3D Radar Research Articles

4D iRIOM: 4D Imaging Radar Inertial Odometry and Mapping

Millimeter wave radar can measure distances, directions, and Doppler velocity for objects in harsh conditions such as fog. The 4D imaging radar with both vertical and horizontal data resembling an image can also measure objects' height. Previous studies have used 3D radars for ego-motion estimation. But few methods leveraged the rich data of imaging radars, and they usually omitted the mapping aspect, thus leading to inferior odometry accuracy. This paper presents a real-time imaging radar inertial odometry and mapping method, iRIOM, based on the submap concept. To deal with moving objects and multipath reflections, we use the graduated non-convexity method to robustly and efficiently estimate ego-velocity from a single scan. To measure the agreement between sparse non-repetitive radar scan points and submap points, the distribution-to-multi-distribution distance for matches is adopted. The ego-velocity, scan-to-submap matches are fused with the 6D inertial data by an iterative extended Kalman filter to get the platform's 3D position and orientation. A loop closure module is also developed to curb the odometry module's drift. To our knowledge, iRIOM based on the two modules is the first 4D radar inertial SLAM system. On our and third-party data, we show iRIOM's favorable odometry accuracy and mapping consistency against the FastLIO-SLAM and the EKFRIO. Also, the ablation study reveal the benefit of inertial data versus the constant velocity model, and scan-to-submap matching versus scan-to-scan matching.

Integration of Remote-Sensing Techniques for the Preventive Conservation of Paleolithic Cave Art in the Karst of the Altamira Cave

Rock art offers traces of our most remote past and was made with mineral and organic substances in shelters, walls, or the ceilings of caves. As it is notably fragile, it is fortunate that some instances remain intact—but a variety of natural and anthropogenic factors can lead to its disappearance. Therefore, as a valuable cultural heritage, rock art requires special conservation and protection measures. Geomatic remote-sensing technologies such as 3D terrestrial laser scanning (3DTLS), drone flight, and ground-penetrating radar (GPR) allow us to generate exhaustive documentation of caves and their environment in 2D, 2.5D, and 3D. However, only its combined use with 3D geographic information systems (GIS) lets us generate new cave maps with details such as overlying layer thickness, sinkholes, fractures, joints, and detachments that also more precisely reveal interior–exterior interconnections and gaseous exchange; i.e., the state of senescence of the karst that houses the cave. Information of this kind is of great value for the research, management, conservation, monitoring, and dissemination of cave art.

3D ground penetrating radar cavity identification algorithm for urban roads using transfer learning

3D ground penetrating radar (GPR) is the main method for the detection of underground cavities in urban roads. The number of road cavity samples detected by 3D radar is small, whereas the intelligent identification model require a large number of learning samples for model training, resulting in inadequate model training. This causes the model to be less accurate in identifying cavities, resulting in many misses and misjudgments. Given the above problems, combined with the detection characteristics of the vertical, the horizontal, and the crossed slices obtained in one detection process of 3D GPR, a 3D GPR cavity intelligent recognition model based on model-based transfer learning is proposed. Firstly, a large amount of 3D GPR data of urban road models with cavities are obtained through forwarding simulation. And the intelligent recognition model was pre-trained on the cavity detection data on three types of slices respectively. Then, through model-based transfer learning, a small amount of real underground cavity data is used to speed up the convergence speed of model training and optimize the structural parameters. It breaks through the limitation of the insufficient number of cavity samples for 3D radar detection on the intelligent learning model training, reduces algorithm training costs, and improves identification accuracy.

Development of “air-ground data fusion” based LiDAR method: Towards sustainable preservation and utilization of multiple-scaled historical blocks and buildings

As important components of urban sustainable development, historical buildings and blocks have critical cultural, economical, and scientific values. Sustainable preservation and utilization of historical buildings/blocks need reliable spatial geometry information, which is commonly generated by three-dimensional (3D) reconstruction. Traditional 3D reconstruction technologies, such as image-based air oblique photography reconstruction technology (OP) and laser-based ground scanning technology (LT), have disadvantages in information integrity and reconstruction accuracy. To overcome these limitations, the current study developed an advanced AGF-LiDAR (“air-ground data fusion” based 3D laser radar method) reconstruction technology. Series of algorithms and hardwares were constructed to realize real-time cooperation between “air-ground” equipment, and real-time fusion of “air-ground” data. The AGF-LiDAR reconstruction technology was verified by practical surveying and mapping of historical buildings, and the real scene scoring system was used in quantitative evaluation. Results show that AGF-LiDAR reconstruction technology can obviously improve reconstruction quality in terms of information integrity and accuracy. The score of AGF-LiDAR reconstruction technology (90 points) was higher than that of traditional manual “air-ground” fusion method (61 points). The AGF-LiDAR reconstruction technology can be applied in various fields, including green urban design, intelligent operation. Overall, the newly developed AGF-LiDAR reconstruction technology is very promising in sustainable heritage preservation.

