VEHICLE RADAR SYSTEM FOR AUTOMATIC CLUTTER REMOVAL IN MULTI-DIMENSIONAL SPACE AND METHOD FOR THIS

Abstract

The present disclosure relates to a vehicle radar system (102) for automatic multi-clutter removal in multidimensional space and a method (400) for this purpose. The system (102) includes a controller (108) that removes one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique. The controller (108) processes one or more radar signals to project radar PCD onto a dense multidimensional space containing a LiDAR voxel grid and to classify the voxel status of an object. The controller (108) dynamically predicts and identifies one or more superimposed radar PCDs that lie on the dense LiDAR voxel grid.The controller (108) filters out the multi-clutter with the binary condition “0” and retains the PCD data with the binary condition “1” to automatically generate ground truth data to train a deep learning model that initiates automatic multi-clutter removal based on ground truth data in the vehicle radar system (102).

Description

The present disclosure relates to the field of clutter filters. In particular, the present disclosure provides for a vehicle radar system for automatic multi-clutter removal in multidimensional space and a method for this. Current, state-of-the-art methods for removing clutter from radar face several challenges, including the need for manual threshold adjustment, which is tedious and difficult to determine accurately. Deep learning-based approaches rely heavily on a ground-truth pipeline to effectively remove clutter. Most existing frameworks require manual annotations, which are prone to errors. Furthermore, existing frameworks primarily focus on removing a single type of clutter. This reliance on precise ground-truth data reduces robustness in dynamic environments. Many techniques have been developed to eliminate the aforementioned problems. For example, patent document CN116386001A describes a method for removing suspended ghost point clouds, a method for automated driving, and associated equipment. The method captures an initial point cloud and image data of a current field of view, followed by semantic segmentation of the image to identify a predefined area. Obstacle point clouds are extracted from the initial data and projected onto the segmented image. If the obstacle area lies within the predefined partition, it is identified as a suspended ghost point cloud and removed. This ensures that the vehicle can avoid ghost point problems in drivable road areas. However, the aforementioned document does not specify a strict dependency on the camera during the inference process for multi-clutter filters.Furthermore, the aforementioned document makes no mention of the loss of depth due to camera projection. The further patent document CN114612329A describes a method, system, device, and medium for ground filtering based on a three-dimensional laser point cloud. This process involves downsampling the point cloud using voxel filtering and then converting the point cloud into a polar coordinate system for sorting. Ground points are filtered by fitting the points of each segment to a line and applying grid conditions based on at least one slope, intersection, and distance. The remaining points are projected onto a 2D image for further processing. However, the cited document contains no information on the automatic generation of the ground truth or the zero-velocity multi-clutter detection. Patent document US20240159871A1 describes predicting a set of new positions of a set of objects in a geographic region based on an initial set of object tracks, in order to obtain a set of predicted object positions, and obtaining a new LiDAR (Light Detection and Ranging) point cloud of the geographic region. A detector model processes the new LiDAR point cloud to obtain a new set of bounding frames around the set of objects detected in the new LiDAR point cloud.Object detection also involves matching the new set of bounding boxes with the set of predicted object positions to generate a set of matches, updating the initial set of object tracks with the new set of bounding boxes according to the set of matches to obtain an updated set of object tracks, and, after updating, filtering the updated set of object tracks to remove those that do not meet a track length threshold to generate a training set of object tracks. However, the cited document contains no information on the automatic generation of ground truth or the detection of zero-velocity multi-clutter. Therefore, there is a need for a vehicle radar system capable of automatic multi-clutter removal in multidimensional space and a method that addresses at least the aforementioned issues. A general objective of the present disclosure is to provide a vehicle radar system and a method for enabling automatic multi-clutter removal by means of LiDAR (Light Detection and Ranging) monitoring in multidimensional spaces. One objective of the present disclosure is to provide an automated, robust and scalable approach applicable to all sensor types, radar manufacturers and multiple clutter types. Another purpose of the present disclosure is to ensure that multi-clutter is minimized, thereby improving the radar system's performance in detecting and tracking relevant targets. Another purpose of the present disclosure is to eliminate errors caused by manual marking or fine-tuning of threshold