Light vehicle identification method and system based on cooperation of multiple sensors

Abstract

To provide a non-motorized vehicle recognition method and system based on multi-sensor collaboration in relation to the technical field of intelligent transportation.SOLUTION: The method includes steps of: constructing a sensor group based on a plurality of sensors, and performing a data collection based on a target range through the sensor group to generate a multi-class regional dataset; transmitting the multi-class regional dataset to a data fusion channel to generate an initial fusion dataset; synchronizing the initial fusion dataset to a data preprocessing unit to perform preprocessing to generate a target fusion dataset; utilizing a feature extraction unit to traverse the target fusion dataset to perform a feature extraction of a target non-motorized vehicle, and generating a target feature information set; and constructing a target recognition unit, and intelligently recognizing the target fusion dataset through the target recognition unit.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a method and system for identifying light vehicles based on the cooperation of multiple sensors. [Background technology]

[0002] While conventional light vehicle (non-motorized vehicle) identification methods and systems utilize the complementary advantages of multiple sensors to improve identification accuracy and stability, many challenges remain. Data redundancy and conflicts between different sensors complicate data processing, while issues such as high cost, high complexity, and poor environmental adaptability limit their widespread application. Therefore, optimizing sensor placement, improving data processing efficiency, and reducing costs remain urgent challenges for this technology. Summary of the Invention [Problem to be solved by the invention]

[0003] The present application aims to provide a method and system for identifying light vehicles based on the cooperation of multiple sensors to solve the problem in the prior art that data processing becomes complicated due to data redundancy and collision between different sensors. [Means for solving the problem]

[0004] In consideration of the above problems, the present application provides a light vehicle identification method and system based on the cooperation of multiple sensors.

[0005] A first aspect of the present application provides a light vehicle identification method based on cooperation of multiple sensors, the method including the following steps: The method includes: an area dataset generating step of constructing a sensor group based on a plurality of sensors, using the sensor group to collect data based on a target range, and generating a plurality of types of area datasets; a fusion dataset generating step of sending the plurality of types of area datasets to a data fusion channel to generate an initial fusion dataset; a target fusion dataset generating step of synchronizing the initial fusion dataset with a data pre-processing unit to perform pre-processing, update the initial fusion dataset, and generate a target fusion dataset; a target feature information set generating step of using a feature extraction unit to scan the target fusion dataset, extract features of a target light vehicle, and generate a target feature information set, where there is a correspondence between the target feature information set and the target fusion dataset; and an identification step of constructing a target identification unit based on the target feature information set, and using the target identification unit to intelligently identify the target fusion dataset.

[0006] A second aspect of the present application provides a light vehicle identification system based on the cooperation of multiple sensors, the system including the following modules: The multiple area dataset generation module constructs a sensor group based on multiple sensors, and uses the sensor group to collect data based on a target range to generate multiple area datasets. The initial fusion dataset generation module transmits the multiple area datasets to a data fusion channel to generate an initial fusion dataset. The target fusion dataset generation module synchronizes the initial fusion dataset with a data pre-processing unit for pre-processing, updates the initial fusion dataset, and generates a target fusion dataset. The target feature information set generation module uses a feature extraction unit to scan the target fusion dataset, extract features of a target light vehicle, and generate a target feature information set, where there is a correspondence between the target feature information set and the target fusion dataset. The intelligent identification module constructs a target identification unit based on the target feature information set, and uses the target identification unit to intelligently identify the target fusion dataset. [Effects of the Invention]

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application configures a sensor group based on multiple sensors, collects data based on a target range using the sensor group, generates multiple types of area datasets, transmits the multiple types of area datasets to a data fusion channel to generate an initial fusion dataset, synchronizes the initial fusion dataset with a data pre-processing unit for pre-processing, updates the initial fusion dataset, generates a target fusion dataset, uses a feature extraction unit to scan the target fusion dataset, extracts features of the target light vehicle, generates a target feature information set, configures a target identification unit based on the target feature information set, and uses the target identification unit to intelligently identify the target fusion dataset. The present invention solves the problem in the prior art that data processing is complicated due to data redundancy and collision between different sensors, and achieves the technical effect of improving the accuracy and stability of identification precision by fusing data from different sensors. [Brief explanation of the drawings]

[0008] In order to more clearly explain the technical solutions in the embodiments of the present invention, the drawings necessary for describing the embodiments are briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative efforts. [Figure 1] FIG. 1 is a schematic diagram of a flowchart of a light vehicle identification method based on cooperation of multiple sensors provided by an embodiment of the present application; [Figure 2] 1 is a schematic diagram of a light vehicle identification system structure based on the cooperation of multiple sensors provided by an embodiment of the present application; FIG. DETAILED DESCRIPTION OF THE INVENTION

[0009] The present application provides a method and system for identifying light vehicles based on the coordination of multiple sensors, thereby solving the problems of the prior art, such as data redundancy between different sensors and complicated data processing due to collisions, and achieving the technical effect of improving the accuracy and stability of identification precision by fusing data from different sensors.

