Radar data storage method and device

Abstract

The invention relates to the technical field of radars, and particularly discloses a radar data storage method and device, and the method comprises the steps: obtaining a historical laser radar data unit of each channel in multiple channels; predicting data density based on the historical laser radar data unit to generate predicted data density; determining a data sampling period based on the predicted data density; and obtaining and storing the to-be-stored laser radar data unit corresponding to each channel in multiple channels based on the data sampling period. According to the method, the data collection period is adjusted by dynamically adapting to the data features of all the channels, and the problems of key data loss and storage resource waste caused by fixed-period collection are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of radar technology, and more specifically to a radar data storage method and apparatus. Background Technology

[0002] With the rapid development of fields such as autonomous driving and environmental monitoring, multi-channel LiDAR systems are widely used because they can simultaneously acquire multi-dimensional environmental information. However, current multi-channel LiDAR data storage generally adopts a fixed-period acquisition strategy, meaning that all channels acquire and store data according to a preset, uniform cycle. This approach has significant drawbacks: Firstly, the environmental characteristics of the detection areas corresponding to different channels vary greatly. For example, some channels may continuously detect areas with dense dynamic targets, while others may face open environments for extended periods. A fixed cycle cannot accommodate these differences, leading to insufficient data acquisition in dynamic areas and the loss of key features, while in open areas, excessive redundant data consumes a large amount of storage resources. Secondly, the data volume of multi-channel systems grows exponentially. Fixed-period acquisition can cause uneven load on the storage system, reducing data read and write efficiency. Furthermore, subsequent data processing requires filtering out a large amount of invalid data, increasing computational costs. Summary of the Invention

[0003] In order to overcome the above-mentioned technical problems existing in the prior art, the present invention provides a radar data storage method and apparatus.

[0004] On one hand, the present invention provides a radar data storage method, applied to a multi-channel lidar system, the method comprising: Acquire the historical lidar data units of each channel in the multi-channel system respectively; Based on the historical lidar data units, the data density is predicted to generate a predicted data density; The data sampling period is determined based on the predicted data density; Based on the data sampling period, acquire and store the LiDAR data units to be stored corresponding to each channel in the multi-channel array.

[0005] Preferably, the historical lidar data unit includes a first historical lidar data unit within a first time window and a second historical lidar data unit within a second time window, wherein the duration of the second time window is longer than the duration of the first time window, and the second time window is a multiple of the first time window; based on the historical lidar data unit, the data density is predicted to generate a predicted data density, including: The degree of echo power fluctuation is determined based on the first historical lidar data unit; The first data density is determined based on the degree of echo power fluctuation. Based on the window size of the first time window, the second historical lidar data is divided into multiple data slices; Extract the radar features corresponding to each segment to generate a radar feature sequence; The radar feature sequence is concatenated with the first data density; The concatenated features are input into a preset LSTM model to obtain the second data density; The first data density and the second data density are fused to obtain the predicted data density.

[0006] Preferably, the first data density and the second data density are fused to obtain the predicted data density, including: Calculate channel similarity based on the first historical lidar data unit; The channel spatial weights are determined based on the channel similarity. The confidence level corresponding to the first data density is determined based on the degree of echo power fluctuation and the channel spatial weight.

[0007] Preferably, fusing the first data density and the second data density to obtain the predicted data density further includes: The confidence level corresponding to the second data density is determined based on the prediction error of the preset LSTM model; The first data density and the second data density are fused based on the confidence levels corresponding to the first data density and the second data density to obtain the predicted data density.

[0008] Preferably, acquiring and storing the lidar data units to be stored corresponding to each channel in the multi-channel array based on the data sampling period includes: The data sampling period is used to determine the LiDAR data units to be stored for each channel in the multi-channel system. The importance of the LiDAR data units to be stored is analyzed to determine the data type. Determine the storage strategy based on the data type; The LiDAR data unit to be stored is stored based on the storage strategy.