values ​​in radar point cloud data. The present disclosure relates to the field of clutter filters. In particular, the present disclosure provides a vehicle radar system for enabling automatic multi-clutter removal in a multidimensional space and a method for this purpose. One aspect of the present disclosure relates to a vehicle radar system for enabling automatic multi-clutter removal in a multidimensional space. The vehicle radar system includes at least one transmitter, at least one receiver, and a controller. The at least one transmitter is configured to generate and transmit one or more radar signals. The at least one receiver is communicatively coupled to the at least one transmitter and configured to receive the one or more radar signals. The controller is communicatively coupled to the at least one receiver and the at least one transmitter. The controller is configured to remove one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique.The controller is configured to process one or more radar signals to project radar PCDs into the multidimensional LiDAR PCD space. Furthermore, the controller is configured to project the radar PCDs onto a dense LiDAR voxel grid in the multidimensional space and classify a voxel status of the object. The voxel status refers to a binary condition that includes at least one of "1" and one of "0", representing an occupied and an unoccupied state, respectively. Additionally, the controller is configured to dynamically predict and identify one or more superimposed radar point cloud datasets on the dense LiDAR voxel grid.Furthermore, the control system is configured to filter out the multi-clutter with the binary condition "0" and retain relevant radar point cloud data with the binary condition "1" to automatically generate ground truth data to train a deep learning model that initiates automatic multi-clutter removal based on the ground truth data in the vehicle radar system. In some embodiments, the point cloud data may include at least one of LiDAR point cloud data and radar point cloud data. In some embodiments, the controller can be configured to automatically categorize the point cloud data into the object and the multi-clutter based on a predefined area. In some embodiments, the object may include at least one of a vehicle, one of a pedestrian and one of a building, wherein the multi-clutter may include at least one of a static ghost object, one of a dynamic ghost object and one of a ground clutter effect. In some embodiments, the controller can be configured to generate the ground truth data using one or more automatically labeled LiDAR occupancy maps to train at least one of the automated multi-clutter filters and DNN (Deep Neural Network) networks. In some embodiments, the controller can be configured to generate a smooth surface on the dense LiDAR voxel grid by applying a Poisson reconstruction technique to the point cloud data, resulting in a coherent mesh surface that enables the detection of at least one noise, a set of irregularities, and one or more outliers in the point cloud data. In some embodiments, the controller can be configured to convert the coherent mesh surface into a voxel representation by dividing the multidimensional space into one or more voxel grids and determining the voxel grid occupancy based on the binary condition. In another aspect, a controller-implemented method for automatic multi-cluster removal in multidimensional space is disclosed, including the removal of one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique. The method involves processing one or more radar signals to project radar PCD onto the multidimensional LiDAR PCD space. It also involves projecting the radar point cloud data onto a dense LiDAR voxel grid in multidimensional space and classifying the voxel status of the object. The voxel status refers to a binary condition that includes at least one of the values ​​"1" and "0", representing an occupied and an unoccupied state, respectively.The method involves the dynamic prediction and identification of one or more superimposed radar point cloud data sets on the dense LiDAR voxel grid. The method further includes filtering out the multi-clutter using the binary condition "0" and retaining relevant radar PCD data using the binary condition "1" to automatically generate ground-truth data for training a deep learning model. This model then initiates automatic multi-clutter removal based on the ground-truth data within the vehicle's radar system. Various tasks, features, aspects, and advantages of the invention will become clearer from the following detailed description of preferred embodiments in conjunction with the accompanying drawings, where similar numbers represent similar components. The accompanying drawings serve to enhance understanding of the present disclosure and are incorporated into this patent specification, forming part thereof. The drawings represent exemplary embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure. Fig. 1 illustrates an exemplary block diagram of a vehicle radar system for enabling automatic multi-clutter removal in a multidimensional space according to embodiments of the present disclosure. Fig. 2 illustrates an exemplary block diagram of a controller according to embodiments of the present disclosure. Fig. 3 illustrates an exemplary block diagram representing a training pipeline for implementing automatic