[0010] The technical solutions in the embodiments of the present application are clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of protection of the present application.

[0011] It should be noted that terms such as "first," "second," etc. in the specification and claims of this application and in the drawings are used to distinguish between similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such terms, as used herein, are interchangeable where appropriate, such that the embodiments of this application described herein may be performed in orders other than those illustrated or described herein. Additionally, the terms "comprise" and "have," and variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or server comprising a series of steps or units is not limited to the explicitly recited steps or units. Rather than being explicitly listed, the present invention may include other steps or modules not expressly listed or inherent to a process, method, product, or device.

[0012] Example 1: As shown in FIG. 1 , the present application provides a light vehicle identification method based on the cooperation of multiple sensors, and the method includes the following steps: In step S100, a sensor group is constructed based on a plurality of sensors, and data is collected by the sensor group based on a target range to generate a plurality of types of regional data sets. In the embodiments of the present application, appropriate sensors are selected based on the actual application needs and the characteristics of the environment, including but not limited to image sensors (cameras), radars, infrared sensors, speed sensors, position sensors, etc., each with different sensing capabilities and characteristics to monitor and acquire different types of data.

[0013] These sensors are then deployed within the target area. The deployment location is selected based on factors such as the area where the target light vehicle may appear, the communication distance and coverage area between sensors, etc. Once deployment is complete, a unique identifier is assigned to each sensor, and a communication protocol and connection method between these sensors are established. Through communication, the sensor group forms a unified network, enabling data sharing and collaboration.

[0014] Once the sensor group is created and communication associations are established, data collection begins. The sensor group collects various data in real time within the target range based on the preset sampling frequency and parameter settings. This data includes information on the trajectory, speed, position, and direction of the light vehicle, as well as data on the surrounding environment such as weather conditions and light intensity.

[0015] The collected data is classified and segmented to generate multiple regional datasets. During classification and segmentation, preprocessing and cluster analysis are performed on the collected data. Data is classified based on data source and data type, for example, image data collected by a camera and velocity data obtained by radar. The classified data is then processed using a clustering algorithm such as k-means to divide the data based on spatial location and feature similarity, forming multiple regional datasets. Finally, the classified and segmented data is combined into multiple regional datasets.

[0016] In step S200, the multiple types of regional data sets are sent to a data fusion channel to generate an initial fusion data set. In this embodiment, the formats of multiple types of area data sets are unified and converted into a unified format through data preprocessing. The multiple types of area data sets are then transmitted to a data fusion channel. During the transmission process, real-time and synchronous data are ensured. By using the same hardware to simultaneously issue trigger acquisition commands, time synchronization of acquisition and measurement by each sensor can be achieved. Data from different sensors can be synchronized in time or space through unified timestamps or spatial calibration.

[0017] The data fusion channel creates an appropriate data fusion strategy, which is determined based on the type, importance, and fusion target of the data. For example, certain important data may be fused using a weighted average method, while some auxiliary data may be fused using simple addition or maximum value operations.

[0018] Based on the fusion strategy, multiple types of domain datasets are fused. The fusion process includes data layer fusion, feature layer fusion, and decision layer fusion. Data layer fusion directly fuses the original data to generate a new dataset. Feature layer fusion fuses extracted features to form a richer feature representation. Decision layer fusion involves each sensor making an initial decision and then making a comprehensive judgment. After the above fusion processes, an initial fusion dataset is generated.

[0019] In step S300, the initial fusion dataset is synchronized with a data pre-processing unit for pre-processing to update the initial fusion dataset and generate a target fusion dataset. In an embodiment of the present application, the initial fusion dataset is synchronized with a data preprocessing unit for preprocessing. In the preprocessing process, data cleansing is first performed to remove noise, outliers, and redundant information from the data. The initial fusion dataset contains noise and outliers caused by factors such as sensor errors and transmission interference. Data cleansing identifies and removes these bad data to improve the purity and accuracy of the data. Data standardization and normalization are then performed to convert the data in the initial fusion dataset into the same dimensions and range. Next, feature extraction and selection are performed to convert the original data into a more representative feature vector, thereby highlighting the underlying rules and patterns in the data. Finally, data augmentation and transformation are performed to increase the volume and diversity of the data.

[0020] After the above pre-processing steps, the initial fusion dataset is updated and a target fusion dataset is generated.

[0021] In step S400, a feature extraction unit is used to traverse the target fusion dataset, extract features of the target light vehicle, and generate a target feature information set, where there is a correspondence between the target feature information set and the target fusion dataset. In the embodiment of the present application, first, ensure that the feature extraction unit is correctly configured, select an appropriate feature extraction algorithm, for example, use the SIFT algorithm, and adjust the parameters according to the specific needs of light vehicle identification to extract feature information that contributes to light vehicle identification.