[0009] Preferably, the importance of the lidar data units to be stored is analyzed to determine the data type, including: Obtain radar data topology map; The LiDAR data units to be stored are abstracted into graph nodes, and the corresponding node features are extracted; Based on preset spatiotemporal association rules, spatiotemporal association edges are established between each graph node and other graph nodes in the radar data topology graph, and the corresponding edge weights are determined to obtain a new radar data topology graph. The degree centrality and betweenness centrality of each graph node are calculated based on the new radar data topology graph. The importance of a node is determined based on the degree centrality and the betweenness centrality. The data type is determined based on the importance of the node.

[0010] Preferably, the data type includes at least core data, and the method further includes; After storing the data corresponding to N data sampling periods, the proportion of the core data is obtained; Whether to update the data sampling period is determined based on the proportion of the core data.

[0011] On the other hand, the present invention provides a radar data storage device, applicable to the method described in the preceding claim, the device comprising: The data acquisition unit is used to acquire the historical lidar data of each channel in the multi-channel array. A density prediction unit is used to predict the data density based on the historical lidar data unit and generate a predicted data density. A period determination unit is used to determine the data sampling period based on the predicted data density; The data storage unit acquires and stores the LiDAR data units to be stored for each channel in the multi-channel array based on the data sampling period.

[0012] The present invention has at least the following technical effects through the technical solution provided by the present invention: By dynamically adapting the data acquisition cycle to the data characteristics of each channel, the problem of critical data loss and storage resource waste caused by fixed-cycle acquisition is effectively solved.

[0013] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a radar data storage method provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a radar data storage device provided in an embodiment of the present invention. Detailed Implementation

[0015] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0016] In this invention, the terms "system" and "network" are used interchangeably. "Multiple" refers to two or more; therefore, in this invention, "multiple" can also be understood as "at least two." "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / ", unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, it should be understood that in the description of this invention, terms such as "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or order.

[0017] With the rapid development of fields such as autonomous driving and environmental monitoring, multi-channel LiDAR systems are widely used because they can simultaneously acquire multi-dimensional environmental information. However, current multi-channel LiDAR data storage generally adopts a fixed-period acquisition strategy, meaning that all channels acquire and store data according to a preset, uniform cycle. This approach has significant drawbacks: Firstly, the environmental characteristics of the detection areas corresponding to different channels vary greatly. For example, some channels may continuously detect areas with dense dynamic targets, while others may face open environments for extended periods. A fixed cycle cannot accommodate these differences, leading to insufficient data acquisition in dynamic areas and the loss of key features, while in open areas, excessive redundant data consumes a large amount of storage resources. Secondly, the data volume of multi-channel systems grows exponentially. Fixed-period acquisition can cause uneven load on the storage system, reducing data read and write efficiency. Furthermore, subsequent data processing requires filtering out a large amount of invalid data, increasing computational costs.

[0018] For the reasons stated above, please refer to Figure 1 This invention provides a radar data storage method applied to a multi-channel lidar system. The method includes: acquiring historical lidar data units for each channel in the multi-channel system; predicting the data density based on the historical lidar data units to generate a predicted data density; determining a data sampling period based on the predicted data density; and acquiring and storing the lidar data units to be stored corresponding to each channel in the multi-channel system based on the data sampling period.

[0019] It should be noted that the radar frame acquired by a channel within one scanning cycle is a lidar data unit. A historical lidar data unit includes multiple historical radar frames, and each lidar data unit contains key information such as point cloud coordinates, echo intensity, timestamp, channel identifier, and signal-to-noise ratio.

[0020] In one possible implementation, firstly, historical LiDAR data units for each channel within a preset time period are extracted in real time through the multi-channel data interface of the LiDAR system and stored in a local cache database according to channel ID. Secondly, a data density prediction module is constructed based on the historical LiDAR data units: for each channel, the number of point clouds per unit time in the historical data (i.e., data density) is counted, and the rate of change of data density is calculated using a sliding window algorithm (window size set to 5 minutes). Combined with a polynomial fitting method, the trend of data density change in the next minute is predicted to generate the predicted data density for that channel. Next, a mapping model between predicted data density and data sampling period is established: preset data density threshold ranges (e.g., low density: 0-50 points / second, medium density: 51-200 points / second, high density: above 201 points / second), and correspondingly different data sampling periods are set. Based on the threshold range into which the predicted data density of each channel falls, its specific data sampling period is determined. Finally, according to the determined data sampling period, the LiDAR data units to be stored in each channel are acquired in real time. After the channel identifier is verified, they are classified and stored using a distributed storage architecture (such as HDFS). At the same time, the data sampling period and timestamp of each data are recorded to facilitate subsequent data traceability and period adjustment.

