A dynamic target recognition method based on millimeter-wave radar

Abstract

This invention relates to a dynamic target identification method based on millimeter-wave radar. It receives echo information returned by millimeter-wave radar, divides it into periods to obtain raw messages containing target information for each period, and parses the raw target messages to obtain information values ​​such as position, velocity, and radar cross-section of each target point within that period. The parsed target information values ​​are preprocessed to filter out static targets and some invalid noise points, resulting in millimeter-wave radar data containing dynamic targets and some ghost points. A ghost point filter is then used on the preprocessed data to remove ghost points, revealing the dynamic targets. The dynamic targets are then placed into a point cloud processor, and the PCL library is used to visualize the point cloud of the dynamic targets. This invention solves the problem of being unable to distinguish valid dynamic targets due to the lack of feature points and the presence of a large number of noise points in millimeter-wave radar data.

Description

Technical Field

[0001] This invention belongs to the field of target recognition and relates to a dynamic target recognition method based on millimeter-wave radar. Background Technology

[0002] Millimeter-wave radar has a strong ability to penetrate smoke and dust, can adapt to most harsh weather conditions, and can meet all-weather operation requirements.

[0003] The paper "Millimeter-Wave Radar Cane: Ablind People Aid With Moving Human Recognition Capabilities," published in IEEE JOURNAL OF ELECTROMAGNETICS, RF, AND MICROWAVES INMEDICINE AND BIOLOGY, VOL. 6, NO. 2 in 2022, uses minute chest displacements caused by physiological activity as a key marker of human presence. This allows the proposed system to accurately identify humans in complex environments with multiple moving targets. Due to the characteristics of millimeter-wave radar, its accuracy surpasses most other wireless methods. However, in daily life, moving targets include not only humans but also various vehicles, animals, and objects. Therefore, the author's method based on human vital signs detection for target identification lacks universality. Furthermore, in dynamic scenarios with a large amount of movement, millimeter-wave radar not only detects valid targets but also generates clutter and noise. The raw data points are dense and featureless, making it impossible to identify valid moving targets.

[0004] In 2015, the paper "Road Edge Recognition Using the Stripe Hough Transform From Millimeter-Wave Radar Images," published in *EE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS*, Vol. 16, No. 2, used millimeter-wave radar for road feature recognition. Millimeter-wave radar images typically show roads as a set of bright spots or blotches. The authors introduced a method called the Stripe Hough Transform (HT) to enhance the extraction of geometric features of the road path, thereby achieving road feature recognition. However, in actual millimeter-wave radar detection, pedestrians and vehicles often appear as one or more target points, and there are also a large number of noise points (as shown in the attached image). Figure 2 (This refers to millimeter-wave radar point cloud data for one cycle). Target identification cannot be performed solely based on the features of millimeter-wave radar images. Summary of the Invention

[0005] Technical problems to be solved

[0006] To avoid the shortcomings of existing technologies, this invention proposes a dynamic target recognition method based on millimeter-wave radar, which solves the problem that effective dynamic targets cannot be distinguished due to the lack of feature points and the presence of a large number of noise points in millimeter-wave radar data.

[0007] Technical solution

[0008] A dynamic target recognition method based on millimeter-wave radar, characterized by the following steps:

[0009] Step 1: Divide the echo information returned by the millimeter-wave radar into periods to obtain all the original messages containing target information detected in each period. Then, perform data parsing on the original messages containing target information to obtain the true value of the information in that period.

[0010] Step 2: Preprocess the true values ​​of the parsed target information to filter out static targets and some invalid noise points, and obtain millimeter-wave radar data containing dynamic targets and some ghost points;

[0011] Target information determined to be stationary based on preprocessing is directly filtered out, while target information in a moving state is retained. The target data left in each frame is packaged into a new ROS topic and published.

[0012] Step 3: For millimeter-wave radar data containing dynamic targets and some ghost points, use a ghost point filter to remove ghost points, thereby obtaining the dynamic targets;

[0013] Step 4: Input the dynamic target object into the point cloud processor and use the PCL library to visualize the point cloud of the dynamic target object.

[0014] The data parsing in step 1 is performed according to the CAN protocol manual of ARS408 to obtain the true value of the information carried in the current message.

[0015] The true values ​​of the information within the period include, but are not limited to, the number of targets detected in the close-range beam scan of the current detection period, the number of targets detected in the long-range beam scan of the current detection period, the target ID number, the target position information, the target velocity information, and the target reflection intensity RCS value.

[0016] The preprocessing in step 2 is to compare the speed value of the currently detected target with the speed value of the current vehicle. When a difference is found, the target is in a moving state.

