the body column y coordinate and the minimum y coordinate of the point cloud range; Wherein, based on the preset allocation quantity, the point cloud range is discretized into a number of sub-areas, specifically including: Based on the number to be allocated on the x-axis and the number to be allocated on the y-axis, the x-coordinate range and the y-coordinate range in the point cloud range are divided to obtain the width and height of each sub-area; Calculate the x-axis index and y-axis index of the sub-region index according to the difference between the x-coordinate of the body column and the minimum x-coordinate of the point cloud range, the difference between the y-coordinate of the body column and the minimum y-coordinate of the point cloud range, and the width and height of each sub-region; Determine the region number of each subregion according to the x-axis index of the subregion index, the y-axis index of the subregion index and the number to be allocated on the y-axis, and obtain each subregion of the point cloud range; Among them, a feature array containing only the z coordinate dimension is initialized to store the maximum difference of the z coordinate in each sub-region; for each sub-region, the maximum difference of the z coordinate of the corresponding region in the feature array is assigned to each point in the sub-region, and the body column feature after adding the regional body column feature is obtained, which specifically includes: ； In the formula, represents the initial body column characteristics, It means to concatenate two tensors in the specified dimension. Represents the body column feature after adding the regional body column feature.

2. The method for 3D target detection based on regional body column feature fusion laser radar according to claim 1, characterized in that: Based on the number to be allocated on the x-axis and the number to be allocated on the y-axis, the x-coordinate range and the y-coordinate range in the point cloud range are divided to obtain the width and height of each sub-area, specifically including: According to the formula , calculate the width of each sub-region; According to the formula , calculate the height of each sub-region; in, Indicates the number of items that need to be allocated on the x-axis. Indicates the amount that needs to be allocated on the y-axis, Indicates the width of each sub-area on the x-axis. Indicates the width of each sub-area on the y-axis.

3. The method for 3D target detection based on regional body column feature fusion laser radar according to claim 2, characterized in that: According to the difference between the body column x coordinate and the minimum x coordinate of the point cloud range, the difference between the body column y coordinate and the minimum y coordinate of the point cloud range, and the width and height of each sub-region, the x-axis index and y-axis index of the sub-region index are calculated, including: According to the formula , calculate the x-axis index of the sub-region index; According to the formula , calculate the y-axis index of the sub-region index; in, The x-axis index representing the sub-region index, The y-axis index representing the sub-region index, Indicates that the maximum value of the restriction variable does not exceed The function of Indicates that the maximum value of the restriction variable does not exceed function.

4. The method for 3D target detection based on regional body column feature fusion laser radar according to claim 3, characterized in that: According to the x-axis index of the sub-region index, the y-axis index of the sub-region index and the number to be allocated on the y-axis, the region number of each sub-region is determined to obtain each sub-region of the point cloud range, specifically including: According to the formula , confirming the area number of each of the sub-areas; Where ID is the area ID.

5. The method for 3D target detection based on regional body column feature fusion laser radar according to claim 4, characterized in that: For each of the sub-regions, according to the coordinate information of each point cloud data in the sub-region, calculating the maximum difference of the z coordinates of each point in the sub-region, specifically includes: According to the formula , calculate the maximum difference of the z coordinates of each point in the sub-area; in, Indicates the maximum difference in the z coordinates of the point cloud in the sub-region, Indicates the maximum z coordinate of the point cloud data in the sub-region, Indicates the minimum z-coordinate of the point cloud data in the sub-region.

6. A 3D target detection system based on regional body column feature fusion laser radar, characterized in that: include: A data acquisition module, used to acquire a point cloud data set of a target scanning area and coordinate information of each point cloud data in the point cloud data set; A range division module is used to determine the point cloud range of the target scanning area according to the coordinate information of each point cloud data; the point cloud range is represented by a body column feature encoding method; Wherein, according to the coordinate information of each point cloud data, determining the point cloud range of the target scanning area specifically includes: According to the formula , determine the x-coordinate range of the body column; According to the formula , determine the y-coordinate range of the body column; in, , , , , express The maximum value in express The maximum value of express The minimum value of express The minimum value of Indicates the difference between the x-coordinate of the body column and the minimum x-coordinate of the point cloud range. Indicates the difference between the body column y coordinate and the minimum y coordinate of the point cloud range; A region discretization module is used to discretize the point cloud range into a number of sub-regions based on a preset allocation quantity; the preset allocation quantity includes the quantity to be allocated on the x-axis and the quantity to be allocated on the y-axis; each sub-region includes a number of point cloud data; each sub-region has a unique region identifier; Among them, the regional discrete module specifically includes: A size calculation submodule, for dividing the x-coordinate range and the y-coordinate range in the point cloud range based on the number to be allocated on the x-axis and the number to be allocated on the y-axis, to obtain the width and height of each sub-area; An index calculation submodule, used to calculate the x-axis index and y-axis index of the sub-region index according to the difference between the x-coordinate of the body column and the minimum x-coordinate of the point cloud range, the difference between the y-coordinate of the body column and the minimum y-coordinate of the point cloud range, and the width and height of each sub-region; A region division submodule, used to determine the region number of each subregion according to the x-axis index of the subregion index, the y-axis index of the subregion index and the number to be allocated on the y-axis, and obtain each subregion of the point cloud range; A coordinate calculation module, for calculating, for each of the sub-regions, the maximum difference of the z coordinates of each point in the sub-region according to the coordinate information of each point cloud data in the sub-region; A feature fusion module is used to initialize a feature array containing only the z coordinate dimension, and store the maximum difference of the z coordinate in each sub-region; for each sub-region, the maximum difference of the z coordinate of the corresponding region in the feature array is assigned to each point in the sub-region, and the body column feature after adding the regional body column feature is obtained; Among them, a feature array containing only the z coordinate dimension is initialized to store the maximum difference of the z coordinate in each sub-region; for each sub-region, the maximum difference of the z coordinate of the corresponding region in the feature array is assigned to each point in the sub-region, and the body column feature after adding the regional body column feature is obtained, which specifically includes: ； In the formula, represents the initial body column characteristics, It means to concatenate two tensors in the specified dimension. It represents the body column feature after adding the regional body column feature; The target detection module is used to obtain the target features of the target to be detected after dynamically adjusting the preset allocation quantity based on the body column features after adding the regional body column features and the area occupied by the target to be detected in the target scanning area.

Images