multi-clutter removal according to embodiments of the present disclosure.Figure 4 illustrates a flowchart of an exemplary method for automatic multi-clutter removal in a vehicle radar system according to embodiments of the present disclosure. Figure 5 illustrates an exemplary computer system in which or with which embodiments of the system according to embodiments of the present disclosure can be implemented. A detailed description of embodiments of the disclosure, illustrated in the accompanying drawings, follows. The embodiments are described in sufficient detail to clearly convey the disclosure. However, the level of detail provided is not intended to limit the expected variations of embodiments; on the contrary, it is intended to cover all modifications, equivalents, and alternatives that fall within the scope of protection of the present disclosures as defined in the accompanying claims. The embodiments described here relate to the field of clutter removal. In particular, the present disclosure provides a system and a method for automatic multi-clutter removal in a vehicle radar system. Various embodiments of the present disclosure are explained in detail with reference to Figs. 1, 2, 3, 4 to 5. Fig. 1 illustrates an exemplary block diagram 100 of a vehicle radar system for enabling automatic multi-clutter removal in a multidimensional space according to embodiments of the present disclosure. With reference to Fig. 1, a vehicle radar system 102 (also referred to as System 102 or Radar System 102, used synonymously here) for automated multi-clutter removal in multidimensional space is shown. System 102 can be used in advanced driver-assistance systems (ADAS) and in autonomous driving. Vehicle radar system 102 can be a sensor that detects and tracks objects in an environment by transmitting radio waves and analyzing the reflected signal. For example, vehicle radar system 102 can be a long-range radar (LRR) or a medium-range radar (MRR). Vehicle radar system 102 can provide an automated, robust, and scalable approach applicable to all sensor types, radar manufacturers, and multiple clutter types.The vehicle radar system 102 can include at least one transmitter 104, at least one receiver 106 and a control unit 108, but is not limited to this. In some embodiments, the transmitter 104 can be configured to generate and transmit one or more radar signals. The object can be a vehicle, a pedestrian, a building, a tree, etc., but is not limited to such. The multi-clutter can consist of at least one static ghost object, one dynamic ghost object, a ground clutter effect, and the like. In one embodiment, a frequency-modulated continuous wave radar (FMCW radar) can be used as a vehicle radar 102 in motor vehicle applications to distinguish between at least one moving target (object) and a stationary clutter using Doppler effect principles. It is understood that the moving target can be referred to synonymously as the object throughout this disclosure.Furthermore, multidimensional space refers to three or more dimensions, which in the case of four dimensions (4D) and beyond may include other variables as additional dimensions. In some embodiments, the receiver 106 can be communicatively coupled to the transmitter 104 and can be configured to receive one or more radar signals from the transmitter 104. The receiver 106 can also be configured to filter out noise and irrelevant signals from the one or more radar signals using advanced processing techniques. The receiver 106 can be configured to amplify and convert radar echoes into a digital form for further processing. In some embodiments, the controller 108 can be communicatively coupled to the receiver 106 and the transmitter 104. The controller 108 can be configured to remove one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique. The controller 108 can be configured to process one or more radar signals, i.e., raw point cloud data, to project radar PCD into the multidimensional LiDAR PCD space to improve target identification and multi-clutter removal. The point cloud data can consist of at least one LiDAR point cloud data and one radar point cloud data. In some embodiments, the multidimensional LiDAR PCD can be incorporated into a dense, multidimensional map of the vehicle's surroundings, accurately representing the shapes and locations of objects and clutter. The LiDAR point cloud data can include, but are not limited to, cars, trees, pedestrians, buildings, and the like, along with details such as the outlines of branches or vehicle edges. The radar point cloud data can be used to detect nearby cars and moving pedestrians with basic spatial position and velocity information, though without the high level of detail found in LiDAR data. The radar point cloud data can primarily be used to identify moving objects and track their velocity. In some embodiments, the controller 108 can be configured to project the radar point cloud data onto a dense LiDAR voxel grid in multidimensional space and classify a voxel status of the object. The voxel status refers to a binary condition that includes at least one of "1" and one of "0". The controller 108 can be configured to dynamically predict and identify one or more superimposed radar point cloud data sets on the dense LiDAR voxel grid based on the binary condition. The controller 108 can be configured to filter out the multi-clutter with the binary condition "0" and retain relevant radar point cloud data with the binary condition "1" to automatically generate ground truth data to train a deep learning model that initiates automatic multi-clutter removal based on the ground truth data in the vehicle radar system 102. In some embodiments, the controller 108 can be configured to generate the ground truth data using one or more automatically labeled LiDAR occupancy maps for training the at least one of the automated clutter filter using DNN networks (DNNs, Deep Neural Networks). In some embodiments, the controller 108 can be configured to generate a smooth surface on the dense LiDAR voxel grid by applying a Poisson reconstruction technique to the point cloud data. This makes the dense LiDAR voxel grid appear as a coherent mesh surface, allowing the detection of at least one noise, a number of irregularities, and one or more outliers in the point cloud data. In some embodiments, the controller 108 can be configured to convert the coherent mesh surface into a voxel representation by dividing the multidimensional space into one or more voxel grids and determining the voxel grid occupancy based on the binary condition. Referring to Fig. 2, the system 102 / the controller 108 can contain the components shown in block diagram 200. The controller 108 can, for example, contain one or more processors 202. The one or more processors 202 can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuits, and / or any devices that manipulate data based on operating instructions. Among other functions, the one or more processors 202 can be configured to retrieve and execute computer-readable instructions stored in a memory 204 of the controller 108. The memory 204 can store one or more computer-readable instructions or routines that can be retrieved and executed to create or release data units via a network service.Memory 204 can contain any non-transient storage device, including, for example, volatile memory such as Random Access Memory (RAM) or non-volatile memory such as EPROM (Erasable Programmable Read-Only Memory), flash memory, and the like. In one embodiment, the system 102 may also include one or more interfaces 206. The one or more interfaces 206 may include a variety of interfaces, such as interfaces for data input and output devices, referred to as I / O devices (input / output), storage devices, and the like. The one or more interfaces 206 may facilitate communication between the sender 104, receiver 106, and controller 108. The one or more interfaces 206 may also provide a communication path for one or more components of the system 102. Examples of such components include, but are not limited to, one or more processing modules 208 and a database 210. In one embodiment, the one or more processing modules 208 can be implemented as a combination of hardware and programming (e.g., programmable instructions) to implement one or more functions of the one or more processing modules 208. In the examples described here, such combinations of hardware and programming can be implemented in various ways. For example, the programming for the one or more processing modules 208 can be processor-executable instructions stored on a non-transitory, machine-readable storage medium, and the hardware for the one or more processing modules 208 can include a processing resource (e.g., a controller) for executing such instructions. In other embodiments, the one or more processing modules 208 can be implemented by electronic circuits.The database 210 can contain data that is either stored or generated as a result of functions implemented by one of the components of the one or more processing modules 208. In some embodiments, the one or more processing modules 208 may include a classification module 212, an automatic ground truth module 214, a calculation module 216, and one or more other modules 218. The one or more other modules 218 may implement functionalities that complement applications / functions performed by the system 102. The classification module 212 can be configured to automatically categorize point cloud data into object and multi-clutter using an automated clutter filter. This categorization can be based on one or more signal properties, including but not limited to Doppler shifts, signal strength, signal range, and the like. The Occupancy Determination Module 214 can be configured to process point cloud data from LiDAR or radar and identify and map occupied spaces within an environment. The Occupancy Determination Module 214 can analyze incoming data to detect objects, obstacles, or free space, thus creating a real-time occupancy grid. By combining information from multiple sensors, the Occupancy Determination Module 214 can ensure the accurate detection of both moving and stationary objects. The automatic Ground Truth Module 216 can be configured to generate labeled data by automatically annotating sensor inputs such as images, LiDAR, or radar point clouds. This eliminates the need for manual annotations, reducing human error and improving scalability. The automatic Ground Truth Module 216 can utilize a deep learning model to classify and label objects in the environment. Automating the ground truth process accelerates the development of robust machine learning models. Figure 3 shows an example of a block diagram for a 300-degree training pipeline for implementing automatic multi-clutter removal. In deep