CN-122023965-A - Training method of millimeter wave detection model, millimeter wave target detection method and device

CN122023965ACN 122023965 ACN122023965 ACN 122023965ACN-122023965-A

Abstract

The application discloses a training method and device of a millimeter wave detection model, and a millimeter wave target detection method and device, and relates to the field of radar target detection, wherein the training method of the millimeter wave detection model comprises the steps of obtaining first point cloud data which is output by scanning a sample space by a millimeter wave radar and second point cloud data which is output by scanning the sample space by a laser radar; the method comprises the steps of obtaining a first point cloud data, obtaining a second point cloud data, processing the second point cloud data by using a laser detection model to obtain a detection labeling result of a sample space, wherein the detection labeling result comprises detection labeling information of at least one sample target in the sample space, taking the first point cloud data as an input sample, taking the detection labeling result as an output sample, and training a millimeter wave detection model, so that the millimeter wave detection model can process millimeter wave point cloud data output by a millimeter wave radar and output a detection result of at least one detection target.

Inventors

ZHANG DONGHUI
SUN MENG

Assignees

北京经纬恒润科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The training method of the millimeter wave detection model is characterized by comprising the following steps of: acquiring first point cloud data output by scanning a sample space by a millimeter wave radar and second point cloud data output by scanning the sample space by a laser radar; Processing the second point cloud data by using a laser detection model to obtain a detection labeling result of the sample space, wherein the detection labeling result comprises detection labeling information of at least one sample target in the sample space; And training a millimeter wave detection model by taking the first point cloud data as an input sample and the detection labeling result as an output sample, so that the millimeter wave detection model can process the millimeter wave point cloud data output by the millimeter wave radar and output a detection result of at least one detection target.
2. The method of claim 1, wherein prior to training the millimeter wave detection model, the method further comprises: According to the corresponding relation between the millimeter wave radar and the laser radar on the scanning time information and the scanning space information, the detection marking result is aligned with the first point cloud data in time and space; Wherein the scan time information characterizes a time difference in which the first point cloud data and the second point cloud data are output; The scan spatial information characterizes a positional distance of the millimeter wave radar and the lidar at a deployment location and an angular difference at a deployment angle.
3. The method of claim 2, wherein the detection labeling result comprises detection labeling results corresponding to the second point cloud data at a plurality of acquisition moments; The time alignment of the detection labeling result and the first point cloud data comprises the following steps: According to the first mode, according to detection labeling results of the second point cloud data at the first acquisition time and the second acquisition time, a first labeling result of the first point cloud data at the third acquisition time is obtained; Wherein the first acquisition time is adjacent to the second acquisition time, and the third acquisition time is between the first acquisition time and the second acquisition time; According to the detection labeling results of the second point cloud data at the first acquisition time and the second acquisition time, a second labeling result of the first point cloud data at the third acquisition time is obtained in a second mode, wherein the first mode is different from the second mode; and fusing detection labeling information belonging to the same sample target according to the first labeling result and the second labeling result to obtain detection labeling results aligned with the first point cloud data in time.
4. A method according to claim 3, wherein the first mode comprises: Using a filtering algorithm, and predicting a detection labeling result of the second point cloud data at the third acquisition time according to a detection labeling result of the second point cloud data at the first acquisition time, a detection labeling result of the second acquisition time and a time difference between the third acquisition time and the first and second acquisition times; And taking a detection labeling result corresponding to the second point cloud data at the third acquisition time as a first labeling result corresponding to the first point cloud data at the third acquisition time.
5. A method according to claim 3, wherein the second mode comprises: In the first point cloud data, according to a detection labeling result corresponding to the first acquisition time and a detection labeling result corresponding to the second acquisition time of the second point cloud data, obtaining first sub point cloud data corresponding to the first acquisition time and second sub point cloud data corresponding to the second acquisition time; And adjusting a detection labeling result corresponding to the second point cloud data at the first acquisition time and a detection labeling result corresponding to the second acquisition time according to sample speed information respectively represented by the first sub point cloud data and the second sub point cloud data so as to obtain a second labeling result corresponding to the first point cloud data at the third acquisition time.
6. The method of claim 2, wherein spatially aligning the detection annotation result with the first point cloud data comprises: According to a coordinate conversion matrix between the laser radar and the millimeter wave radar, performing space conversion on the detection labeling information of the sample target to obtain a detection labeling result which is aligned with the first point cloud data in space; The coordinate transformation matrix is determined according to the deployment positions and the deployment angles of the laser radar and the millimeter wave radar.
7. The method according to any one of claims 1 to 6, wherein the plurality of laser detection models are provided, and the detection labeling result output by each laser detection model comprises detection labeling results corresponding to the second point cloud data at a plurality of acquisition moments; the processing the second point cloud data by using a laser detection model to obtain a detection labeling result of the sample space comprises the following steps: fusing initial labeling information of the sample targets in the detection labeling results output by each laser detection model to obtain fused labeling results, wherein the fused labeling results comprise fused labeling information of at least one sample target; Performing target tracking on sample targets corresponding to different acquisition moments in the fusion labeling result to remove redundant targets belonging to the same sample target among different acquisition moments; and adjusting the fusion labeling information of the rest sample targets by utilizing a fine adjustment model to obtain detection labeling information of at least one sample target in the sample space.
8. A millimeter wave target detection method, characterized by being applied to the millimeter wave detection model of claim 1, comprising: Obtaining millimeter wave point cloud data output by a millimeter wave radar; processing the millimeter wave point cloud data by using the millimeter wave detection model to obtain a detection result output by the millimeter wave detection model, wherein the detection result comprises detection information of at least one detection target; The input sample of the millimeter wave detection model comprises first point cloud data obtained by scanning a sample space by a millimeter wave radar, the output sample of the millimeter wave detection model comprises detection marking information corresponding to the first point cloud data, the detection marking information is obtained by processing second point cloud data by a laser detection model corresponding to a laser radar, and the second point cloud data are point cloud data obtained by scanning the sample space by the laser radar.
9. A training device for a millimeter wave detection model, the device comprising: The point cloud obtaining unit is used for obtaining first point cloud data which are output by scanning a sample space by the millimeter wave radar and second point cloud data which are output by scanning the sample space by the laser radar; The mark obtaining unit is used for processing the second point cloud data by utilizing a laser detection model to obtain a detection mark result of the sample space, wherein the detection mark result comprises detection mark information of at least one sample target in the sample space; the model training unit is used for training the millimeter wave detection model by taking the first point cloud data as an input sample and the detection labeling result as an output sample, so that the millimeter wave detection model can process the millimeter wave point cloud data output by the millimeter wave radar and output the detection result of at least one detection target.
10. A millimeter wave target detection device, characterized by being applied to the millimeter wave detection model of claim 1, comprising: the point cloud obtaining unit is used for obtaining millimeter wave point cloud data output by the millimeter wave radar; The model processing unit is used for processing the millimeter wave point cloud data by using the millimeter wave detection model to obtain a detection result output by the millimeter wave detection model, wherein the detection result comprises detection information of at least one detection target; The input sample of the millimeter wave detection model comprises first point cloud data obtained by scanning a sample space by a millimeter wave radar, the output sample of the millimeter wave detection model comprises detection marking information corresponding to the first point cloud data, the detection marking information is obtained by processing second point cloud data by a laser detection model corresponding to a laser radar, and the second point cloud data are point cloud data obtained by scanning the sample space by the laser radar.

