Point cloud generation method and system and vehicle-mounted terminal

Abstract

The invention provides a point cloud generation method and system and a vehicle-mounted terminal. The point cloud generation method comprises the following steps: constructing a sunlight noise amplitude image in real time according to a radar acquisition image; then extracting a target contour based on continuous multi-frame sunlight noise amplitude images and original flight time; and further obtaining a noise supplementary point cloud based on the target contour, and generating a target point cloud through confidence coefficient weighted fusion based on the original point cloud and the noise supplementary point cloud. According to the invention, the hardware architecture of the existing pure solid-state laser radar can be well considered, and the integrity of the generated point cloud can be greatly improved without depending on hardware improvement; moreover, through the dynamic statistics of the sunlight noise, the method can adapt to various dynamic illumination changes, thereby reducing the false detection rate, and further guaranteeing the integrity of the generated point cloud data.

Description

Technical Field

[0001] This application relates to the field of radar technology, and in particular to a point cloud generation method, system and vehicle terminal. Background Technology

[0002] Existing solid-state lidar suffers from the following problems in strong sunlight interference scenarios: the infrared band in sunlight is similar to the wavelength of the laser emitted by the lidar, easily generating noise signals at the receiver and drowning out the true target echo, especially at long distances or with low reflectivity objects. Current technologies mainly suppress noise through narrowband filtering, time-domain thresholding, or increasing laser power, but these methods have limited effectiveness under extreme lighting conditions and rely on hardware improvements (such as increasing filter precision or laser intensity), leading to increased costs and power consumption. When the noise intensity exceeds the signal strength, traditional algorithms cannot distinguish between noise and the true echo, resulting in sparse point clouds or data loss, affecting the reliability of ranging. Summary of the Invention

[0003] This application provides a point cloud generation method, system, and vehicle-mounted terminal to solve the above-mentioned technical problems.

[0004] Specifically, this application provides a point cloud generation method, including the following steps: constructing a sunlight noise amplitude image in real time based on radar-acquired images; extracting a target contour based on multiple consecutive frames of sunlight noise amplitude images and the original flight time; and obtaining a noise-supplemented point cloud based on the target contour, and generating a target point cloud by confidence-weighted fusion of the original point cloud and the noise-supplemented point cloud.

[0005] The above technical solution does not rely on hardware upgrades and greatly improves the completeness of point cloud generation in direct sunlight scenarios, especially for the detection of distant low-reflectivity targets; and through dynamic noise statistics, it can adapt to various dynamic sunlight changes, reduce the false detection rate, and further improve the completeness of point cloud.

[0006] Furthermore, the construction of the sunlight noise amplitude image includes: acquiring radar-acquired images and statistically analyzing the noise level of each pixel in the radar-acquired images in real time, so as to construct a sunlight noise amplitude image based on the noise level.

[0007] In the above technical solution, a sunlight noise amplitude image is constructed by statistically analyzing the noise level of each pixel in the radar-acquired image in real time. This accurately reflects the intensity differences of sunlight interference in different areas, allowing subsequent processing to take corresponding measures for areas with different noise intensities, thereby improving the targeting and effectiveness of anti-interference. At the same time, the real-time statistical method can adapt to various sunlight changes and promptly capture the dynamic changes in noise, providing a foundation for accurate extraction of target contours and generation of point clouds. Furthermore, this step does not require hardware upgrades, avoiding the problems of increased costs and power consumption caused by adding hardware.

[0008] Furthermore, the extraction of the target contour includes: performing feature extraction based on multiple consecutive frames of sunlight noise amplitude images, and performing feature matching based on the feature extraction results; analyzing the target's motion trajectory based on the feature matching results, while using geometric constraints to identify and remove noise points to obtain effective features; and extracting an initial contour based on the effective features.

