A map construction method based on dual-pitch rotation 2D laser radar data fusion

Abstract

The application provides a kind of map construction method based on dual pitch rotation 2D laser radar data fusion, calibrates front radar and rear radar;RANSAS algorithm is used to respectively carry out interframe point cloud data coarse registration on front radar and rear radar;Point cloud similarity matching algorithm is used to supplement radar interframe point cloud data, and interframe point cloud data after fusion is obtained;Repeat the steps to obtain the interframe point cloud data after fusion of next time;ICP algorithm is used to accurately register the interframe point cloud data after fusion. Use double laser radar, i.e. front radar and rear radar, the front radar is limited to the length of the vehicle body and the information of the composting slope that the radar mounting position cannot collect is complete, so the rear radar data is used for supplement, so as to solve the problem of front radar visual blind area, and realize accurate interframe point cloud registration.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to a map construction method based on dual-pitch-rotation 2D lidar data fusion. Background Technology

[0002] Indoor organic fertilizer turning and turning is labor-intensive and operates in harsh environments. Manual labor is not only costly but also poses a significant health risk. Due to the poor signal, humidity, and dust in the fermentation room, technologies such as BeiDou and GPS cannot provide real-time location information, making unmanned turning and turning operations difficult. Therefore, this invention proposes a SLAM-based indoor map construction and automatic control scheme. It utilizes two 2D lidar sensors capable of pitch and rotation to construct a three-dimensional map of the turning and turning environment, laying the foundation for unmanned turning and turning operations.

[0003] In recent years, with the improvement of people's living standards, the scale of livestock and poultry farming has continued to expand, generating approximately 1.73 billion tons of livestock and poultry manure and waste annually. Practice has proven that using livestock and poultry manure as organic fertilizer is a fundamental way to solve waste pollution. Organic fertilizer turning equipment is crucial, but existing turning equipment is energy-intensive and has low automation, resulting in a large workload and low efficiency in organic fertilizer processing. Furthermore, current turning machines require extensive manual operation, and the fermentation process of organic fertilizer produces malodorous and harmful gases such as hydrogen sulfide and indole, jeopardizing the physical and mental health of workers. To address these issues, many scholars have conducted in-depth research on unmanned organic fertilizer turning equipment and proposed the idea of ​​using lidar to construct indoor maps.

[0004] Problems and shortcomings of existing technologies: Currently, 3D LiDAR is too expensive, the 3D mapping algorithm is complex, and the point cloud data calculation is large. Although traditional 2D LiDAR is cheaper, it scans in a plane and cannot acquire ground feature information, so it cannot effectively judge ground features. At the same time, due to the long vehicle body, if the radar is installed in the middle, the radar scanning range is too small. If the radar is installed at the front, it is limited by the long vehicle body, and the radar's ability to collect information from the rear is greatly restricted. Summary of the Invention

[0005] To overcome the aforementioned technical deficiencies, this invention proposes a more accurate map construction method based on dual-laser pitch and rotation 2D radar data fusion.

[0006] The present invention provides a map construction method based on dual-elevation and rotation 2D lidar data fusion as follows: Step 1: Calibrate the front and rear radars; Step 2: Use the RANSAC algorithm to perform coarse registration of inter-frame point cloud data for the front and rear radars respectively; Step 3: Use a point cloud similarity matching algorithm to supplement the radar inter-frame point cloud data to obtain fused inter-frame point cloud data; Step 4: Repeat steps 2-3 to obtain the fused inter-frame point cloud data for the next time step; Step 5: Use the ICP algorithm to accurately register the fused inter-frame point cloud data obtained in Step 3 and Step 4.

[0007] Furthermore, the specific process of step 2 is as follows: The front-end radar acquires m inter-frame point cloud data p1, p2, ..., pm at time t, and m inter-frame point cloud data q1, q2, ..., qm at time t+Δt. The RANSAC algorithm is used to perform coarse registration on the inter-frame point cloud data at time t and time t+Δt to obtain the coarsely registered inter-frame point cloud data L1, L2, ..., Lz of the front-end radar at time t+Δt. Similarly, m inter-frame point cloud data p1', p2', ..., pm' of the rear radar at time t+δt are collected, and m inter-frame point cloud data q1', q2', ..., qm' of the rear radar at time t+δt + Δt are collected. The RANSAC algorithm is used to perform coarse registration of the inter-frame point clouds at time t+δt and time t+δt+Δt, resulting in the coarsely registered inter-frame point cloud data L'1, L'2, ..., L'z of the rear radar at time t+δt+Δt.