[0021] This invention effectively solves the problems of critical data loss and storage resource waste caused by fixed-cycle acquisition by dynamically adapting the data acquisition cycle to the data characteristics of each channel. On the one hand, in data-dense areas, increasing the data acquisition cycle ensures complete recording of environmental details and dynamic target information, providing high-quality data support for subsequent applications such as target detection and trajectory prediction. On the other hand, reducing the data acquisition cycle in data-sparse areas significantly reduces the amount of redundant data stored, improving the space utilization and data read / write efficiency of the storage system. Simultaneously, the differentiated processing strategy for multiple channels enables the system to better adapt to complex and changing detection environments, enhancing the practicality and reliability of the multi-channel LiDAR system in real-world scenarios.

[0022] Furthermore, the accuracy of data density prediction directly determines the rationality of the data acquisition cycle. In practical applications, relying solely on historical LiDAR data units within a single time window—for example, using only short-term or long-term (e.g., 1-hour) data for prediction—is problematic. While short-term window data can reflect recent data fluctuations, it is susceptible to transient noise, leading to oscillating prediction results. Long-term window data, while reflecting overall trends, lags behind sudden changes in data density (such as the sudden appearance of dense target groups), failing to capture short-term, drastic fluctuations in a timely manner. Especially in multi-channel LiDAR systems, the data characteristics of each channel exhibit a complex coupling relationship between short-term fluctuations and long-term trends due to environmental influences. Single-window prediction methods struggle to balance both, resulting in significant deviations between predicted and actual data densities. Consequently, the data acquisition cycle determined based on this prediction is unreasonable, impacting data storage effectiveness.

[0023] In this embodiment of the invention, the historical lidar data unit includes a first historical lidar data unit within a first time window and a second historical lidar data unit within a second time window, wherein the duration of the second time window is longer than the duration of the first time window, and the second time window is a multiple of the first time window; predicting the data density based on the historical lidar data unit to generate a predicted data density includes: determining the echo power fluctuation level based on the first historical lidar data unit; determining a first data density based on the echo power fluctuation level; dividing the second historical lidar data into multiple data slices based on the window size of the first time window; extracting the radar features corresponding to each slice to generate a radar feature sequence; concatenating the radar feature sequence with the first data density; inputting the concatenated features into a preset LSTM model to obtain a second data density; and fusing the first data density and the second data density to obtain the predicted data density.

[0024] In one possible implementation, historical lidar data units are first divided into a first historical lidar data unit with a first time window (e.g., 1s, T1) and a second historical lidar data unit with a second time window (e.g., 60s, T2=60×T1). For the first data density: physical parameters such as echo power, target distance, and reflectivity are extracted from each data unit based on the first historical data and substituted into a preset physical model (e.g., data density = k × echo power / ( The initial physical density is calculated by taking the coefficient k, which is related to reflectivity and determined through calibration experiments. Then, the initial physical density is corrected by considering the echo power fluctuation (calculated using standard deviation). (The larger the fluctuation, the smaller the correction coefficient; for example, the correction coefficient is 0.8 when the fluctuation is >0.5), ultimately yielding the first data density. For the second data density (long-term prediction): the second historical lidar data unit is divided into 60 data slices according to T1. Features such as the first data density, density change rate, and ambient temperature (affecting laser propagation) corresponding to each data slice are extracted to form a feature sequence. This sequence is input into a pre-trained LSTM model (trained with multi-channel long-term data, containing 3 layers of LSTM units, focusing on capturing periodic and trend changes), outputting the second data density. In the fusion stage: a dynamic weighting method is used. When the physical parameters of the first historical data are relatively stable (e.g., reflectivity fluctuation <5%), the weight of the first data density is set to 0.7, and the weight of the second data density is set to 0.3. When the physical parameters are unstable, the weight of the first data density decreases to 0.4, and the weight of the second data density increases to 0.6, resulting in the predicted data density.