learning, the 300-degree training pipeline refers to the end-to-end process that includes all the steps required to train a model, with LiDAR monitoring only during the training phase. The 300-degree pipeline starts with raw data and includes various stages until the model is fully trained and ready for evaluation or deployment. The raw data is processed concurrently while the model is being trained, and the newly trained model is deployed to the vehicle with improved capabilities.The training pipeline 300 includes the LiDAR point cloud 302, a thin 3D occupancy map 304, a dense 3D occupancy map 306, a spatially accurate representation of an environment 308, a radar point cloud 310, an automated multi-clutter filter 312 and a filtered radar point cloud 314. In some embodiments, the LiDAR point cloud 302 can be a collection of 3D data points generated by LiDAR systems, representing the surface and features of the environment with high spatial accuracy. The thin 3D occupancy map 304 can be configured to acquire data from a LiDAR sensor. LiDAR sensors provide multidimensional PCD, such as a 3D point cloud, by measuring the distance to surrounding objects using laser pulses. Furthermore, the thin 3D occupancy map 304 filters the point cloud to remove noise and outliers, which may involve removing points that are too far away or applying a statistical filter to remove points that do not fit the overall distribution of the data. If required, a coordinate transformation can be performed to transform the point cloud from the sensor's coordinate frame to a global coordinate frame. In some embodiments, the construction of the thin 3D occupancy map 304 also includes voxel occupancy classification and ground point removal. Voxel occupancy classification determines whether each voxel (a 3D equivalent of a pixel) in a raster is occupied, free, or unknown. In some embodiments, the dense 3D occupancy map 306 can be generated based on the application of a Poisson reconstruction technique to the thin 3D occupancy map 304. The Poisson reconstruction technique can be applied to mark dense voxels resulting from a mesh to obtain a dense 3D occupancy. Poisson surface reconstruction is a method by which 3D shapes are generated from a collection of points. In some implementations, the generation of dense voxels is based on the collection of cloud data points in 3D space. Each point typically has an associated normal vector that indicates the direction in which the surface points at that point. The Poisson reconstruction technique produces a smooth surface that best fits the given cloud data points, eliminating noise and irregularities within the data points. Furthermore, the reconstruction process results in a triangular mesh that can be denser than the original point cloud data. This increased density is achieved because the Poisson reconstruction algorithm fills gaps and creates a continuous surface, effectively increasing the number of vertices and faces in the mesh. Once the triangular mesh is obtained from the Poisson reconstruction, it can be converted into a voxel representation.In voxelization, the 3D space is divided into a grid of cubes (voxels) and it is determined which voxels are occupied by the reconstructed surface. In some embodiments, the radar and LiDAR sensors are calibrated together to project the radar point cloud data 310 onto a dense LiDAR voxel grid 306. Dense 3D voxel grids 306 represent a binary state of the object, present or absent, in each voxel, represented by the voxel state x, y, z; x, y, z represent 3D points of the voxel grid, and voxel state represents the binary condition of the object, i.e., present or absent. Binary state 1 indicates that the object is present, and 0 represents the absence of the object. The LiDAR ground surface is removed during the preprocessing phase, and thus the ground surface is represented with state 0. Overlaid radar point cloud data in the binary voxel grid filters out the radar points that are in binary state 0 and retains points in binary state 1. The voxel grid acts as a binary mask. In some embodiments, a spatially accurate representation of an environment 308 is a dense 3D map that provides a comprehensive, detailed representation of an environment, supporting various analyses and applications across multiple fields. In some embodiments, an automated multi-clutter filter 312 involves training a 3D U-mesh model using a deep learning architecture for volumetric data to learn the characteristics of noisy radar point cloud data (PC, data) affected by clutter. The automated multi-clutter filter 312 is trained on labeled datasets where the radar point clouds are identified using the automatically generated labels of the LiDAR-based multi-clutter removal procedure to identify relevant features within the noise. During training, the network learns to segment and classify the clutter, thereby improving its ability to distinguish between useful signals and unwanted noise. The trained 3D U-mesh model provides valuable insights for effective multi-clutter removal.Furthermore, the filtered radar point cloud 314 includes the processing of radar data from which unwanted noise and multi-clutter have been removed. In some embodiments, the Training Pipeline 300 can effectively process multi-clutter distance from radar point cloud data across different sensor types and manufacturers. System 102 dynamically predicts and