[0008] Furthermore, the specific process of step 3 is as follows: 3a) Divide the inter-frame point cloud data L1, L1, ..., Lz of the front radar at time t+△t into n regions, namely A1, A2, ..., An. Calculate the covariance V1, V2, ..., Vn of the inter-frame point cloud data of the front radar in the n regions respectively, and then calculate the Mahalanobis distance d1, d2, ..., dn corresponding to each covariance; The post-radar inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt are divided into n regions, namely B1, B2, ..., Bn, and the sizes of B1, B2, ..., Bn correspond one-to-one with A1, A2, ..., An; the Mahalanobis distances d1', d2', ..., dn' corresponding to B1, B2, ..., Bn are calculated sequentially using the method for calculating Mahalanobis distance d1. 3b) Each region of the front-end radar is paired one-to-one with each of the n regions of the rear-end radar. The specific process is as follows: Calculate the absolute value of the difference between the Mahalanobis distance d1' of B1 and the Mahalanobis distance d1 of A1. D11 = |d1' - d1|, and so on, calculate D12 = |d1' - d2|, D13 = |d1' - d3|, ..., D1n = |d1' - dn|; Similarly, calculate D21, D22, ..., D2n, D21, D22, ..., D2n, ..., Dn1, Dn2, ..., Dnn in sequence, and compare D11, D12, ..., Dnn with the preset similarity threshold S2. If the similarity is greater than the preset similarity threshold S2, the match is successful, and the number of successfully matched regions is recorded. If the number of successfully matched regions is less than the preset number of matches threshold S3, proceed to step 3e). If the number of successfully matched regions is greater than the preset number of matches threshold S3, proceed to step 3c). 3c) Calculate the point cloud supplementary transformation matrix H1; 3d) The z inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt of the rear radar are supplemented into the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar using the point cloud supplementation transformation matrix H1 to complete the fusion and obtain the k fused inter-frame point cloud data O1, O2, ..., Ok; 3e) When the number of iterations is less than or equal to the preset number of iterations, the z inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt of the rear radar and the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar are re-divided into regions, and steps 3a) to 3b) are repeated. When the number of iterations is greater than the preset number of iterations, the iteration is stopped, and the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar are directly and precisely registered, while the number of iterations is reset to zero.

[0009] Furthermore, in step 3a), the specific calculation process of the Mahalanobis distance d1 is as follows: V1= T , = ; d1=[( ) T V1 -1 ( )] 1 / 2 μ1 = (μ11, μ12, ..., μ1e) T ; x 1 is a vector of mean μ1 and covariance V1, i.e. x 1=( x 11, x 12,..., x 1e) T Similarly, calculate the Mahalanobis distances d2, d3, ..., dn in sequence.

[0010] Further, in step 3c), the calculation process of the point cloud supplementation transformation matrix H1 is as follows: Define a three-dimensional coordinate vector = In a three-dimensional coordinate system Rotation about the z-axis The rotation matrix Rz is obtained from angle 1. 1)= In a three-dimensional coordinate system Rotate about the x-axis The rotation matrix Rx is obtained from angle 1'. )= In a three-dimensional coordinate system Rotation around the y-axis The rotation matrix Ry is obtained from the 1'' angle. '')= Obtain the point cloud supplementary rotation matrix R1= ; The point cloud is supplemented by translation matrix T1 by shifting x1 along the x-axis, y1 along the y-axis, and z1 along the z-axis. ; Inter-frame point cloud data only undergoes rotation and translation transformations, without deformation, and the scaling factor is 1, thus obtaining the point cloud supplementary transformation matrix H1= .

[0011] Furthermore, the specific process of step 4 is as follows: Repeat steps 2 to 3 to obtain the k inter-frame point cloud data W1, W2, ... Wk after fusion at time t+δt+△t', where △t'>△t.