[0025] This invention integrates historical data features from short-term and long-term time windows. It captures instantaneous fluctuations and subtle changes in data using data from the first time window, while simultaneously uncovering long-term trends and periodic patterns using data from the second time window and an LSTM model. This effectively overcomes the limitations of single-window prediction. The predicted data density more accurately reflects the actual data distribution characteristics, especially in scenarios of sudden environmental changes (such as the sudden appearance of dense vehicle groups). It can quickly respond to short-term surges in data density while avoiding over-prediction by incorporating long-term trends. This makes the subsequent data collection cycle more aligned with actual needs, further improving the adaptability and reliability of multi-channel LiDAR data storage.

[0026] Furthermore, the above implementation is still limited to prediction based on the data itself, that is, pure data prediction, which cannot be combined with the physical laws of lidar, that is, it cannot be combined with the attenuation law of echo power with detection distance during the process of laser pulse transmission to reception. This law reflects the combined influence of three major factors: geometric attenuation, atmospheric attenuation, and scattering characteristics. In another possible implementation, although short-term window data can reflect recent data fluctuations, it is easily affected by instantaneous noise, which leads to oscillations in the prediction results. Therefore, the embodiments of the present invention do not predict based on short-term window data, but directly based on the theoretical echo power corresponding to each radar frame in the short-term window. Furthermore, a physical model can be constructed based on the attenuation law, and the theoretical echo power can be determined based on the physical model. Then, the degree of fluctuation of the theoretical echo power can be determined. Based on the degree of fluctuation of the theoretical echo power, the corresponding first data density can be determined from a preset correlation table. Then, the first data density and the second data density determined by long-term window data are fused. It should be noted that the method of determining the second data density based on long-term window data may not be further optimized to extract features such as SNR, Doppler frequency shift, power residual and spatiotemporal coordinates (x,y,z,t) corresponding to the channel based on long-term window data, and make predictions based on these extracted features.

[0027] This invention constructs a physical model based on the physical attenuation law of lidar, directly calculates the theoretical echo power of the radar frame within a short window, and determines the first data density. This not only avoids the interference of instantaneous noise on short-term data from the source, ensuring that the first data density is in line with the physical essence and has stronger stability, but also achieves a deep integration of physical laws and long-term data trends.

[0028] After determining the first and second data densities, they need to be fused. One possible approach is to use fixed weights for fusion, but this method does not consider the impact of data correlation between multiple channels on the reliability of the first data density. In reality, adjacent channels of multi-channel LiDAR often detect overlapping or related areas, and their data characteristics exhibit strong correlation. If the first data density of a certain channel deviates significantly from the concurrent data density of adjacent channels, it indicates that the first data density may be affected by noise interference and has low reliability; conversely, if it is highly consistent with the data density of adjacent channels, its reliability is high.

[0029] In this embodiment of the invention, fusing the first data density and the second data density to obtain the predicted data density includes: calculating channel similarity based on the first historical lidar data unit; determining channel spatial weights based on the channel similarity; and determining the confidence level corresponding to the first data density based on the echo power fluctuation degree and the channel spatial weights.

[0030] In one possible implementation, firstly, channel similarity is calculated based on the first historical lidar data unit: features such as signal-to-noise ratio and power residual are extracted from the first historical lidar data unit, where the power residual is determined based on the difference between the theoretical and actual echo power values; based on the extracted features, cosine similarity between channels is calculated. For each channel, the number of channels with a similarity greater than 0.7 (denoted as N) is counted. If N≥3, the spatial weight of that channel is set to 1.0; if 1≤N<3, the spatial weight is set to 0.8; if N=0, the spatial weight is set to 0.5 (indicating that the channel data is highly independent and has low reference value). Next, the confidence level of the first data density is calculated in conjunction with the degree of echo power fluctuation: confidence level = spatial weight × (1 - fluctuation coefficient), where the fluctuation coefficient is the normalized value of the corresponding echo power fluctuation. For example, if a channel has a spatial weight of 1.0 and a fluctuation coefficient of 0.2, then the confidence level of its first data density is 0.8, indicating that the data has high reliability.

