A method and system for millimeter wave radar point cloud recognition using capsule network

Abstract

The present invention relates to the field of intelligent driving technology, and specifically discloses a method for millimeter-wave radar point cloud recognition using a capsule network, comprising: acquiring point cloud data of a millimeter-wave radar; recognizing the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data; and outputting the classification result of the millimeter-wave radar point cloud data. The present invention also discloses a system for millimeter-wave radar point cloud recognition using a capsule network. The method for millimeter-wave radar point cloud recognition using a capsule network provided by the present invention improves the effect of radar point cloud recognition by using a capsule network, a deep learning algorithm that is more suitable for radar characteristics.

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving technology, and more specifically, to a method for millimeter-wave radar point cloud recognition using a capsule network and a system for millimeter-wave radar point cloud recognition using a capsule network. Background Art

[0002] With the continuous development of science and technology and people's constant pursuit of living standards, more and more research is being conducted in the field of intelligent driving. Intelligent driving is mainly divided into perception system, decision-making system and control system. The perception system mainly uses sensors to perceive and detect moving targets (vehicles, non-motorized vehicles, pedestrians) and stationary targets (telephone poles, traffic lights, road concrete curbs, iron fences, elevated sound insulation boards) in the road environment to obtain information in the environment.

[0003] The capsule network is a neural network structure proposed by Geoffrey Hinton to address some of the shortcomings of convolutional neural networks. First, convolutional neural networks lack the ability to recognize spatial relationships between objects, whereas 4D millimeter-wave radar excels in spatial perception. Furthermore, convolutional neural networks lack the ability to recognize rotated objects. However, radar, due to its principle of transmitting and receiving radio waves, inherently needs to process data generated during rotation. By improving some operational steps in the dynamic routing algorithm of the capsule network and incorporating a multi-head attention mechanism, the system can recognize radar signals with higher accuracy than convolutional neural networks. Summary of the Invention

[0004] In order to address the deficiencies in the prior art, the present invention provides a method and system for millimeter-wave radar point cloud recognition using capsule networks. By using capsule networks, a deep learning algorithm that is more suitable for radar characteristics, the effect of radar point cloud recognition is improved.

[0005] As a first aspect of the present invention, a method for millimeter-wave radar point cloud recognition using a capsule network is provided, comprising:

[0006] Obtain point cloud data from millimeter-wave radar;

[0007] Identifying the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data;

[0008] Output the classification result of the millimeter-wave radar point cloud data.

[0009] Furthermore, the identifying the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data further includes:

[0010] Step S1: Start;

[0011] Step S2: setting the length, width and height of the millimeter-wave radar point cloud data, intercepting the point cloud data within the set length, width and height space and storing it as a high-dimensional vector u;

[0012] Step S3: Multiply the high-dimensional vector u by the weight w to obtain a weighted vector U, as shown in the following formula (1);

[0013] U=w T u (1)

[0014] Step S4: summing the weighted vector U to obtain the summed vector z, as shown in the following formula (2);

[0015] z=∑U (2)

[0016] Step S5: Use a nonlinear function to transform the summed vector z into a vector v, as shown in the following formula (3);

[0017] v=unlinear(z) (3)

[0018] Step S6: Preset the number of iterations r and the inter-layer vector b;

[0019] Step S7: Determine whether the inter-layer vector b reaches the preset number of iterations r. If so, execute step S14; if not, execute step S8.