[0012] Furthermore, the specific process of step 5 is as follows: 5a) Calculate the exact registration transformation matrix H2; 5b) Using the precise registration transformation matrix H2, transform the k inter-frame point cloud data O1, O2, ... Ok after fusion at time t+δt+Δt to obtain O1', O2', ..., Ok' respectively; 5c) Calculate the average distance d ={||O1'-W1|| 2 +|| O2'-W2|| 2 +……+|| Ok'-Wk|| 2} / k; if d If the value is less than the precise registration threshold S4, the calculation is complete. d If the value is greater than or equal to the precise registration threshold S4, then repeat steps 2 to 4.

[0013] Furthermore, the calculation process of the precise registration transformation matrix H2 in step 5a) is as follows: In a three-dimensional coordinate system Rotation about the z-axis The rotation matrix Rz is obtained from the two angles. 2)= In a three-dimensional coordinate system Rotate about the x-axis The rotation matrix Rx is obtained from the 2' angle. )= In a three-dimensional coordinate system Rotation around the y-axis The rotation matrix Ry is obtained from the 2'' angle. '')= To obtain the accurate registration rotation matrix R2= ; The precise registration translation matrix T2 is obtained by translating x2 on the x-axis, y2 on the y-axis, and z2 on the z-axis. ; The inter-frame point cloud data only undergoes rotation and translation transformations, without deformation, and the scaling factor is 1, thus obtaining the accurate registration transformation matrix H2= .

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses dual lidar, namely front and rear radar. The front radar is limited by the vehicle length and radar installation position and cannot collect complete information about the compost back slope. Therefore, the data from the rear radar is used to supplement it, thereby solving the problem of blind spots in the field of view of the front radar. 2. This invention proposes a point cloud similarity matching algorithm. The inter-frame point cloud data collected by the front radar and the rear radar are divided into regions, and the Mahalanobis distance of each region is calculated. Similarity matching is then performed to supplement the point cloud data of the rear radar to the corresponding position of the front radar to achieve data fusion. Finally, point cloud registration correction is performed to achieve accurate inter-frame point cloud registration. Attached Figure Description

[0015] Figure 1 This is a diagram illustrating the map construction method using a dual-pitch-rotation 2D lidar as described in this invention. Figure 2 This is a supplementary illustration of the map construction method using a dual-pitch-rotation 2D lidar as described in this invention; Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and specific implementation steps.