[0031] This invention introduces channel similarity analysis to determine spatial weights and combines this with the echo power fluctuation level to calculate the confidence level of the first data density, making the reliability assessment of the first data density more closely reflect the actual correlation characteristics of multi-channel data. When the channel data has high consistency with adjacent channels and stable fluctuations, its first data density is assigned a higher confidence level; conversely, the weight is reduced, effectively avoiding interference from unreliable data on the fusion results.

[0032] Of course, in another possible implementation, if the first data density is determined based on physical laws, its confidence level should be determined based on the power residual. The smaller the power residual, the closer the confidence level is to 1. In practical applications, the highest confidence level corresponding to the first data density is 0.95.

[0033] Furthermore, the confidence level of the second data density also needs to be considered. In practical applications, the prediction performance of the LSTM model is affected by factors such as the distribution of the input data, which may lead to large prediction errors (such as when encountering special environmental scenarios not included in the training samples). If the reliability of the second data density is not assessed and a fixed or empirically based weight is directly assigned, the second data density may still significantly affect the final result even when the model's prediction bias is large. This could cause the fused prediction data density to deviate from reality, affecting the rationality of the data sampling period.

[0034] In this embodiment of the invention, fusing the first data density and the second data density to obtain the predicted data density further includes: determining the confidence level corresponding to the second data density based on the prediction error of the preset LSTM model; and fusing the first data density and the second data density based on the confidence level corresponding to the first data density and the confidence level corresponding to the second data density to obtain the predicted data density.

[0035] In a possible implementation, for the second data density, its corresponding confidence level is determined based on the prediction error of a preset LSTM model: after the LSTM model is trained, the test set data is used for verification, and the absolute error between the second data density output by the model and the actual data density in each group of data is recorded. The mean value of this error (denoted as μ) and the standard deviation (denoted as σ) are calculated; during actual prediction, for the currently input radar feature sequence, the real-time error (denoted as e) between the second data density output by the model and the recent actual data density of this channel (such as the previous 30 seconds) is calculated. If e ≤ μ - σ, it indicates that the prediction accuracy of the second data density is extremely high, and its confidence level is set to 1.0; if μ - σ < e ≤ μ, the confidence level is set to 0.8; if μ < e ≤ μ + σ, the confidence level is set to 0.5; if e > μ + σ, it indicates that the prediction deviation of the second data density is large, and the confidence level is set to 0.2. Secondly, combining the confidence level (C) corresponding to the first data density determined above, and the confidence level (C2) corresponding to the second data density obtained in this step, the first data density and the second data density are fused using a weighted fusion formula, that is, the predicted data density = (C1 × the first data density + C2 × the second data density) / (C1 + C2). Through this method, the fusion result is made more consistent with the actual reliability of the two data densities, improving the accuracy of the predicted data density.

[0036] In the embodiment of the present invention, by introducing the prediction error of the LSTM model to evaluate the confidence level of the second data density, the fusion process can dynamically adapt to the real-time prediction performance of the model. When the model predicts accurately, the second data density is given a higher weight, giving full play to its advantage of capturing long-term trends; when the model prediction deviation is large, its weight is reduced to avoid misleading the final result.

[0037] Once the predicted data density is determined, the corresponding data sampling period can be determined based on the preset correlation, and data storage processing can be performed based on this sampling period. This achieves on-demand allocation of storage resources and avoids the problems of data redundancy or missing key information under fixed periods. However, there are still obvious limitations in practical applications: using a uniform storage strategy (such as the same storage medium, redundancy, compression algorithm, etc.) to process all channels' data to be stored according to the data acquisition period ignores the significant differences in the actual information value of different data units. For example, data units containing key information such as dynamic targets and complex terrain are crucial for core tasks such as environmental perception and target recognition, requiring high reliability and low latency storage guarantees; while static background data in open areas has lower information value and does not need to occupy too many high-quality storage resources. This "one-size-fits-all" storage method may not only lead to the risk of loss or damage of key data due to insufficient storage resource allocation, but also cause low-value data to over-occupy high-cost storage resources, resulting in low storage efficiency and increased costs. At the same time, subsequent data processing requires sifting key information from a large amount of low-value data, increasing computational overhead and affecting the overall system performance. Therefore, the key to improving the overall performance of multi-channel lidar data storage systems lies in how to formulate differentiated storage strategies based on the information value of data units, in addition to dynamic data acquisition cycles.