False Detections Revising Algorithm for Millimeter Wave Radar SLAM in Tunnel

Millimeter wave (MMW) radar simultaneous localization and mapping (SLAM) technology is an emerging technology in a tunnel vehicle accident rescue scene. It is a powerful tool for statistic-trapped vehicle detection with limited vision caused by darkness, heat, and smoke. A variety of SLAM frameworks have been proven to be able to obtain 3-degree-of-freedom (3-dof) pose estimation results using 2-dimention (2D) MMW radar in open space. In the application of millimeter wave radar for pose estimation and mapping in a closed environment, closed space structures and artificial targets together constitute high-intensity multi-path scattering measurement data, leading to radar false detections. Radar false detections caused by multi-path scattering are generally considered to be detrimental to radar applications, such as multi-target tracking. However, few studies analyze the mechanism of multi-path effects on radar SLAM, especially in closed spaces. In order to address the problem, this paper first presents a radar multi-path scattering theory to study the generation mechanism difference of false and radar true detection and their influences on radar SLAM 2D pose estimation and mapping in tunnel. According to the scattering mechanism differences on SLAM, a radar azimuth scattering angle signature is proposed, which allows distinguishing radar false detections from real ones. This is useful in avoiding using unreliable radar false detections to solve a radar SLAM problem. In addition, two different radar false detection revising methods combined with the CSM (correlative scan matching) algorithm are proposed in this paper. The HTMR-CSM (hard-threshold-multi-path-revised correlative scan matching) algorithm only depends on a hard threshold of radar azimuth scattering angle signature to eliminate all radar false detections as much as possible before CSM. Another idea is the STMR-CSM (soft-threshold-multi-path-revised correlative scan matching) algorithm. All the radar false detections are classified according to the distribution model of radar azimuth accuracy, and part of more reliable radar false detections are retained to estimate a more accurate pose. All the ideas in this paper are validated by using an MMW 2D radar mounted on a rail-guided robot in a tunnel. Two cars on fire were set as the targets. The experimental results show that the STMR-CSM algorithm that keeps the reliable radar false detections improves the positioning accuracy by 20% compared with CSM.

Research on Multi-Sensor Fusion SLAM Algorithm Based on Improved Gmapping

Simultaneous Localization and Mapping (SLAM) is the core technology of the intelligent robot system, and it is also the basis for its autonomous movement. In recent years, it has been found that SLAM using a single sensor has certain limitations, such as Inertial Measurement Unit (IMU) noise and serious drift, and 2D radar can only detect environmental information on the same horizontal plane. In this regard, this paper constructs a multi-sensor back-end fusion SLAM algorithm that combines vision, laser, encoder and IMU information. Experiments have proved that compared with using a single sensor, the application of a multi-sensor fusion system makes the edges of the constructed map clearer and the noise reduced. Aiming at the problem of increased calculation caused by particle degradation and too many particles, this paper improves the Gmappping algorithm, and uses the combination of selective resampling and Kullback-Leibler Distance (KLD) sampling to complete resampling. It has been proved by experiments that compared with the original algorithm of Gmapping, the application of the improved algorithm increases the particle convergence speed by 39.85% in the process of indoor mapping. Aiming at the problems that the traditional loop detection algorithm is easily affected by environmental factors, resulting in low detection accuracy, and the loop detection algorithm based on deep convolutional neural network has a large amount of calculation and takes a long time to detect. The main research of this paper is to apply a deep learning-based loop detection algorithm on the multi-sensor fusion framework, and use the combination of high-dimensional and low-dimensional features of the image for loop detection. This paper uses different algorithms to conduct comparative experiments on the dataset CityCentre. The experimental results show that compared with the traditional algorithms Bag of Words (BoW), AlexNet algorithm, VGG19 algorithm, and ResNet32 algorithm, the accuracy of the algorithm proposed in this paper has increased by 31.26%, 14.21%, 3.05%, and 1.56%, respectively. In addition, the comparison experiment results of SLAM mapping with the original Real-Time Appearance-Based Mapping (RTAB-MAP) algorithm prove that the loop closure detection algorithm based on deep learning proposed in this paper can enable the system to better build a globally consistent map, including more environmental information.