[0009] In the above technical solution, feature extraction and matching based on continuous multi-frame images of sunlight noise amplitude can utilize the continuity and correlation of the target in multiple frames to more accurately identify target features. Compared with single-frame images, multi-frame analysis can reduce the influence of random noise and improve the reliability of feature extraction. Feature matching can analyze the target's motion trajectory, which is crucial for identifying and tracking dynamic targets. In strong sunlight interference scenarios, accurate motion trajectory information helps distinguish between real targets and noise, improving the accuracy of target recognition. Furthermore, using geometric constraints to identify and remove noise points can effectively remove the interference of sunlight noise on target contour extraction. After removing noise, extracting the target contour based on the remaining effective features can greatly improve the accuracy of contour extraction, especially in the case of distant or low-reflectivity objects, which can more clearly delineate the target boundary and provide an accurate foundation for subsequent point cloud generation.

[0010] Furthermore, the extraction of the target contour also includes: calculating the depth value corresponding to each pixel based on the initial contour, and fusing the depth value with the current radar's original flight time, so as to optimize the initial contour by combining the fusion result to obtain the target contour.

[0011] In the above technical solution, by fusing depth values ​​with original time-of-flight (ToF), the radar's depth information can be fully utilized to optimize the target contour. Depth information can provide more information about the target's three-dimensional structure, making the optimized target contour more accurate and complete. This fusion method can adapt to various complex lighting and target conditions. Even under direct sunlight or when the target has low reflectivity, the initial contour can be corrected using depth information, improving the accuracy and reliability of the target contour.

[0012] Furthermore, the acquisition of noise-supplemented point cloud includes: marking the target contour based on the difference in sunlight echo intensity to acquire noise-supplemented point cloud.

[0013] The above technical solution can supplement the data missing in the original point cloud due to noise interference. In strong sunlight interference scenarios, the original point cloud may be sparse or data may be lost. Noise-supplemented point cloud can effectively make up for these deficiencies and improve the integrity of the point cloud. This method can be marked according to the differences in sunlight echo intensity of different targets and is applicable to various types of targets.

[0014] Furthermore, after acquiring the noise-supplemented point cloud, the method further includes: performing spatiotemporal alignment between the original point cloud and the noise-supplemented point cloud.

[0015] In the above technical solution, by aligning time and space, it can be ensured that the original point cloud and the noise-supplemented point cloud have a unified time reference, and at the same time, they overlap as much as possible in space to ensure the accuracy of subsequent point cloud generation.

[0016] Furthermore, the generation of the target point cloud includes: performing confidence assessments on the spatiotemporally aligned original point cloud and the noise-supplemented point cloud respectively, and performing confidence-weighted fusion based on the confidence assessment results to obtain the fused point cloud as the target point cloud.

[0017] In the above technical solution, by fusing the original point cloud and the noise-supplemented point cloud with confidence weighting, the advantages of the two types of point clouds can be comprehensively utilized. By weighting according to the confidence of different point clouds, the data with high confidence plays a greater role in the fusion process, thereby improving the accuracy and reliability of the target point cloud.

[0018] Based on the same concept, this application also provides a point cloud generation system, including: a construction module for constructing a sunlight noise amplitude image in real time based on radar-acquired images; an extraction module for extracting a target contour based on multiple consecutive frames of sunlight noise amplitude images and the original flight time; and a generation module for obtaining a noise-supplemented point cloud based on the target contour, and generating a target point cloud by weighted fusion of the original point cloud and the noise-supplemented point cloud with confidence.

[0019] The above technical solution is compatible with existing pure solid-state LiDAR hardware architecture, requiring no additional filtering or modules, and can greatly improve the completeness of point cloud generation in direct sunlight scenarios; while through dynamic noise statistics, it can adapt to various dynamic sunlight changes, reduce false detection rate, and further improve the completeness and accuracy of generated point clouds.

[0020] The system also includes: Furthermore, the point cloud generation system also includes an alignment module; the alignment module is used to perform spatiotemporal alignment of the original point cloud and the noise-supplemented point cloud.