filters overlapping radar point clouds. Furthermore, using a binary classification system, System 102 assigns a condition of "0" to clutter points, indicating their distance, while retaining relevant point cloud data with a condition of "1". The Training Pipeline 300 utilizes advanced machine learning techniques to learn from various types of clutter, ensuring adaptability across different environments and scenarios. Furthermore, a combination of statistical analysis and neural networks enhances predictive accuracy. Thanks to its scalability, the System 102 can be seamlessly integrated with various radar technologies, making it suitable for widespread deployment in radar systems. By automating the multi-clutter removal process, the System 102 significantly improves object detection and tracking capabilities, contributing to safer and more efficient navigation in autonomous vehicles. Referring to Fig. 4, a method 400 for automatic multi-clutter removal in a vehicle radar system 102 can comprise a plurality of blocks 402 to 410. The method 400 can be implemented by the system 102 or its controller 108. In block 402, the method 400 involves removing one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique. In Block 404, the procedure 400 involves processing one or more radar signals to project radar PCD onto the multidimensional LiDAR PCD space. In Block 406, the procedure 400 involves projecting the radar point cloud data onto a dense lidar voxel grid in the multidimensional space and classifying a voxel status of the object, where the voxel status refers to a binary condition that includes at least one of "1" and "0" and represents an occupied status or an unoccupied status. Method 400, in block 408, involves the dynamic prediction and identification of one or more superimposed radar point cloud datasets on the dense LiDAR voxel grid. Method 400, in block 410, further involves filtering out the multi-clutter with binary condition "0" and retaining relevant radar PCD data with binary condition "1" to automatically generate ground-truth data for training a deep learning model. This model then initiates automatic multi-clutter removal based on the ground-truth data in the vehicle radar system 102. System 102 and methods 400 can be implemented on a computer system. With reference to Fig. 5, the block diagram represents a computer system 500 comprising an external storage device 510, a bus 520, main memory 530, read-only memory 540, mass storage device 550, a communication port 560, and a processor 570. It will be clear to those skilled in the art that the computer system 500 can include more than one processor 570 and communication ports 560. The processor 570 can include various modules associated with embodiments of the present disclosure.The 560 communication port can be any of the recommended standard 232 ports for use with a modem-based dial-up connection, a 10 / 100 Ethernet port, a Gigabit or 10 Gigabit port using copper or fiber optic cabling, a serial port, a parallel port, or any other existing or future port. The 560 communication port can be selected according to the network, such as a LAN (Local Area Network), a WAN (Wide Area Network), or any network to which the 500 computer system is connected. In one embodiment, the memory 530 can be RAM or other dynamic memory device generally known in the field. The read-only memory (ROM) 540 can be one or more arbitrary static memory devices, such as a PROM chip (PROM, Programmable Read-Only Memory) for storing static information. The mass storage 550 can be any current or future mass storage solution that can be used to store information and / or instructions. Examples of mass storage solutions include, but are not limited to, Parallel Advanced Technology Attachment (PATA) or Serial Advanced Technology Attachment (SATA) hard disks or solid-state drives (internal or external, such as those with USB and / or FireWire interfaces), one or more optical disks, RAID storage (RAID, Redundant Array of Independent Disks), such as an array of disks (e.g., SATA arrays). In one embodiment, the bus 520 communicatively couples the one or more processors 570 with the other main memory, storage, and communication blocks. The bus 520 can be, for example, a PCI / PCI Extended (PCI-X) bus (PCI, Peripheral Component Interconnect), a SCSI interface (SCSI, Small Computer System Interface), USB, or similar, to connect expansion cards, drives, and other subsystems, as well as other buses, such as a front-side bus (FSB) that connects the processor 570 to the computer system 500. In another embodiment, operator and management interfaces, such as a display, a keyboard, and a cursor control device, can also be coupled to the bus 520 to support direct operator interaction with the computer system 500. Other operator and management interfaces can be provided via network connections connected through the communication port 560. In some embodiments, the external storage device 510 can be any type of external hard disk, floppy disk drive, compact disc read-only storage (CD-ROM), compact disc re-writable (CD-RW), or digital video disc read-only storage (DVD-ROM). The components described above are merely intended to illustrate various possibilities. The aforementioned computer system 500 is not intended to limit the scope of this disclosure in any way. While the foregoing describes various embodiments of the present