[0038] In this embodiment of the invention, acquiring and storing the LiDAR data units to be stored corresponding to each channel in a multi-channel system based on the data sampling period includes: determining the LiDAR data units to be stored corresponding to each channel in a multi-channel system based on the data sampling period; performing importance analysis on the LiDAR data units to be stored to determine the data type; determining a storage strategy based on the data type; and storing the LiDAR data units to be stored based on the storage strategy.

[0039] In one possible implementation, according to the data sampling period, LiDAR data units to be stored are acquired. These data units include basic information such as point cloud coordinates, echo intensity, timestamps, and channel identifiers. Next, an importance analysis is performed on the LiDAR data units to be stored to determine the data type: features are extracted from the LiDAR data units, and the extracted features are input into a preset importance evaluation model to obtain a scoring result; based on the scoring result, the LiDAR data units to be stored are divided into core data (8-10 points, such as multi-channel overlapping area data containing fast-moving targets at close range), important data (5-7 points, such as single-channel key area data containing static targets at long range), and general data (0-4 points, such as open area data without significant targets), thus completing the determination of the data type. Next, corresponding storage strategies are formulated based on the determined data types: core data uses multi-replica redundant storage (3 copies), high-speed solid-state storage (SSD) is selected as the storage medium, lossless compression is used to ensure data integrity, and the retention period is set at 1 year; important data uses dual-replica storage, hybrid storage (SSD cache + HDD) is selected as the storage medium, medium-loss compression is used, and the retention period is set at 3 months; general data uses single-replica storage, large-capacity hard disk drive (HDD) is selected as the storage medium, high-loss compression is used, and the retention period is set at 1 month. Finally, based on the above storage strategies, the LiDAR data units to be stored in each channel are classified and stored in the corresponding storage media according to their data types, and the data type, storage location, and strategy parameters are recorded through the metadata management module to achieve differentiated storage management of the data.

[0040] This invention achieves precise allocation of storage resources by analyzing the importance of the LiDAR data units to be stored and employing a differentiated storage strategy. Core data is stored using a highly redundant, high-speed, and long-term storage method, ensuring its security and accessibility and providing reliable data support for critical applications such as autonomous driving decision-making and environmental event backtracking. Important and general data are stored using appropriate solutions based on their value, significantly reducing overall storage costs and improving the space utilization of the storage system. Simultaneously, categorized storage allows subsequent data retrieval and analysis to directly focus on core data, reducing the amount of invalid data processed and improving data application efficiency.

[0041] The above implementation relies on the local features (such as the number of targets and distance) of the LiDAR data unit to be stored during the importance analysis process, ignoring the spatiotemporal correlation between the data unit and other data. In reality, multi-channel LiDAR data exhibits strong spatiotemporal correlation: for example, a data unit may have few targets (local features indicating low importance), but it is a crucial node connecting multiple core data units, thus possessing high actual importance. Assessing importance solely based on local features may lead to misclassification of such valuable data as low importance, resulting in inappropriate storage strategies that affect the overall data integrity and correlation, and reduce the accuracy of subsequent multi-channel data fusion and scene reconstruction. In this embodiment of the invention, the importance analysis of the LiDAR data units to be stored and the determination of the data type include: obtaining a radar data topology graph; abstracting the LiDAR data units to be stored into graph nodes and extracting the corresponding node features; establishing spatiotemporal association edges between each graph node and other graph nodes in the radar data topology graph based on preset spatiotemporal association rules, and determining the corresponding edge weights to obtain a new radar data topology graph; calculating the degree centrality and betweenness centrality of each graph node based on the new radar data topology graph; determining the node importance based on the degree centrality and betweenness centrality; and determining the data type based on the node importance.