[0021] In the above technical solution, by performing time alignment and spatial alignment on the two types of point cloud data respectively, the consistency between the two in time and space can be ensured, thus guaranteeing the accurate generation of the target point cloud in the future.

[0022] Based on the same concept, this application also provides a vehicle-mounted terminal, which includes a processor and a memory. The memory stores at least one instruction, at least one program, code set, or instruction set. The at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the point cloud generation method.

[0023] Compared with the prior art, the beneficial effects of this application are as follows: This application can better take into account the existing hardware architecture of pure solid-state LiDAR without relying on hardware improvements to greatly improve the integrity of the generated point cloud; and through dynamic statistics of sunlight noise, it can adapt to various dynamic lighting changes to reduce the false detection rate and further ensure the integrity of the generated point cloud data. Attached Figure Description

[0024] Figure 1 This is a flowchart of the point cloud generation method described in this application.

[0025] Figure 2 This is a framework diagram of the point cloud generation system described in this application. Detailed Implementation

[0026] The following describes in further detail a point cloud generation method, system, and vehicle terminal of this application with reference to specific embodiments and accompanying drawings.

[0027] Please see Figure 1 This application provides a point cloud generation method, including the following steps S100-S300.

[0028] Step S100: Construct a real-time image of sunlight noise amplitude based on radar-acquired images.

[0029] Step S200: Extract the target contour based on the sunlight noise amplitude image of multiple consecutive frames and the original time of flight.

[0030] Step S300: Obtain noise-supplemented point cloud based on the target contour, and generate target point cloud by confidence-weighted fusion of the original point cloud and the noise-supplemented point cloud.

[0031] In some feasible implementations, images obtained from radar scanning are acquired, and the noise level of each pixel in the image is statistically analyzed in real time within the histogram. Then, an amplitude image based on sunlight noise (i.e., the sunlight noise amplitude image) is constructed based on the noise level to reflect the intensity differences of sunlight interference in different areas. Next, feature points are extracted and matched from multiple consecutive frames of amplitude images. Combined with the target motion trajectory and geometric constraints, irrelevant components in the sunlight noise are removed, and a stable target contour is extracted. Furthermore, based on the matching results, pixel-level depth information is calculated using a camera-like triangulation principle, and optimized by fusing raw time-of-flight data from the lidar. Contour accuracy; further, based on the difference in echo intensity of sunlight on the surface of objects with different reflectivity, statistically significant areas in the target contour are marked as potential targets (i.e., the noise-supplemented point cloud), supplementing the point cloud data missing due to insufficient signal-to-noise ratio; further, the traditional ToF point cloud (i.e., the original point cloud) and the noise-supplemented point cloud are time-aligned and spatially aligned, so that confidence assessment (such as distance, reflection intensity, noise characteristics) is performed based on the spatiotemporally aligned point cloud, and weighted fusion is performed based on the confidence assessment results to obtain the final fused point cloud as the target point cloud, ensuring the accuracy and completeness of data in strong sunlight interference scenarios.

[0032] Next, steps S100-S300 above will be described in detail.

[0033] The step S100 of constructing the sunlight noise amplitude image includes: acquiring a radar-acquired image and calculating the noise level of each pixel in the radar-acquired image in real time, so as to construct a sunlight noise amplitude image based on the noise level.

[0034] In some feasible implementations, the noise level of each pixel in the histogram of the radar scan image (i.e., the radar acquired image) is statistically analyzed in real time. Based on the statistical results of the noise level of each pixel, a sunlight noise amplitude image is constructed. In this sunlight noise amplitude image, the value of each pixel corresponds to the intensity of sunlight interference at that location. In this sunlight noise amplitude image, different grayscale values ​​or color values ​​represent the degree of sunlight noise interference in that area.