[0017] This invention relates to a map construction method based on dual-elevation and rotation 2D lidar data fusion, the specific process of which is as follows: Step 1: Calibrate the front-end and rear-end radars using the NDT registration method. Step 2: Use the RANSAC algorithm to perform coarse registration of inter-frame point cloud data for the front and rear radars respectively. The front-end radar acquires m inter-frame point cloud data p1, p2, ..., pm at time t, and m inter-frame point cloud data q1, q2, ..., qm at time t+Δt. The RANSAC algorithm is used to perform coarse registration on the inter-frame point cloud data at time t and time t+Δt to obtain the coarsely registered inter-frame point cloud data L1, L2, ..., Lz of the front-end radar at time t+Δt. Similarly, m inter-frame point cloud data p1', p2', ..., pm' of the rear radar at time t+δt are collected, and m inter-frame point cloud data q1', q2', ..., qm' of the rear radar at time t+δt + Δt are collected. The RANSAC algorithm is used to perform coarse registration of the inter-frame point clouds at time t+δt and time t+δt+Δt, resulting in the coarsely registered inter-frame point cloud data L'1, L'2, ..., L'z of the rear radar at time t+δt+Δt. Step 3: Use a point cloud similarity matching algorithm to supplement the radar inter-frame point cloud data to obtain fused inter-frame point cloud data; 3a) Divide the inter-frame point cloud data L1, L1, ..., Lz of the front radar at time t+△t into n regions, namely A1, A2, ..., An. Calculate the covariance V1, V2, ..., Vn of the inter-frame point cloud data of the front radar in the n regions respectively, and then calculate the Mahalanobis distance d1, d2, ..., dn corresponding to each covariance; The specific calculation process for the Mahalanobis distance d1 is as follows: V1= T , = ; d1=[( ) T V1 -1 ( )] 1 / 2 μ1 = (μ11, μ12, ..., μ1e) T ; x 1 is a vector of mean μ1 and covariance V1, i.e. x 1=( x 11, x 12,..., x 1e) T ; Similarly, calculate the Mahalanobis distances d2, d3, ..., dn in sequence; The post-radar inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt are divided into n regions, namely B1, B2, ..., Bn, and the sizes of B1, B2, ..., Bn correspond one-to-one with A1, A2, ..., An (i.e., the size of B1 is the same as the size of A1, and so on). The Mahalanobis distances d1', d2', ..., dn' corresponding to B1, B2, ..., Bn are calculated sequentially according to the method for calculating Mahalanobis distance d1. 3b) Each region of the front-end radar is paired one-to-one with each of the n regions of the rear-end radar. The specific process is as follows: Calculate the absolute value of the difference between the Mahalanobis distance d1' of B1 and the Mahalanobis distance d1 of A1. D11 = |d1' - d1|, and so on, calculate D12 = |d1' - d2|, D13 = |d1' - d3|, ..., D1n = |d1' - dn|; Similarly, calculate D21, D22, ..., D2n, D21, D22, ..., D2n, ..., Dn1, Dn2, ..., Dnn in sequence, and compare D11, D12, ..., Dnn with the preset similarity threshold S2. If the similarity is greater than the preset similarity threshold S2, the match is successful, and the number of successfully matched regions is recorded. If the number of successfully matched regions is less than the preset number of matches threshold S3, proceed to step 3e). If the number of successfully matched regions is greater than the preset number of matches threshold S3, proceed to step 3c). 3c) Calculate the point cloud supplementary transformation matrix H1. The calculation process is as follows: Define a three-dimensional coordinate vector = In a three-dimensional coordinate system Rotation about the z-axis The rotation matrix Rz is obtained from angle 1. 1)= In a three-dimensional coordinate system Rotate about the x-axis The rotation matrix Rx is obtained from angle 1'. )= In a three-dimensional coordinate system Rotation around the y-axis The rotation matrix Ry is obtained from the 1'' angle. '')= Obtain the point cloud supplementary rotation matrix R1= ; The point cloud is supplemented by translation matrix T1 by shifting x1 along the x-axis, y1 along the y-axis, and z1 along the z-axis. ; Inter-frame point cloud data only undergoes rotation and translation transformations, without deformation, and the scaling factor is 1, thus obtaining the point cloud supplementary transformation matrix H1= ; 3d) The z inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt of the rear radar are supplemented into the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar using the point cloud supplementation transformation matrix H1 to complete the fusion and obtain the k fused inter-frame point cloud data O1, O2, ..., Ok; 3e) When the number of iterations is less than or equal to the preset number of iterations, the z inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt of the rear radar and the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar are re-divided into regions, and steps 3a) to 3b) are repeated. When the number of iterations is greater than the preset number of iterations, the iteration is stopped, and the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar are directly and precisely registered, while the number of iterations is reset to zero. Step 4: Repeat steps 2-3 to obtain the k inter-frame point cloud data W1, W2, ... Wk after fusion at time t+δt+△t' (△t'>△t). Step 5: Use the ICP algorithm to perform precise registration of the fused inter-frame point cloud data obtained in Step 3 and Step 4. 5a) Calculate the exact registration transformation matrix H2. The calculation process is as follows: In a three-dimensional coordinate system Rotation about the z-axis The rotation matrix Rz is obtained from the two angles. 2)= In a three-dimensional coordinate system Rotate about the x-axis The rotation matrix Rx is obtained from the 2' angle. )= In a three-dimensional coordinate system Rotation around the y-axis The rotation matrix Ry is obtained from the 2'' angle. '')= To obtain the accurate registration rotation matrix R2= ; The precise registration translation matrix T2 is obtained by translating x2 on the x-axis, y2 on the y-axis, and z2 on the z-axis. ; The inter-frame point cloud data only undergoes rotation and translation transformations, without deformation, and the scaling factor is 1, thus obtaining the accurate registration transformation matrix H2= ; 5b) Using the precise registration transformation matrix H2, transform the k inter-frame point cloud data O1, O2, ... Ok after fusion at time t+δt+Δt to obtain O1', O2', ..., Ok' respectively; 5c) Calculate the average distance d ={||O1'-W1|| 2 +|| O2'-W2|| 2 +……+|| Ok'-Wk|| 2} / k; if d If the value is less than the precise registration threshold S4, the calculation is complete. d If the value is greater than or equal to the precise registration threshold S4, then repeat steps 2 to 4.