Pixel-CRN: A New Machine Learning Approach for Convective Storm Nowcasting

The short-term convective storm forecasting (i.e., nowcasting) mainly rely on weather radar which can resolve the 3D structure of convective storms. With the rapid development of numerical models, modern models can produce 3D reanalysis data which gives atmospheric background information of convective storms. Current deep learning nowcasting models use only 2D radar images for nowcasting and often require massive historical data for training. But it may not be operationally feasible to collect long-term radar data to train a new model. Hence, how to establish a nowcasting model using only a small dataset has become an important issue. In addition, existing models do not effectively use the state-of-the-art model reanalysis data, which is a shortcoming of these models. To tackle these problems, this article develops a pixel-wise convolutional recurrent neural network (Pixel-CRN) for precipitation nowcasting. It has three key designs: (1) Through a concise pixel-wise sampling and oversampling technique, Pixel-CRN can be trained using only a small dataset. (2) In spatial learning, Pixel-CRN embeds a spatial convolution subnet into the recurrent unit, which can input raw 3D radar and model reanalysis data, thus valuable atmospheric background information can be learned to assist nowcasting. (3) In spatiotemporal learning, according to the information bottleneck principle, Pixel-CRN builds a heterogeneous encoder-decoder structure to squeeze multi-channel 3D input data into latent space, and recurrently generates 30 and 60 min nowcasts. Compared with existing deep learning nowcasting methods, the experimental results show that Pixel-CRN can provide skillful results with a rather small training dataset.

Crack Unet: Crack Recognition Algorithm Based on Three-Dimensional Ground Penetrating Radar Images

Three-dimensional (3D) ground-penetrating radar is an effective method for detecting internal crack damage in pavement structures. Inefficient manual interpretation of radar images and high personnel requirements have substantially restrained the generalization of 3D ground-penetrating radar. An improved Crack Unet model based on the Unet semantic segmentation model is proposed herein for 3D ground-penetrating radar crack image processing. The experiment showed that the MPA, MioU, and accuracy of the model were improved, and it displayed better capacity in the radar image crack segmentation task than current mainstream algorithms do, such as deepLabv3, PSPNet, and Unet. In the test dataset without cracks, Crack Unet is on the same level as deepLabv3 and PSPNet, which can meet engineering requirements and display a significant improvement compared with Unet. According to the ablation experiment, the MPA and MioU of Unet configured with PMDA, MC-FS, and RS modules were larger than those of Unet configured with one or two modules. The PMDA module adopted by the Crack Unet model showed a higher MPA and MioU than the SE module and the CBAM module did, respectively. The results show that the Crack Unet model has a better segmentation ability than the current mainstream algorithms do in the task of the crack segmentation of radar images, and the performance of crack segmentation is significantly improved compared with the Unet model. The Crack Unet model has excellent engineering application value in the task of the crack segmentation of radar images.

Test-Retest Reliability of Task Performance for Golf Swings of Medium- to High-Handicap Players.

Background: Golf swing performance in medium- to high-handicap players must be reliably measured to use this variable in both research studies and in applied settings. Nevertheless, there are no studies published on this topic and test−retest evidence is only available for low-handicap players. The aim of this study was to determine the number of attempts necessary to obtain a reliable measurement protocol for swing performance variables in medium- to high-handicap players. Methods: Ten amateur players (55.67 (13.64) years, 78.4 (11.4) kg, 1.75 (7.95) m) took part in a test−retest study in two experimental sessions one week apart. In each one, fifteen swings with a six iron and a driver were evaluated with a 3D Doppler tracking golf radar. Results: The results showed that variables related to side carry could not be reliably measured in medium- to high-handicap players in only fifteen trials (ICC < 0.26, SEM > 12.05 m and MDC > 33.41 m). The rest of the performance variables related to the club and ball trajectories could be reliably measured with a 3D Doppler radar with between seven and ten swings. Conclusions: At least seven swings are recommended for the driver and ten for the six iron to measure golf swing performance.