[0035] It should be noted that when a pure solid-state LiDAR is working, it scans and detects the surrounding environment. The data it acquires can be presented and processed in the form of images, which are composed of pixels. In addition, since the sunlight interference experienced by pixels at different locations may vary, by performing noise statistics on each pixel individually, it is possible to accurately understand the noise situation at each location. For example, in some areas, due to direct sunlight, the noise level reflected by the pixels in that area may be relatively high, while in areas where sunlight is blocked, the noise level of the pixels is relatively low.

[0036] In the above technical solution, a sunlight noise amplitude image is constructed by statistically analyzing the noise level of each pixel in the radar-acquired image in real time. This accurately reflects the intensity differences of sunlight interference in different areas, allowing subsequent processing to take corresponding measures for areas with different noise intensities, thereby improving the targeting and effectiveness of anti-interference. At the same time, the real-time statistical method can adapt to various sunlight changes and promptly capture the dynamic changes in noise, providing a foundation for accurate extraction of target contours and generation of point clouds. Furthermore, this step does not require hardware upgrades, avoiding the problems of increased costs and power consumption caused by adding hardware.

[0037] Furthermore, the extraction of the target contour in step S200 includes: performing feature extraction based on multiple consecutive frames of sunlight noise amplitude images, and performing feature matching based on the feature extraction results; analyzing the target's motion trajectory based on the feature matching results, while using geometric constraints to identify and remove noise points to obtain effective features; and extracting an initial contour based on the effective features.

[0038] In some feasible implementations, appropriate feature point extraction algorithms, such as SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Features), are applied to extract feature points from each frame of the continuous multi-frame sunlight noise amplitude image. After extracting the feature points of each frame, feature matching algorithms, such as FLANN (Fast Library for Approximate Nearest Neighbors), are used to match feature points between adjacent frames or specified frames. The matching process is based on the descriptors of the feature points to find pairs of points with similar features. Furthermore, the motion trajectory of the target is analyzed based on the results of the feature point matching. At the same time, geometric constraints (such as the shape, size, and speed of the target) are used. For example, if the motion trajectory of a certain feature point does not conform to the normal motion trajectory of the target, it is judged as a noise point and removed.

[0039] Furthermore, after removing noise points, based on the remaining valid feature points, a contour extraction algorithm is used, such as an edge detection-based method (e.g., Canny edge detection) or a region growing-based method, to extract the initial contour.

[0040] In the above technical solution, feature extraction and matching based on continuous multi-frame images of sunlight noise amplitude can utilize the continuity and correlation of the target in multiple frames to more accurately identify target features. Compared with single-frame images, multi-frame analysis can reduce the influence of random noise and improve the reliability of feature extraction. Feature matching can analyze the target's motion trajectory, which is crucial for identifying and tracking dynamic targets. In strong sunlight interference scenarios, accurate motion trajectory information helps distinguish between real targets and noise, improving the accuracy of target recognition. Furthermore, using geometric constraints to identify and remove noise points can effectively remove the interference of sunlight noise on target contour extraction. After removing noise, extracting the target contour based on the remaining effective features can greatly improve the accuracy of contour extraction, especially in the case of distant or low-reflectivity objects, which can more clearly delineate the target boundary and provide an accurate foundation for subsequent point cloud generation.

[0041] Furthermore, the extraction of the target contour in step S200 also includes: calculating the depth value corresponding to each pixel based on the initial contour, and fusing the depth value with the original flight time of the current radar, so as to optimize the initial contour by combining the fusion result to obtain the target contour.

[0042] In some feasible implementations, pixel-level depth information is calculated based on the initial contour using a camera-like triangulation principle. This principle is similar to a binocular vision system, which calculates the depth value corresponding to each pixel by matching known camera parameters and feature points. The calculated pixel-level depth information is then fused with the radar's raw time-of-flight (ToF, which calculates the distance to an object by measuring the time it takes for a light pulse to travel from emission to reflection) data. Weighted averaging, Kalman filtering, and other methods can be used to combine the advantages of both types of data to optimize the accuracy of the initial contour, thereby obtaining the target contour.