[0018] This invention relates to a map construction method based on dual-pitch-rotation 2D lidar data fusion. It employs a pitch-rotation device to enable 2D radar to acquire three-dimensional point cloud information. By using dual lidars, the point cloud data from the rear lidar is used to supplement the point cloud data from the front lidar, which solves the problem of small scanning range and scanning limitation by vehicle length of a single pitch-rotation radar. When using dual lidars, the data from the rear lidar is used to supplement the scanning blind spots of the front lidar, such as the back slope area, thereby increasing the scanning range and avoiding the impact of incomplete mapping on subsequent automatic navigation.

[0019] In the above scheme, when scanning blind spots with radar: if a single radar is used, due to the long vehicle body, installing the radar in the middle would result in a very small radar scanning range; if the radar is installed at the front, the long vehicle body would severely limit the radar's ability to collect information from the rear, making it impossible to collect complete information. Therefore, this invention employs dual radars. Using dual radars... Figure 1 and Figure 2 As shown, in Figure 1 At position 1, the car is driving on a flat slope and cannot scan the back slope information. Figure 2 At position 1', the car is on the verge of going downhill, and the front radar still cannot scan the complete back slope information. During the descent to position 2, the front radar's scanning of the rear is severely limited due to the car's length and forward mounting position, making it unable to collect complete back slope information. Therefore, inter-frame point cloud data from the rear radar is needed to supplement this information.

[0020] The present invention has been described in detail above with reference to specific embodiments, but it is not limited to the above content. Any simple modifications based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A map construction method based on dual-tilt-rotation 2D lidar data fusion, characterized in that: The construction method is as follows: Step 1: Calibrate the front and rear radars; Step 2: Use the RANSAC algorithm to perform coarse registration of inter-frame point cloud data for the front and rear radars respectively; Step 3: Use a point cloud similarity matching algorithm to supplement the radar inter-frame point cloud data to obtain fused inter-frame point cloud data; Step 4: Repeat steps 2-3 to obtain the fused inter-frame point cloud data for the next time step; Step 5: Use the ICP algorithm to accurately register the fused inter-frame point cloud data obtained in Step 3 and Step 4; The specific process of step 2 is as follows: The front-end radar acquires m inter-frame point cloud data p1, p2, ..., pm at time t, and m inter-frame point cloud data q1, q2, ..., qm at time t+Δt. The RANSAC algorithm is used to perform coarse registration on the inter-frame point cloud data at time t and time t+Δt to obtain the coarsely registered inter-frame point cloud data L1, L2, ..., Lz of the front-end radar at time t+Δt. Similarly, m inter-frame point cloud data p1', p2', ..., pm' of the rear radar at time t+δt are collected, and m inter-frame point cloud data q1', q2', ..., qm' of the rear radar at time t+δt + Δt are collected. The RANSAC algorithm is used to perform coarse registration of the inter-frame point clouds at time t+δt and time t+δt+Δt, resulting in the coarsely registered inter-frame point cloud data L'1, L'2, ..., L'z of the rear radar at time t+δt+Δt. The specific process of step 3 is as follows: 3a) Divide the inter-frame point cloud data L1, L1, ..., Lz of the front radar at time t+△t into n regions, namely A1, A2, ..., An. Calculate the covariance V1, V2, ..., Vn of the inter-frame point cloud data of the front radar in the n regions respectively, and then calculate the Mahalanobis distance d1, d2, ..., dn corresponding to each covariance; The post-radar inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt are divided into n regions, namely B1, B2, ..., Bn, and the sizes of B1, B2, ..., Bn correspond one-to-one with A1, A2, ..., An; the Mahalanobis distances d1', d2', ..., dn' corresponding to B1, B2, ..., Bn are calculated sequentially using the method for calculating Mahalanobis distance d1. 3b) Each region of the front-end radar is paired one-to-one with each of the n regions of the rear-end radar. The specific process is as follows: Calculate the absolute value of the difference between the Mahalanobis distance d1' of B1 and the Mahalanobis distance d1 of A1. D11 = |d1' - d1|, and so on, calculate D12 = |d1' - d2|, D13 = |d1' - d3|, ..., D1n = |d1' - dn|; Similarly, calculate D21, D22, ..., D2n, D21, D22, ..., D2n, ..., Dn1, Dn2, ..., Dnn in sequence, and compare D11, D12, ..., Dnn with the preset similarity threshold S2. If the similarity is greater than the preset similarity threshold S2, the match is successful, and the number of successfully matched regions is recorded. If the number of successfully matched regions is less than the preset number of matches threshold S3, proceed to step 3e). If the number of successfully matched regions is greater than the preset number of matches threshold S3, proceed to step 3c). 3c) Calculate the point cloud supplementary transformation matrix H1; 3d) The z inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt of the rear radar are supplemented into the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar using the point cloud supplementation transformation matrix H1 to complete the fusion and obtain the k fused inter-frame point cloud data O1, O2, ..., Ok; 3e) When the number of iterations is less than or equal to the preset number of iterations, the z inter-frame point cloud data L'1, L'2, ..., L'z at time t+δt+Δt of the rear radar and the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar are re-divided into regions, and steps 3a) to 3b) are repeated. When the number of iterations is greater than the preset number of iterations, the iteration is stopped, and the z inter-frame point cloud data L1, L2, ..., Lz at time t+Δt of the front radar are directly and precisely registered, while the number of iterations is reset to zero.