[0017] The ghost point filter is used to remove ghost points, thus obtaining the dynamic target as follows:

[0018] 3.1 Maintain a grid table, where each grid cell stores information values ​​for multiple different targets; the stored information includes: the number of targets in this grid cell, the vector coordinates (x, y) of each target, and the time when each target appeared;

[0019] 3.2 Subscribe to the ROS topic obtained from the preprocessing in step 2, obtain the target data information for the frame period of interest, and traverse each target data information in that frame period;

[0020] 3.3 Convert the horizontal and vertical vector coordinates of the currently traversed target A into raster coordinates, and save the vector coordinate values ​​of target A and the current time into the raster at the current raster coordinate position;

[0021] 3.4 If the number of targets at the current grid position does not reach the threshold, do not proceed to the next step, and return to step 3.3; proceed to the next step only when the threshold is reached.

[0022] 3.5. Perform old period determination on the current grid, traverse the stored target information, and delete targets whose appearance time is in the old period;

[0023] 3.6. Perform the threshold determination in step 3.4 again. If the number of targets in the current grid is still greater than the threshold, put the vector coordinates of target A into the point cloud output queue; otherwise, return to step 3.3.

[0024] Repeat steps 3.3-3.6 until all targets in the current cycle have been traversed.

[0025] Step 4 specifically involves: displaying the targets in the point cloud output queue using PCL point cloud and then clearing the point cloud output queue.

[0026] Beneficial effects

[0027] This invention proposes a dynamic target recognition method based on millimeter-wave radar. It receives echo information returned by millimeter-wave radar, divides it into periods to obtain all raw messages containing target information detected within each period, and parses the raw target messages to obtain information values ​​such as position, velocity, and radar cross-section of each target point within that period. The parsed target information values ​​are preprocessed to filter out static targets and some invalid noise points, resulting in millimeter-wave radar data containing dynamic targets and some ghost points. A ghost point filter is then used on the preprocessed data to remove ghost points, obtaining the dynamic targets. The dynamic targets are then placed into a point cloud processor, and the PCL library is used to visualize the point cloud of the dynamic targets.

[0028] In practical millimeter-wave radar detection, the method of this invention addresses the problem that pedestrians and vehicles are often one or more target points, and there is also a large amount of noise in the data. Target identification can be performed solely by analyzing the features of the millimeter-wave radar image, solving the problem of being unable to distinguish effective dynamic targets due to the lack of feature points and the presence of a large number of noise points in the millimeter-wave radar data. Attached Figure Description

[0029] Figure 1 A framework diagram of a millimeter-wave radar dynamic target tracking method.

[0030] Figure 2 : Raw point cloud image from a one-cycle millimeter-wave radar.

[0031] Figure 3 The dynamic target point cloud image within the current period is obtained by filtering using the method of this invention.

[0032] Figure 4 : The experimental platform of the embodiment. Detailed Implementation

[0033] The present invention will now be further described in conjunction with the embodiments and accompanying drawings:

[0034] The framework of this method is attached. Figure 1 As shown, the raw data detected by millimeter-wave radar is processed by a preprocessor, a ghost point filter, and a point cloud processor, and finally output as a dynamic target object in the form of a point cloud.

[0035] The present invention will be further illustrated below with specific implementation examples. Figure 1 This is a framework diagram of the method; attached. Figure 2 It involves acquiring raw point cloud images from a specific frame period within a dataset; (Attached) Figure 3 The method of the present invention is used to attach Figure 2 Dynamic target point cloud images obtained by filtering out radar data; (Attached) Figure 4 The experimental platform consisted of a millimeter-wave radar mounted at the front of a vehicle; the data was collected from roads surrounding a university. The algorithm was implemented using ROS based on Ubuntu 18.04, version Melodic. The specific steps included:

[0036] 1. Receive echo information returned by millimeter-wave radar, divide the echo information into periods to obtain the original target messages detected in each period, and perform data parsing on the original target messages to obtain the true values ​​of a series of information, including the number of targets detected in the close-range beam scan of the current detection period, the number of targets detected in the long-range beam scan of the current detection period, the target ID number, the target position information, the target velocity information, and the target's RCS value. Specifically, as shown below:

[0037] 1.1. The period mentioned refers to the detection period of the millimeter-wave radar, and the ARS408 detection period used is 72ms. The starting message of each detection period is: indicating the number of targets detected in the current period (let's say n); after this message, there will be n messages indicating information about each detected target, and these n messages are grouped into the same period.

[0038] 1.2. Data Parsing: The messages emitted by the millimeter-wave radar follow the CAN protocol. Each message consists of several hexadecimal characters, and their specific meaning cannot be determined without parsing. Therefore, it is necessary to perform corresponding data parsing according to the ARS408 CAN protocol manual to obtain the true value of the information carried by the current message.

[0039] 2. The parsed target information values ​​are preprocessed to filter out static targets and some invalid noise points, resulting in millimeter-wave radar data containing dynamic targets and some ghost points. The specific preprocessing is as follows:

[0040] 2.1. For each frame of millimeter-wave radar data, there is a corresponding speed value of the vehicle in the current frame; for each target in the frame, there is a corresponding speed value of the target relative to the millimeter-wave radar at the current moment.