3D inverse synthetic aperture radar image quality improvement using sparse signal representation

AbstractGeneralisation of one‐dimensional dictionary learning (1DDL) algorithms to Multidimensional (MD) mode and its utilisation in MD data applications, increases the speed and reduces the computational complexity. An example of such an application is 3D inverse synthetic aperture radar (ISAR) image reconstruction and noise reduction. In this study, in addition to MD mode generalisation, the formulation structure of the multidimensional dictionary learning (MDDL) problem is discussed followed by two novel algorithms to solve it. The first one is based on the K‐singular value decomposition algorithm for 1DDL, which uses alternating minimisation and singular value decomposition. The second algorithm is the extension of the sequential generalisation of K‐means 1DDL algorithm to the MD mode. Moreover, the MD tensor denoising method based on the MDDL algorithm (MDDL‐ALG) is proposed. As an application, the proposed method is used to denoise the 3D ISAR image. The numerical simulations reveal that the proposed methods, in addition to reducing the memory consumption and the computational complexity, also enjoy higher convergence rate in comparison to 1D algorithms. Specifically, convergence speed of MD algorithms, depending on the training data size, is up to at least 10 times faster than the equivalent 1D counterparts. As revealed through the simulations, the amount of signal to noise ratio recovered by the proposed methods is almost 2 dB higher than the case using a pre‐designed dictionary for denoising. Moreover, it outperforms about 10 dB over the case with the conventional 3D‐IFFT method for image construction.

Three-Dimensional Convective–Stratiform Echo-Type Classification and Convectivity Retrieval from Radar Reflectivity

Abstract The Echo Classification from COnvectivity (ECCO) algorithm identifies convective and stratiform types of radar echo in three dimensions. It is based on the calculation of reflectivity texture—a combination of the intensity and the heterogeneity of the radar echoes on each horizontal plane in a 3D Cartesian volume. Reflectivity texture is translated into convectivity, which is designed to be a quantitative measure of the convective nature of each 3D radar grid point. It ranges from 0 (100% stratiform) to 1 (100% convective). By thresholding convectivity, a more traditional qualitative categorization is obtained, which classifies radar echoes as convective, mixed, or stratiform. In contrast to previous algorithms, these echo-type classifications are provided on the full 3D grid of the reflectivity field. The vertically resolved classifications, in combination with temperature data, allow for subclassifications into shallow, mid-, deep, and elevated convective features, and low, mid-, and high stratiform regions—again in three dimensions. The algorithm was validated using datasets collected over the U.S. Great Plains during the PECAN field campaign. An analysis of lightning counts shows ∼90% of lightning occurring in regions classified as convective by ECCO. A statistical comparison of ECCO echo types with the well-established GPM radar precipitation-type categories show 84% (88%) of GPM stratiform (convective) echo being classified as stratiform (convective) or mixed by ECCO. ECCO was applied to radar grids for the continental United States, the United Arab Emirates, Australia, and Europe, illustrating its robustness and adaptability to different radar grid characteristics and climatic regions.

3D radar stealth composite hierarchical grid structure with extremely broadband absorption performance and effective load bearing

Multifunctional structures with both excellent microwave absorption performance and mechanical properties have attracted great attentions recently due to wide applications in aircraft and ships. In this research, a novel absorption structure was designed and fabricated through molding and interlocking technology by perfectly combining the composite hierarchical grid structure with excellent mechanical properties and extremely broadband 3D absorption metamaterial. The mechanism of structural parameters affecting absorption and mechanical properties of the proposed 3D radar absorption composite hierarchical grid structure (3D-RACHGS) were investigated. The measured result of reflectivity showed that −10 dB absorption bandwidth of 3D-RACHGS achieves 34.8 GHz (2.6–4 GHz, 6.6–40 GHz) with an average reflectivity of −17.4 dB. Further simulation demonstrated that the broadband absorption is attributed to multiple stationary wave modes. The measurement result of out-of-plane compression showed that the specific strength of the 3D-RACHGS achieves 123.5 × 103 m2/s2, which proves its effective load bearing. Failure mechanism map was also constructed to predict the failure mode of the 3D-RACHGS with various structural parameters. The proposed 3D-RACHGS will contribute to the vigorous development of the novel structural absorption materials and have immense potential to be used in new generation multifunctional skin and inlet.

Drone-Based 3D Synthetic Aperture Radar Imaging with Trajectory Optimization.

This paper presents a trajectory determination and optimization method of multirotors equipped with a single-channel radar to obtain 3D Synthetic Aperture Radar imaging. The result is a realistic trajectory that allows to obtain an imaging of the assumed quality in less time than using a multi-pass trajectory. The optimization criteria, in addition to the cross-range resolution, are the Peak Sidelobe Ratio (PSLR), Integrated Sidelobe Ratio (ISLR), and time of flight. The algorithm is based on a realistic motion model of the radar platform. This paper presents all the steps of the algorithm and provides simulation results that show its practical applicability. The advantage of the presented approach over the existing ones is indicated and further research directions are proposed.