[0020] Step S8: Use the softmax function to operate on the current inter-layer vector b to obtain the operated vector c, as shown in the following formula (4);

[0021] c=softmax(b) (4)

[0022] Step S9: multiplying the vector v by the calculated vector c to obtain an updated vector z, and then executing steps S10 and S11 respectively;

[0023] Step S10: Use the squash function to operate on the updated vector z, and the result is recorded as vector z1, as shown in the following formula (5);

[0024] z1=squash(z) (5)

[0025] Step S11: using Multi-HeadAttention to operate on the updated vector z, and the result is recorded as vector z2;

[0026] Step S12: Add vector z1 and vector z2 to form an updated vector z, as shown in the following formula (6);

[0027] z=z1+z2 (6)

[0028] Step S13: Add the current inter-layer vector b to the product of the updated vector z and the vector v to form an updated vector b, as shown in the formula (7) below, and then execute step S7;

[0029] b=b+zv (7)

[0030] Step S14: Return the vector z after the iteration is completed;

[0031] Step S15: using the cross entropy function on the iterated vector z to obtain the classification result of the millimeter-wave radar point cloud data;

[0032] Step S16: End.

[0033] Furthermore, the millimeter wave radar includes a 4D millimeter wave radar.

[0034] As a second aspect of the present invention, a system for millimeter-wave radar point cloud recognition using a capsule network is provided, which is used to implement the method for millimeter-wave radar point cloud recognition using a capsule network as described above, comprising:

[0035] Acquisition module, used to obtain point cloud data of millimeter wave radar;

[0036] an identification module, configured to identify the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data;

[0037] The output module is used to output the classification results of the millimeter-wave radar point cloud data.

[0038] The method and system for millimeter-wave radar point cloud recognition using capsule networks provided by the present invention have the following advantages:

[0039] 1. The improved capsule network is more sensitive to the spatial relationship recognition ability of objects;

[0040] 2. The improved capsule network is more sensitive to the recognition ability of objects after translation and rotation;

[0041] 3. Another advantage of the improved capsule network is that when it only uses a small part of the data, it can achieve the same good results as the convolutional neural network that uses all the data, reducing the amount of data collection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation of the present invention.

[0043] Figure 1 This is a flow chart of the method for millimeter-wave radar point cloud recognition using capsule networks according to the present invention.

[0044] Figure 2 This is a flowchart of a specific implementation of the method for millimeter-wave radar point cloud recognition using capsule networks in the present invention. DETAILED DESCRIPTION

[0045] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of the method and system for millimeter-wave radar point cloud recognition using capsule networks proposed in accordance with the present invention. It should be understood that the described embodiments are only a subset of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.

[0046] In this embodiment, a method for millimeter wave radar point cloud recognition using capsule network is provided. Figure 1 As shown in FIG, the method for using capsule network to perform millimeter wave radar point cloud recognition includes:

[0047] Obtain point cloud data from millimeter-wave radar;

[0048] Identifying the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data;

[0049] Output the classification result of the millimeter-wave radar point cloud data.

[0050] Preferably, if Figure 2 As shown, the identifying of the millimeter-wave radar point cloud data to obtain a classification result of the millimeter-wave radar point cloud data further includes:

[0051] Step S1: Start;

[0052] Step S2: setting the length, width and height of the millimeter-wave radar point cloud data, intercepting the point cloud data within the set length, width and height space and storing it as a high-dimensional vector u;

[0053] Step S3: Multiply the high-dimensional vector u by the weight w to obtain a weighted vector U, as shown in the following formula (1);

[0054] U=w T u (1)

[0055] Step S4: summing the weighted vector U to obtain the summed vector z, as shown in the following formula (2);

[0056] z=∑U (2)

[0057] Step S5: Use a nonlinear function to transform the summed vector z into a vector v, as shown in the following formula (3);

[0058] v=unlinear(z) (3)

[0059] Step S6: Preset the number of iterations r and the inter-layer vector b;

[0060] Step S7: Determine whether the inter-layer vector b reaches the preset number of iterations r. If so, execute step S14; if not, execute step S8.