[0041] 2.2. By comparing the difference between the speed value of the currently detected target and the speed value of the current vehicle, it can be determined whether each target is stationary or moving relative to the current vehicle.

[0042] 2.3. Static target information is filtered out, while moving target information is retained. The target data retained from each frame is packaged into a new ROS topic and published.

[0043] 3. The data obtained after preprocessing in step 2 is filtered out using a ghost point filter to remove ghost points, thereby obtaining the dynamic target. The details are as follows:

[0044] 3.1. Maintain a raster table, where each raster can store information values ​​for multiple different targets. The stored information includes: the number of targets in this raster, the vector coordinates (x, y) of each target, and the time when each target appeared.

[0045] 3.2. Subscribe to the ROS topic obtained from the preprocessing in step 2 to obtain the target data information for one frame period of interest. Iterate through each target data information in that frame period.

[0046] 3.3. Convert the horizontal and vertical vector coordinates of the currently traversed target (let's call it target A) into raster coordinates, and save the vector coordinate values ​​of target A and the current time into the raster at the current raster coordinate position.

[0047] 3.4. If the number of targets at the current grid position has not reached the threshold, do not proceed to the next step, and return to step 3.3; if the threshold is reached, proceed to the next step (3.5).

[0048] 3.5. Determine the old period for the current grid, traverse the stored target information, and delete targets whose appearance time is in the old period.

[0049] 3.6. Perform the threshold determination in step 3.4 again. If the number of targets in the current grid is still greater than the threshold, add the vector coordinates of target A to the point cloud output queue; otherwise, return to step 3.3.

[0050] 3.7. Repeat steps 3.3-3.6 until all targets in the current cycle have been traversed.

[0051] 4. Place the dynamic target object into the point cloud processor and use the PCL library to visualize the point cloud of the dynamic target object. The details are as follows:

[0052] 4.1. Display the targets in the point cloud output queue using PCL point cloud technology. The display effect is shown in the attached figure. Figure 3 The identified dynamic targets (in this case, pedestrians) are displayed in the form of point clouds.

[0053] 4.2. Clear the point cloud output queue.

Claims

1. A dynamic target recognition method based on millimeter-wave radar, characterized in that... The steps are as follows: Step 1: Divide the echo information returned by the millimeter-wave radar into periods to obtain all the original messages containing target information detected in each period. Perform data parsing on the original messages containing target information to obtain the true value of the information in that period. The true value of the information in the period includes the number of targets detected in the close-range beam scan of the current detection period, the number of targets detected in the long-range beam scan of the current detection period, the target ID number, the target position information, the target velocity information, and the target reflection intensity RCS value. Step 2: Preprocess the true values ​​of the parsed target information to filter out static targets and some invalid noise points, and obtain millimeter-wave radar data containing dynamic targets and some ghost points; The preprocessing in step 2 is to compare the speed value of the currently detected target with the speed value of the current vehicle. When a difference is found, it indicates that the target is moving. Target information determined to be stationary based on preprocessing is directly filtered out, while target information in a moving state is retained. The target data left in each frame is packaged into a new ROS topic and published. Step 3: For millimeter-wave radar data containing dynamic targets and some ghost points, use a ghost point filter to remove ghost points, thereby obtaining the dynamic targets; Step 4: Input the dynamic target object into the point cloud processor and use the PCL library to visualize the point cloud of the dynamic target object.

2. The dynamic target recognition method based on millimeter-wave radar according to claim 1, characterized in that: The data parsing in step 1 is performed according to the CAN protocol manual of ARS408 to obtain the true value of the information carried in the current message.

3. The dynamic target recognition method based on millimeter-wave radar according to claim 1, characterized in that: The ghost point filter is used to remove ghost points, thus obtaining the dynamic target as follows: 3.1 Maintain a raster table, where each raster stores information values ​​for multiple different targets; the stored information includes: the number of targets in this raster, the vector coordinates (x, y) of each target, and the time when each target appeared; 3.2 Subscribe to the ROS topic obtained from the preprocessing in step 2, obtain the target data information for the frame period of interest, and traverse each target data information in that frame period; 3.3 Convert the horizontal and vertical vector coordinates of the currently traversed target A into raster coordinates, and save the vector coordinate values ​​of target A and the current time into the raster at the current raster coordinate position; 3.4 If the number of targets at the current grid position does not reach the threshold, do not proceed to the next step, and return to step 3.3; proceed to the next step only when the threshold is reached. 3.

5. Perform old period determination on the current grid, traverse the stored target information, and delete targets whose appearance time is in the old period; 3.

6. Perform the threshold determination in step 3.4 again. If the number of targets in the current grid is still greater than the threshold, put the vector coordinates of target A into the point cloud output queue; otherwise, return to step 3.

3. Repeat steps 3.3-3.6 until all targets in the current cycle have been traversed.

4. The dynamic target recognition method based on millimeter-wave radar according to claim 1, characterized in that: Step 4 specifically involves: displaying the targets in the point cloud output queue using PCL point cloud and then clearing the point cloud output queue.

Images