A vision transformer for lightning intensity estimation using 3D weather radar

Lightning has strong destructive powers; its blast wave, high temperature, and high voltage can pose a great threat to human production, life, and personal safety. The destructive power of high-intensity lightning is much greater than that of low-intensity lightning. The estimation of lightning intensity can provide an important reference for determining the lightning protection level and lightning disaster risk assessment. Lightning is a type of small-scale severe convective weather phenomenon. Weather radar is one of the best monitoring systems that can frequently sample the detailed three-dimensional (3D) structures of convective storms, with a small spatial scale and short lifetime at high temporal and spatial resolutions. Therefore, it is possible to extract the 3D spatial feature strongly correlated with lightning from 3D weather radar for estimating lightning intensity. This paper proposes a Vision Transformer model for lightning intensity estimation that can automatically estimate lightning intensity from 3D weather radar data. In an experiment, we transferred the task of estimating lightning intensity into a multicategory classification task. A framework was designed to produce lightning feature samples for model input from 3D weather radar and lightning location data. Then, the Synthetic Minority Over-Sampling Technique (SMOTE) algorithm was used to balance and optimize the sample distribution. Finally, samples were input into the proposed lightning intensity estimation model based on Vision Transformer for training and evaluation. Experimental results show that the proposed model based on Vision Transformers performs well with lightning intensity estimation.

Three-dimensional SAR imaging with sparse linear array using tensor completion in embedded space

Due to the huge data storage and transmission pressure, sparse data collection strategy has provided opportunities and challenges for 3D SAR imaging. However, sparse data brought by the sparse linear array will produce high-level side-lobes, as well as the aliasing and the false-alarm targets. Simultaneously, the vectorizing or matrixing of 3D data makes high computational complexity and huge memory usage, which is not practicable in real applications. To deal with these problems, tensor completion (TC), as a convex optimization problem, is used to solve the 3D sparse imaging problem efficiently. Unfortunately, the traditional TC methods are invalid to the incomplete tensor data with missing slices brought by sparse linear arrays. In this paper, a novel 3D imaging algorithm using TC in embedded space is proposed to produce 3D images with efficient side-lobes suppression. With the help of sparsity and low-rank property hidden in the 3D radar signal, the incomplete tensor data is taken as the input and converted into a higher order incomplete Hankel tensor by multiway delay embedding transform (MDT). Then, the tucker decomposition with incremental rank has been applied for completion. Subsequently, any traditional 3D imaging methods can be employed to obtain excellent imaging performance for the completed tensor. The proposed method achieves high resolution and low-level side-lobes compared with the traditional TC-based methods. It is verified by several numerical simulations and multiple comparative studies on real data. Results clearly demonstrate that the proposed method can generate 3D images with small reconstruction error even when the sparse sampling rate or signal to noise ratio is low, which confirms the validity and advantage of the proposed method.

Design of a new algorithm MFCFT using ISAR imaging to improve the Cross-Range Resolution and Computation Time

SAR imaging is a 2D Radar imaging that produces high resolution images of targets like Ships and Airplane. It uses RD, RID and RIC methods for smoothly rotating, highly rotating and maneuvering targets respectively to generate the images. In this paper, the imaging is done using a new algorithm MFCFT to improve the Cross range resolution and computation time and comparison with different algorithms is shown through the simulated results. Keywords: MFCFT, RD, RID, RIC

Activity Recognition Based on Millimeter-Wave Radar by Fusing Point Cloud and Range–Doppler Information

Millimeter-wave radar has demonstrated its high efficiency in complex environments in recent years, which outperforms LiDAR and computer vision in human activity recognition in the presence of smoke, fog, and dust. In previous studies, researchers mostly analyzed either 2D (3D) point cloud or range–Doppler information from radar echo to extract activity features. In this paper, we propose a multi-model deep learning approach to fuse the features of both point clouds and range–Doppler for classifying six activities, i.e., boxing, jumping, squatting, walking, circling, and high-knee lifting, based on a millimeter-wave radar. We adopt a CNN–LSTM model to extract the time-serial features from point clouds and a CNN model to obtain the features from range–Doppler. Then we fuse the two features and input the fused feature into the full connected layer for classification. We built a dataset based on a 3D millimeter-wave radar from 17 volunteers. The evaluation result based on the dataset shows that this method has higher accuracy than utilizing the two kinds of information separately and achieves a recognition accuracy of 97.26%, which is about 1% higher than other networks with only one kind of data as input.