[0042] In one possible implementation, firstly, a radar data topology graph is constructed. Secondly, the LiDAR data units to be stored are processed: each LiDAR data unit is abstracted into a new graph node, and its timestamp, spatial coordinates, channel ID, SNR, echo power, and other features are extracted as node features. Based on preset spatiotemporal association rules, for example, if two nodes satisfy a time difference ≤ 1 second and a spatial distance ≤ 10 meters (same channel) or ≤ 5 meters (adjacent channels), a spatiotemporal association edge is established. The edge weight is set as the normalized inverse ratio of time difference to spatial distance; the higher the value, the stronger the association. The spatiotemporal distance between the new node and all nodes in the radar data topology graph is calculated, and association edges that meet the conditions are established and assigned weights to form an extended radar data topology graph containing the new node, i.e., the new radar data topology graph. Next, the node importance is calculated: the degree centrality and betweenness centrality of each new node are calculated using graph theory algorithms. After normalizing both, they are weighted and summed according to a weight of 40% for degree centrality and 60% for betweenness centrality to obtain the overall node importance. Finally, the data type is determined based on the comprehensive importance threshold (≥0.8 is core data, 0.5-0.79 is important data, and <0.5 is general data).

[0043] It should be noted that abstracting each LiDAR data unit to be stored into a new graph node means abstracting the LiDAR data acquisition unit of each channel within the data acquisition cycle into an independent graph node, that is, the number of new graph nodes is consistent with the number of channels.

[0044] This invention, through the introduction of radar data topology analysis, identifies the spatiotemporal relationships of LiDAR data units to be stored, overcoming the limitations of single local feature evaluation and making importance judgments more aligned with the actual value of the data within the overall system. For LiDAR data units with inconspicuous local features but crucial correlational roles, they can be accurately identified and assigned higher importance, ensuring the adoption of appropriate storage strategies and thus guaranteeing the overall correlation and integrity of multi-channel LiDAR data.

[0045] In the process of constructing a radar data topology graph and generating a new radar data topology graph, corresponding spatiotemporal related edges and edge weights are constructed, while semantic relatedness, i.e., the correlation between node features, is ignored. In another possible implementation, based on the node features of the graph node and the node features of other graph nodes in the radar data topology graph, the similarity between the graph node and other graph nodes is calculated, and semantic related edges and corresponding edge weights are generated based on the similarity.

[0046] Furthermore, if the corresponding data acquisition cycle needs to be calculated every time data storage is performed, the computational overhead will increase dramatically. To reduce computational overhead, it is designed to acquire data multiple times according to each data acquisition cycle and perform the corresponding storage before re-executing the data acquisition cycle update.

[0047] In this embodiment of the invention, the data type includes at least core data, and the method further includes: storing the data corresponding to N data sampling periods, obtaining the proportion of the core data, and determining whether to update the data sampling period based on the proportion of the core data.

[0048] In one possible implementation, N is preset as the number of data acquisition cycles. After N cycles of acquisition and storage operations are completed, a core data percentage statistics process is triggered. All data units stored within the N cycles are retrieved, and the ratio of the total number of core data units to the total number of existing data units is calculated to obtain the core data percentage (denoted as R). Next, a reasonable threshold range for the core data percentage is set (e.g., 30%-60%), and a feedback adjustment rule is established: if R < 30% or R > 60%, it indicates that the current data acquisition cycle can no longer meet the actual needs and needs to be updated. The update method is the same as the method for determining the current data acquisition cycle, and will not be elaborated in this embodiment; if 30% ≤ R ≤ 60%, the current data acquisition cycle remains unchanged. Finally, the updated data acquisition cycle is used as the basis for the next round (N cycles of data), and the update time, the data acquisition cycles before and after the update, and the core data percentage are recorded to generate log data for subsequent optimization and adjustment rules.

[0049] This invention, through dynamic monitoring of the core data ratio and feedback adjustment of the data acquisition cycle, enables the system to continuously adapt to changes in the detection environment, ensuring that the core data ratio remains within a reasonable range. This enhances the adaptive capability of the multi-channel LiDAR system, allowing the data acquisition and storage strategies to maintain high efficiency and rationality over the long term. While meeting the data application needs of different scenarios, it maximizes the balance between storage resource consumption and data quality.