[0043] In the above technical solution, by fusing depth values ​​with original time-of-flight (ToF), the radar's depth information can be fully utilized to optimize the target contour. Depth information can provide more information about the target's three-dimensional structure, making the optimized target contour more accurate and complete. This fusion method can adapt to various complex lighting and target conditions. Even under direct sunlight or when the target has low reflectivity, the initial contour can be corrected using depth information, improving the accuracy and reliability of the target contour.

[0044] Furthermore, the step S300 of obtaining the noise supplement point cloud includes: marking the target contour based on the difference in sunlight echo intensity to obtain the noise supplement point cloud.

[0045] In some feasible implementations, based on the difference in the echo intensity of sunlight on the surfaces of objects with different reflectivity (such as high-reflectivity metal and low-reflectivity asphalt), regions with statistical significance in the sunlight noise amplitude image (i.e., in the target contour) are marked as potential targets as noise-supplemented point clouds to supplement the point cloud data missing due to insufficient signal-to-noise ratio.

[0046] For example, an echo intensity threshold can be set, and areas with echo intensity above the threshold can be marked as high reflectivity areas, while areas with echo intensity below the threshold can be marked as low reflectivity areas. These statistically significant areas are then marked as potential targets. These potential targets may be missing parts in the original point cloud due to insufficient signal-to-noise ratio, but they can be identified in the sunlight noise amplitude image through echo intensity differences. Data of the areas marked as potential targets is extracted from the sunlight noise amplitude image. This data contains information such as the target's location and echo intensity. The extracted data is then converted into a point cloud format, where each data point corresponds to a coordinate in three-dimensional space, thus obtaining a noise-supplemented point cloud.

[0047] The above technical solution can supplement the data missing in the original point cloud due to noise interference. In strong sunlight interference scenarios, the original point cloud may be sparse or data may be lost. Noise-supplemented point cloud can effectively make up for these deficiencies and improve the integrity of the point cloud. This method can be marked according to the differences in sunlight echo intensity of different targets and is applicable to various types of targets.

[0048] Furthermore, after obtaining the noise-supplemented point cloud in step S300, the method further includes: performing spatiotemporal alignment between the original point cloud and the noise-supplemented point cloud.

[0049] In some feasible implementations, the original point cloud is the traditional ToF point cloud. If the acquisition times of the original point cloud and the noise-supplemented point cloud are inconsistent, interpolation or synchronization operations need to be performed according to their respective timestamps to make them correspond in time. For example, if the acquisition frequency of the original point cloud is 10Hz and the acquisition frequency of the noise-supplemented point cloud is 20Hz, the noise-supplemented point cloud needs to be downsampled or interpolated to align it with the original point cloud in time.

[0050] Furthermore, feature descriptors such as Fast Persistent Feature Histograms (FPFH) are used to extract features from the original point cloud and the noise-supplemented point cloud. Then, algorithms such as Sample Consensus Initial Alignment (SAC-IA) are used for initial alignment to obtain a rough transformation matrix. Based on the transformation matrix of the initial alignment, the Iterative Closest Point (ICP) algorithm is used for precise alignment, continuously optimizing the transformation matrix to make the two point clouds overlap as much as possible in space.

[0051] In the above technical solution, by aligning time and space, it can be ensured that the original point cloud and the noise-supplemented point cloud have a unified time reference, and at the same time, they overlap as much as possible in space to ensure the accuracy of subsequent point cloud generation.

[0052] Furthermore, the generation of the target point cloud in step S300 includes: performing confidence assessments on the spatiotemporally aligned original point cloud and the noise-supplemented point cloud respectively, and performing confidence-weighted fusion based on the confidence assessment results to obtain the fused point cloud as the target point cloud.