Transforming 2D Radar Remote Sensor Information from a UAV into a 3D World-View

Since unmanned aerial vehicles (UAVs) have been established in geoscience as a key and accessible tool, a wide range of applications are currently being developed. However, not only the design of UAVs themselves is vital to carry out an accurate investigation, but also the sensors and the data processing are key parts to be considered. Several publications including accurate sensors are taking part in pioneer research programs, but less is explained about how they were designed. Besides the commonly used sensors such as a camera, one of the most popular ones is radar. The advantages of a radar sensor to perform research in geosciences are the robustness, the ability to consider large distances and velocity measurements. Unfortunately, these sensors are often expensive and there is a lack of methodological papers that explain how to reduce these costs. To fill this gap, this article aims to show how: (i) we used a radar sensor from the automotive field; and (ii) it is possible to reconstruct a three-dimensional scenario with a UAV and a radar sensor. Our methodological approach proposes a total of eleven stages to process the radar data. To verify and validate the process, a real-world scenario reconstruction is presented with a system resolution reaching from two to three times the radar resolution. We conclude that this research will help the scientific community to include the use of radars in their research projects and programs, reducing costs and increasing accuracy.

Right-lateral offset associated with the most recent earthquake on the Ikeda fault of the Median Tectonic Line, southwest Japan, revealed by ground-penetrating radar profiling

The Median Tectonic Line (MTL) is an arc-parallel strike-slip fault that accommodates much of the arc-parallel component of the oblique convergence of the Philippine Sea and Eurasian plates at the Nankai Trough. The MTL in Shikoku is one of the fastest-slipping faults in Japan, with a late Quaternary right-lateral slip rate of 5–10 mm/yr. To estimate the right-lateral slip amounts of the past faulting events on the MTL, we acquired 2D and pseudo-3D ground-penetrating radar (GPR) sections across the ENE-trending Ikeda fault of the MTL in eastern Shikoku. We conducted the GPR surveys at the Higashi-Miyoshi site, where two terrace riser offsets mark the active fault trace. The 2D lines were about 28–64 m long, and the pseudo-3D data were sized 20 m × 30 m with a 0.5-m inline spacing. We used 50 MHz GPR antennas and conducted wide-angle measurements to estimate the electromagnetic wave velocity. We identified three paleochannels on the final depth-converted GPR sections, and two of them are deflected by the fault. A paleochannel at 0.6–1.4 m depth is observed on all inline sections of the pseudo-3D GPR data. We built a 3D model of this paleochannel and estimated the right-lateral and vertical displacements of ~ 3.5 m and ~ 0.5 m, respectively. This paleochannel offset is probably caused by the most recent surface-rupturing earthquake on the Ikeda fault, which may be the 1596 Keicho-Fushimi earthquake. This study demonstrates the usefulness of the GPR surveys to identify geological features displaced laterally and vertically by the most recent surface-rupturing earthquake.

A 3-D Storm Motion Estimation Method Based on Point Cloud Learning and Doppler Weather Radar Data

In recent years, deep learning techniques have been developed in the field of storm nowcasting, and they primarily focus on 2D radar product processing. Deep learning’s application to obtaining motion fields is limited by the lack of a motion field ground truth dataset for training. In this paper, we propose a method for storm motion estimation based on a point cloud deep learning network, which can perform 3D motion field estimation on two consecutive frames of weather radar volumetric data. To address the lack of ground truth data, we propose a synthetic data generation method based on Perlin noise and also generate a dataset to train the network. Our model’s architecture is based on FlowNet 3D. In this sense, we import an attention architecture to improve its ability to embed motion features. We also redesign the loss function to make it suitable for our task. In the experiment section, we first evaluate the performance of the trained model on our synthetic dataset. The experimental results demonstrate the following: first, that the trained network has the ability to estimate 3D motion from radar point cloud data; second, that it is a robust solution for radar data with different resolutions and observation ranges; and finally, that for the examined task, our proposed architecture has a 5-15% lower end point error than the original. Then, we directly apply the model trained on synthetic data to real cases, demonstrating the feasibility of our trained model on real data. Although motion field is a low-level product when considering a storm’s evolutionary process, it has the potential to serve a higher-level purpose, such as lightning or hail nowcasting. The point cloud approach provides a new perspective on weather radar data processing.