disclosure, other and further embodiments of the present disclosure may be conceived without departing from its basic scope. The scope of the present disclosure is determined by the following claims. The present disclosure is not limited to the described embodiments, versions, or examples, which are included to enable the person skilled in the art to implement and use the present disclosure in conjunction with information and knowledge available to the person skilled in the art. The present disclosure provides a system and a method for automatic multi-clutter removal in a vehicle radar system. The present disclosure provides an automated, robust and scalable approach applicable to all sensor types, radar manufacturers and multiple clutter types. This disclosure ensures that multi-clutter is minimized, thereby improving the radar system's performance in detecting and tracking relevant targets. This disclosure provides an automatically labeled ground-truth pipeline during the training phase, eliminating the need for tedious and error-prone manual threshold adjustments. QUOTES INCLUDED IN THE DESCRIPTION This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature CN 116386001A

[0003] CN 114612329A

[0004] US 20240159871A1

[0005]

Claims

Vehicle radar system (102) for enabling automatic multi-clutter removal in a multidimensional space, the vehicle radar system (102) comprising: at least one transmitter (104) configured to generate and transmit one or more radar signals; at least one receiver (106) communicatively coupled to the at least one transmitter (104) and configured to receive the one or more radar signals; and a controller (108) communicatively coupled to the at least one receiver (104) and the at least one transmitter (106) and configured to: remove one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique; and process the one or more radar signals to convert radar PCD into the multidimensional LiDAR PCD space. project;Projecting the radar PCD onto a dense LiDAR voxel grid in multidimensional space and classifying a voxel status of an object, wherein the voxel status refers to a binary condition comprising at least one of "1" and "0", representing an occupied status and an unoccupied status, respectively; dynamically predicting and identifying one or more superimposed radar PCDs on the dense LiDAR voxel grid; and filtering out the multi-clutter with the binary condition "0" and retaining relevant radar PCD data with the binary condition "1" to automatically generate ground-truth data to train a deep learning model that initiates automatic multi-clutter removal based on the ground-truth data in the vehicle radar system (102). Vehicle radar system (102) according to claim 1, wherein the PCD comprises at least one of: LiDAR point cloud data and radar point cloud data. Vehicle radar system (102) according to claim 1, wherein the controller (108) is configured to automatically categorize the radar PCD into the object and the multi-clutter based on a predefined area. Vehicle radar system (102) according to claim 1, wherein the object comprises at least one of: a vehicle, a pedestrian and a building, and wherein the multi-clutter comprises at least one of: a static ghost object, a dynamic ghost object and a ground clutter. Vehicle radar system (102) according to claim 1, wherein the controller (108) is configured to: generate ground truth data using one or more automatically labeled LiDAR occupancy maps for training at least one of: an automated clutter filter (312) using deep neural networks (DNNs). Vehicle radar system (102) according to claim 1, wherein the controller (108) is configured to: generate a smooth surface on the dense LiDAR voxel grid by applying a Poisson reconstruction technique to the PCD, resulting in a coherent mesh surface that enables the detection of at least one noise, a set of irregularities and one or more outliers in the PCD. Vehicle radar system (102) according to claim 6, wherein the controller (108) is configured to: convert the coherent mesh surface into a voxel representation by dividing the multidimensional space into one or more voxel grids; and determine an allocation of the voxel grid based on the binary condition. Method (300) for automatic multi-clutter removal in a multidimensional space, the method comprising: removing one or more ground points from a multidimensional LiDAR (Light Detection and Ranging) space containing point cloud data (PCD) using a LiDAR ground point removal technique by a controller (108); processing one or more radar signals by the controller (108) to project radar PCD onto the multidimensional LiDAR PCD space; projecting the radar PCD by the controller (108) onto a dense LiDAR voxel grid in the multidimensional space and classifying a voxel status of an object, wherein the voxel status refers to a binary condition comprising at least one of '1' and '0' and an occupied status orrepresents an unoccupied state; dynamically predict and identify one or more superimposed radar point cloud data in the dense LiDAR voxel grid by the controller (108); and filter out the multi-clutter with the binary condition "0" and retain relevant radar PCD data with the binary condition "1" by the controller (108) to automatically generate ground truth data to train a deep learning model that initiates automatic multi-clutter removal based on the ground truth data in a vehicle radar system (102).

Images