Description

Training method of millimeter wave detection model, millimeter wave target detection method and device Technical Field The present application relates to the field of electronic scanning technologies, and in particular, to a training method of a millimeter wave detection model, a millimeter wave target detection method and a millimeter wave target detection device. Background At present, a fully-supervised target detection model of the 4D millimeter wave radar is usually trained based on manually-marked training samples. For example, the point cloud data of the laser radar which collects the point cloud data simultaneously with the 4D millimeter wave radar is manually marked, and the produced marking frame is multiplexed into the marking frame of the 4D millimeter wave radar. However, the mode relies on manual labeling of laser radar point cloud data, so that the construction efficiency of a training sample is low, and the training efficiency of a target detection model of the 4D millimeter wave radar is seriously affected. Disclosure of Invention In view of the above problems, the application provides a training method of a millimeter wave detection model, a millimeter wave target detection method and a millimeter wave target detection device, so as to achieve the purpose of improving the efficiency of the millimeter wave detection model. The specific scheme is as follows: The first aspect of the present application provides a training method for a millimeter wave detection model, including: acquiring first point cloud data output by scanning a sample space by a millimeter wave radar and second point cloud data output by scanning the sample space by a laser radar; Processing the second point cloud data by using a laser detection model to obtain a detection labeling result of the sample space, wherein the detection labeling result comprises detection labeling information of at least one sample target in the sample space; And training a millimeter wave detection model by taking the first point cloud data as an input sample and the detection labeling result as an output sample, so that the millimeter wave detection model can process the millimeter wave point cloud data output by the millimeter wave radar and output a detection result of at least one detection target. In one possible implementation, before training the millimeter wave detection model, the method further includes: According to the corresponding relation between the millimeter wave radar and the laser radar on the scanning time information and the scanning space information, the detection marking result is aligned with the first point cloud data in time and space; Wherein the scan time information characterizes a time difference in which the first point cloud data and the second point cloud data are output; The scan spatial information characterizes a positional distance of the millimeter wave radar and the lidar at a deployment location and an angular difference at a deployment angle. In one possible implementation, the detection labeling result comprises detection labeling results corresponding to the second point cloud data at a plurality of acquisition moments; The time alignment of the detection labeling result and the first point cloud data comprises the following steps: According to the first mode, according to detection labeling results of the second point cloud data at the first acquisition time and the second acquisition time, a first labeling result of the first point cloud data at the third acquisition time is obtained; Wherein the first acquisition time is adjacent to the second acquisition time, and the third acquisition time is between the first acquisition time and the second acquisition time; According to the detection labeling results of the second point cloud data at the first acquisition time and the second acquisition time, a second labeling result of the first point cloud data at the third acquisition time is obtained in a second mode, wherein the first mode is different from the second mode; and fusing detection labeling information belonging to the same sample target according to the first labeling result and the second labeling result to obtain detection labeling results aligned with the first point cloud data in time. In one possible implementation, the first manner includes: Using a filtering algorithm, and predicting a detection labeling result of the second point cloud data at the third acquisition time according to a detection labeling result of the second point cloud data at the first acquisition time, a detection labeling result of the second acquisition time and a time difference between the third acquisition time and the first and second acquisition times; And taking a detection labeling result corresponding to the second point cloud data at the third acquisition time as a first labeling result corresponding to the first point cloud data at the third acquisition time. In one possible implementation, the second m