[0053] In some feasible implementations, based on the measurement distance of the original point cloud, points that are closer generally have higher measurement accuracy and higher distance confidence, and vice versa. Points with higher reflection intensity indicate that the target has a stronger ability to reflect laser light, and the measurement results are relatively reliable with higher reflection intensity confidence, and vice versa. Further analysis of the noise characteristics of the noise-supplemented point cloud reveals that regions with lower noise levels have higher noise level confidence, and vice versa.

[0054] Based on the confidence assessment above, the two spatiotemporally aligned point clouds are weighted and fused to obtain a fused point cloud.

[0055] In the above technical solution, by fusing the original point cloud and the noise-supplemented point cloud with confidence weighting, the advantages of the two types of point clouds can be comprehensively utilized. By weighting according to the confidence of different point clouds, the data with high confidence plays a greater role in the fusion process, thereby improving the accuracy and reliability of the target point cloud.

[0056] In summary, the point cloud generation method described in this application does not require hardware upgrades, greatly improves the completeness of point cloud generation in direct sunlight scenarios, especially for the detection of distant, low-reflectivity targets; and through dynamic noise statistics, it can adapt to various dynamic sunlight changes, reduce the false detection rate, and further improve the completeness of point cloud.

[0057] Based on the same concept, please refer to Figure 2This application also provides a point cloud generation system, comprising: a construction module for constructing a sunlight noise amplitude image in real time based on radar-acquired images; an extraction module for extracting a target contour based on multiple consecutive frames of sunlight noise amplitude images and the original flight time; and a generation module for obtaining a noise-supplemented point cloud based on the target contour, and generating a target point cloud by weighted fusion of the original point cloud and the noise-supplemented point cloud with confidence.

[0058] Furthermore, the point cloud generation system also includes an alignment module; the alignment module is used to perform spatiotemporal alignment of the original point cloud and the noise-supplemented point cloud.

[0059] In one feasible implementation, firstly, environmental images are acquired using radar scanning. Then, for each pixel in the image, its noise level on the histogram is calculated in real time. Based on these noise levels, a sunlight noise amplitude image is constructed, which clearly shows the intensity differences in sunlight interference in different areas. Further, feature point extraction and matching operations are performed on multiple consecutive frames of the sunlight noise amplitude image. During this process, components irrelevant to the target are removed from the sunlight noise by combining the target's motion trajectory and geometric constraints. Based on the matching results, pixel-level depth information is calculated using a principle similar to camera triangulation, and this information is fused with the raw time-of-flight data from the lidar to optimize the target contour extraction accuracy and obtain a stable target contour.

[0060] Furthermore, since sunlight produces different echo intensities on surfaces with varying reflectivity, statistically significant regions are identified within the target outline and marked as potential targets. The point cloud data corresponding to these potential targets constitutes the noise-supplemented point cloud, which can supplement data missing from the original point cloud due to insufficient signal-to-noise ratio. Then, the original point cloud obtained from conventional Time-of-Flight (ToF) measurements and the noise-supplemented point cloud are aligned temporally and spatially. After alignment, the confidence levels of the two types of point clouds are evaluated from multiple aspects, including distance, reflection intensity, and noise characteristics. Based on the confidence level evaluation results, the spatiotemporally aligned original point cloud and noise-supplemented point cloud are weighted and fused to obtain the fused point cloud, i.e., the target point cloud, thus ensuring the accuracy and completeness of the data under strong sunlight interference scenarios.

[0061] The above technical solution is compatible with existing pure solid-state LiDAR hardware architecture, requiring no additional filtering or modules, and can greatly improve the completeness of point cloud generation in direct sunlight scenarios; while through dynamic noise statistics, it can adapt to various dynamic sunlight changes, reduce false detection rate, and further improve the completeness and accuracy of generated point clouds.

[0062] Based on the same concept, this application also provides a vehicle-mounted terminal, which includes a processor and a memory. The memory stores at least one instruction, at least one program, code set, or instruction set. The at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the point cloud generation method.