[0050] Please refer to Figure 2 Based on the same inventive concept, the present invention provides a radar data storage device, the device comprising: a data acquisition unit for acquiring historical lidar data units of each channel in the multi-channel array; a density prediction unit for predicting the data density based on the historical lidar data units to generate a predicted data density; a period determination unit for determining a data sampling period based on the predicted data density; and a data storage unit for acquiring and storing lidar data units to be stored corresponding to each channel in the multi-channel array based on the data sampling period.

[0051] It should be understood that the radar data storage device provided in this embodiment of the invention and the radar data storage method provided in the above embodiment are based on the same inventive concept. For more specific working principles of each unit in this embodiment of the invention, please refer to the above embodiment, which will not be repeated in this embodiment of the invention.

[0052] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.

[0053] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0054] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0055] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. A radar data storage method, applied to a multi-channel lidar system, characterized in that, include: Acquire the historical lidar data units of each channel in the multi-channel system respectively; Based on the historical lidar data units, the data density is predicted to generate a predicted data density; The data sampling period is determined based on the predicted data density; Based on the data sampling period, acquire and store the LiDAR data units to be stored corresponding to each channel in the multi-channel array.

2. The radar data storage method according to claim 1, characterized in that, The historical lidar data unit includes a first historical lidar data unit within a first time window and a second historical lidar data unit within a second time window, wherein the duration of the second time window is longer than the duration of the first time window, and the second time window is a multiple of the first time window; based on the historical lidar data unit, the data density is predicted to generate a predicted data density, including: The degree of echo power fluctuation is determined based on the first historical lidar data unit; The first data density is determined based on the degree of echo power fluctuation. Based on the window size of the first time window, the second historical lidar data is divided into multiple data slices; Extract the radar features corresponding to each segment to generate a radar feature sequence; The radar feature sequence is concatenated with the first data density; The concatenated features are input into a preset LSTM model to obtain the second data density; The first data density and the second data density are fused to obtain the predicted data density.

3. A radar data storage method according to claim 2, characterized in that, The first data density and the second data density are fused to obtain the predicted data density, including: Calculate channel similarity based on the first historical lidar data unit; The channel spatial weights are determined based on the channel similarity. The confidence level corresponding to the first data density is determined based on the degree of echo power fluctuation and the channel spatial weight.

4. A radar data storage method according to claim 2, characterized in that, The method further includes fusing the first data density and the second data density to obtain the predicted data density, and also includes: The confidence level corresponding to the second data density is determined based on the prediction error of the preset LSTM model; The first data density and the second data density are fused based on the confidence levels corresponding to the first data density and the second data density to obtain the predicted data density.

5. A radar data storage method according to claim 1, characterized in that, Based on the data sampling period, acquire and store the LiDAR data units to be stored corresponding to each channel in the multi-channel array, including: The data sampling period is used to determine the LiDAR data units to be stored for each channel in the multi-channel system. The importance of the LiDAR data units to be stored is analyzed to determine the data type. Determine the storage strategy based on the data type; The LiDAR data unit to be stored is stored based on the storage strategy.

6. A radar data storage method according to claim 5, characterized in that, The importance of the LiDAR data units to be stored is analyzed to determine the data type, including: Obtain radar data topology map; The LiDAR data units to be stored are abstracted into graph nodes, and the corresponding node features are extracted; Based on preset spatiotemporal association rules, spatiotemporal association edges are established between each graph node and other graph nodes in the radar data topology graph, and the corresponding edge weights are determined to obtain a new radar data topology graph. The degree centrality and betweenness centrality of each graph node are calculated based on the new radar data topology graph. The importance of a node is determined based on the degree centrality and the betweenness centrality. The data type is determined based on the importance of the node.

7. A radar data storage method according to claim 5, characterized in that, The data type includes at least core data, and the method further includes; After storing the data corresponding to N data sampling periods, the proportion of the core data is obtained; Whether to update the data sampling period is determined based on the proportion of the core data.

8. A radar data storage device, applicable to the method described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to acquire the historical lidar data of each channel in the multi-channel array. A density prediction unit is used to predict the data density based on the historical lidar data unit and generate a predicted data density. A period determination unit is used to determine the data sampling period based on the predicted data density; The data storage unit acquires and stores the LiDAR data units to be stored for each channel in the multi-channel array based on the data sampling period.

Images