2. The map construction method based on dual-tilt-rotation 2D lidar data fusion according to claim 1, characterized in that: In step 3a), the specific calculation process of the Mahalanobis distance d1 is as follows: V1= T ， = ； d1=[( ) T V1 -1 ( )] 1 / 2 ,μ1=(μ11,μ12,...,μ1e) T ; x 1 is a vector of mean μ1 and covariance V1, i.e. x 1=( x 11, x 12,..., x 1e) T Similarly, calculate the Mahalanobis distances d2, d3, ..., dn in sequence.

3. The map construction method based on dual-tilt-rotation 2D lidar data fusion according to claim 1, characterized in that: In step 3c), the calculation process of the point cloud supplementary transformation matrix H1 is as follows: Define a three-dimensional coordinate vector = In a three-dimensional coordinate system Rotation about the z-axis The rotation matrix Rz is obtained from angle 1. 1)= In a three-dimensional coordinate system Rotate about the x-axis The rotation matrix Rx is obtained from angle 1'. )= In a three-dimensional coordinate system Rotation around the y-axis The rotation matrix Ry is obtained from the 1'' angle. '')= Obtain the point cloud supplementary rotation matrix R1= ; The point cloud is supplemented by translation matrix T1 by shifting x1 along the x-axis, y1 along the y-axis, and z1 along the z-axis. ; Inter-frame point cloud data only undergoes rotation and translation transformations, without deformation, and the scaling factor is 1, thus obtaining the point cloud supplementary transformation matrix H1= .

4. The map construction method based on dual-tilt-rotation 2D lidar data fusion according to claim 1, characterized in that: The specific process of step 4 is as follows: Repeat steps 2 to 3 to obtain the k inter-frame point cloud data W1, W2, ... Wk after fusion at time t+δt+△t', where △t'>△t.

5. The map construction method based on dual-tilt-rotation 2D lidar data fusion according to claim 4, characterized in that: The specific process of step 5 is as follows: 5a) Calculate the exact registration transformation matrix H2; 5b) Using the precise registration transformation matrix H2, transform the k inter-frame point cloud data O1, O2, ... Ok after fusion at time t+δt+Δt to obtain O1', O2', ..., Ok' respectively; 5c) Calculate the average distance d ={||O1'-W1|| 2 +|| O2'-W2|| 2 +……+|| Ok'-Wk|| 2 } / k; if d If the value is less than the precise registration threshold S4, the calculation is complete. d If the value is greater than or equal to the precise registration threshold S4, then repeat steps 2 to 4.

6. The map construction method based on dual-tilt-rotation 2D lidar data fusion according to claim 5, characterized in that: The calculation process for the precise registration transformation matrix H2 in step 5a) is as follows: In a three-dimensional coordinate system Rotation about the z-axis The rotation matrix Rz is obtained from the two angles. 2)= In a three-dimensional coordinate system Rotate about the x-axis The rotation matrix Rx is obtained from the 2' angle. )= In a three-dimensional coordinate system Rotation around the y-axis The rotation matrix Ry is obtained from the 2'' angle. '')= To obtain the accurate registration rotation matrix R2= ; The precise registration translation matrix T2 is obtained by translating x2 on the x-axis, y2 on the y-axis, and z2 on the z-axis. ; The inter-frame point cloud data only undergoes rotation and translation transformations, without deformation, and the scaling factor is 1, thus obtaining the accurate registration transformation matrix H2= .

Images