[0063] In some embodiments, the memory and processor are interconnected via a bus; the processor may be one or more CPUs. If the processor is a single CPU, it may be a single-core CPU or a multi-core CPU. The processor is used to control various functional modules of the vehicle terminal and process signals. The memory includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), and CD-ROM (Compact Disc Read-Only Memory). This memory is used to store computer programs, operating systems, various applications, and data, such as storing computer programs used to implement the point cloud generation method.

[0064] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0065] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0066] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0067] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules according to the embodiments of this application. This application can also be implemented as an apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0069] Although the description of this application has been made in conjunction with the specific embodiments described above, it will be apparent to those skilled in the art that many substitutions, modifications, and variations can be made based on the foregoing. Therefore, all such substitutions, modifications, and variations are included within the spirit and scope of the appended claims.

Claims

1. A point cloud generation method, characterized in that, Includes the following steps: A real-time image of sunlight noise amplitude is constructed based on radar-acquired images; Target contours are extracted based on continuous multi-frame images of sunlight noise amplitude and the original time of flight. Furthermore, a noise-supplemented point cloud is obtained based on the target contour, and a target point cloud is generated by confidence-weighted fusion of the original point cloud and the noise-supplemented point cloud.

2. The point cloud generation method according to claim 1, characterized in that, The construction of the sunlight noise amplitude image includes: The radar-acquired images are acquired, and the noise level of each pixel in the radar-acquired images is statistically analyzed in real time to construct a sunlight noise amplitude image based on the noise level.

3. The point cloud generation method according to claim 2, characterized in that, The extraction of the target contour includes: Feature extraction is performed based on multiple consecutive frames of sunlight noise amplitude images, and feature matching is performed based on the feature extraction results; The motion trajectory of the target is analyzed based on the feature matching results, and noise points are identified and removed using geometric constraints to obtain effective features. And, an initial contour is extracted based on the effective features.

4. The point cloud generation method according to claim 3, characterized in that, The extraction of the target contour also includes: The depth value corresponding to each pixel is calculated based on the initial contour, and the depth value is fused with the original flight time of the current radar to optimize the initial contour by combining the fusion result, thereby obtaining the target contour.

5. The point cloud generation method according to claim 4, characterized in that, The acquisition of the noise-supplemented point cloud includes: The target contour is marked based on the difference in sunlight echo intensity to obtain a noise-supplemented point cloud.

6. The point cloud generation method according to claim 5, characterized in that, After acquiring the noise-supplemented point cloud, the following steps are also included: Spatiotemporal alignment is performed on the original point cloud and the noise-supplemented point cloud.

7. The point cloud generation method according to claim 6, characterized in that, The generation of the target point cloud includes: The confidence levels of the original point cloud and the noise-supplemented point cloud after spatiotemporal alignment are evaluated respectively. Based on the confidence evaluation results, a confidence-weighted fusion is performed to obtain the fused point cloud as the target point cloud.

8. A system employing the point cloud generation method as described in any one of claims 1-7, characterized in that, include: The module is used to construct a real-time image of sunlight noise amplitude based on radar-acquired images; An extraction module is used to extract target contours based on consecutive frames of sunlight noise amplitude images and the original time of flight; The generation module is used to obtain a noise-supplemented point cloud based on the target contour, and generate a target point cloud by weighted fusion of the original point cloud and the noise-supplemented point cloud through confidence.

9. The system according to claim 8, characterized in that, The system also includes: The alignment module is used to perform spatiotemporal alignment between the original point cloud and the noise-supplemented point cloud.

10. A vehicle-mounted terminal, characterized in that, The vehicle-mounted terminal includes a processor and a memory. The memory stores at least one instruction, at least one program, a code set, or an instruction set. The at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the point cloud generation method as described in any one of claims 1-7.

Images