[0061] Step S8: Use the softmax function to operate on the current inter-layer vector b to obtain the operated vector c, as shown in the following formula (4);

[0062] c=softmax(b) (4)

[0063] Step S9: multiplying the vector v by the calculated vector c to obtain an updated vector z, and then executing steps S10 and S11 respectively;

[0064] Step S10: Use the squash function to operate on the updated vector z, and the result is recorded as vector z1, as shown in the following formula (5);

[0065] z1=squash(z) (5)

[0066] Step S11: using Multi-HeadAttention to operate on the updated vector z, and the result is recorded as vector z2;

[0067] Step S12: Add vector z1 and vector z2 to form an updated vector z, as shown in the following formula (6);

[0068] z=z1+z2 (6)

[0069] Step S13: Add the current inter-layer vector b to the product of the updated vector z and the vector v to form an updated vector b, as shown in the formula (7) below, and then execute step S7;

[0070] b=b+zv (7)

[0071] Step S14: Return the vector z after the iteration is completed;

[0072] Step S15: using the cross entropy function on the iterated vector z to obtain the classification result of the millimeter-wave radar point cloud data;

[0073] Step S16: End.

[0074] Preferably, the millimeter wave radar includes a 4D millimeter wave radar.

[0075] As another embodiment of the present invention, a system for millimeter-wave radar point cloud recognition using a capsule network is provided, comprising:

[0076] Acquisition module, used to obtain point cloud data of millimeter wave radar;

[0077] an identification module, configured to identify the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data;

[0078] The output module is used to output the classification results of the millimeter-wave radar point cloud data.

[0079] The present invention provides a method for millimeter-wave radar point cloud recognition using a capsule network, which can improve recognition performance and reduce the amount of required data.

[0080] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Any technician familiar with the present profession can make slight changes or modifications to equivalent embodiments using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for millimeter wave radar point cloud recognition using capsule network, characterized in that: include: Obtain point cloud data from millimeter-wave radar; Identifying the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data; Outputting the classification result of the millimeter-wave radar point cloud data; The step of identifying the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data further includes: Step S1: Start; Step S2: setting the length, width and height of the millimeter-wave radar point cloud data, intercepting the point cloud data within the set length, width and height space and storing it as a high-dimensional vector u; Step S3: Multiply the high-dimensional vector u by the weight w to obtain a weighted vector U, as shown in the following formula (1); （1） Step S4: summing the weighted vector U to obtain the summed vector z, as shown in the following formula (2); （2） Step S5: Use a nonlinear function to transform the summed vector z into a vector v, as shown in the following formula (3); v = unlinear(z) (3) Step S6: Preset the number of iterations r and the inter-layer vector b; Step S7: Determine whether the inter-layer vector b reaches the preset number of iterations r. If so, execute step S14; if not, execute step S8. Step S8: Use the softmax function to operate on the current inter-layer vector b to obtain the operated vector c, as shown in the following formula (4); c=softmax(b) (4) Step S9: multiplying the vector v by the calculated vector c to obtain an updated vector z, and then executing steps S10 and S11 respectively; Step S10: Use the squash function to operate on the updated vector z, and the result is recorded as vector z1, as shown in the following formula (5); z1=squash(z) (5) Step S11: using Multi-Head Attention to operate on the updated vector z, and the result is recorded as vector z2; Step S12: Add vector z1 and vector z2 to form an updated vector z, as shown in the following formula (6); z=z1+z2 (6) Step S13: Add the current inter-layer vector b to the product of the updated vector z and the vector v to form an updated vector b, as shown in the following formula (7), and then execute step S7; b=b+zv (7) Step S14: Return the vector z after the iteration is completed; Step S15: using the cross entropy function on the iterated vector z to obtain the classification result of the millimeter-wave radar point cloud data; Step S16: End.

2. The method for millimeter wave radar point cloud recognition using capsule network according to claim 1, characterized in that: The millimeter wave radar includes a 4D millimeter wave radar.

3. A system for millimeter-wave radar point cloud recognition using a capsule network, used to implement the method for millimeter-wave radar point cloud recognition using a capsule network as described in any one of claims 1-2, characterized in that: include: Acquisition module, used to obtain point cloud data of millimeter wave radar; an identification module, configured to identify the point cloud data of the millimeter-wave radar to obtain a classification result of the millimeter-wave radar point cloud data; The output module is used to output the classification results of the millimeter-wave radar point cloud data.

Images