[0022] Next, the feature extraction unit starts scanning the target fusion dataset, sequentially accessing each record or each data point in the dataset in a certain order, for example, row by row or column by column. During the scanning process, the feature extraction unit analyzes the attributes and features of each data point. Based on the preset feature selection criteria, the feature extraction unit selects features from each data point that contribute to identifying light vehicles, including shape, size, etc., which reflect the uniqueness of the vehicle and its difference from other objects.

[0023] After feature selection and extraction, the feature extraction unit generates a target feature information set, which includes all feature information extracted from the target fusion dataset and has a one-to-one correspondence with the target fusion dataset.

[0024] In step S500, a target identification unit is constructed based on the target feature information set, and the target fusion data set is intelligently identified by the target identification unit.

[0025] In the embodiment of the present application, a convolutional neural network model is selected to construct the target identification unit. The convolutional model is trained by collecting a large number of datasets containing light vehicle images obtained from actual shooting and annotation. These datasets include images of light vehicles in various scenes, angles, and lighting conditions, ensuring that the model can learn sufficient feature information.

[0026] A dataset containing a large number of light vehicle images is input to a convolutional neural network (CNN), which extracts image features at each layer and converts the original image into a higher-order feature representation. After processing multiple convolutional layers, pooling layers, and fully connected layers, the network outputs a probability distribution indicating the likelihood that the input image belongs to each light vehicle category. To improve the model's classification performance, the model is optimized by adjusting the network structure and optimizing hyperparameter settings, and a technique called regularization is used to prevent overfitting.

[0027] Through the above process, the construction of the target identification unit is completed, and the intelligent identification of light vehicle targets is realized.

[0028] Furthermore, step S100 in the method provided by the embodiment of the present application may include: Based on the monitoring needs information, the plurality of sensors are arranged within a target range, and the sensors in the same area are associated with each other through communication, and based on the associated information, the sensor group is constructed; collecting data from a target area with the sensor group to generate a plurality of area data sets; The method further includes performing a cluster analysis on the plurality of regional data sets to generate the plurality of types of regional data sets.

[0029] In this embodiment, the type of data to be collected, such as the traffic volume, speed, driving trajectory, and parking location of light vehicles, is determined based on the monitoring needs of light vehicles. Based on this information, appropriate sensor types, such as image sensors (cameras), radar sensors, and infrared sensors, are selected and accurately deployed within the target area. After deployment is complete, wireless communication technology, such as ZigBee, is used to connect the sensors in the same area.

[0030] After the sensor group is established, data collection is initiated within the target area. The sensors capture real-time information such as images, speed, and location of light vehicles, and generate multiple regional datasets. Each regional dataset contains detailed data on light vehicles within the corresponding area, such as traffic volume statistics, speed distribution, and driving trajectory maps.

[0031] Next, cluster analysis is performed on the multiple region datasets to identify the characteristics and behavior patterns of light vehicles in different regions. For example, the K-Means algorithm is used to divide the target area into high-traffic volume regions, low-traffic volume regions, high-speed driving regions, and low-speed driving regions based on the traffic volume and speed data of light vehicles, and multiple types of region datasets are generated.

[0032] Furthermore, step S200 in the method provided by the embodiment of the present application Standardizing timestamps for the plurality of types of area data sets, detecting time deviations of these area data sets based on the standardized timestamps, and generating time deviation data; Perform time calibration according to the time deviation data, verify the time consistency according to the calibration result, and establish a fusion time consistency branch; constructing a virtual spatial coordinate system, scanning the plurality of types of area data sets to perform positional registration, and generating a registration coordinate set; verifying spatial consistency based on the registration coordinate set and establishing a fusion spatial consistency branch; The method further includes constructing the data fusion channel based on the fused time-aligned branch and the fused space-aligned branch, where the fused time-aligned branch and the fused space-aligned branch have a connection order.

[0033] In an embodiment of the present application, timestamps are standardized because there are various data sources, different collection devices, and differences in the format and precision of timestamps for each data set. When standardizing timestamps, a unified timestamp format, such as a format based on an international standard such as ISO 8601, is first determined. Then, the timestamps for all data sets are converted to this unified format to ensure consistency in the representation format. Finally, the precision of the timestamps is unified according to the needs of the analysis, for example, all timestamps are unified to the second or minute level.

[0034] After timestamp standardization is complete, time deviations between multiple types of area data sets are detected based on these standardized timestamps. Time deviations can occur due to various factors in the data collection, transmission, or processing processes. When detecting time deviations between multiple types of area data sets, timestamps that indicate the same or similar times from different data sets are paired. The timestamps are paired by comparing their proximity, and a time window is set. Timestamps within this window are considered paired. Next, the time difference between the paired timestamps is calculated to obtain the time interval between them. Next, statistical analysis is performed on these time differences, such as calculating the average value and standard deviation, to understand the distribution and degree of time deviation. Finally, obviously abnormal time deviation values ​​are identified, and time deviation data is obtained.

[0035] Then, time calibration is performed using the time deviation data to adjust the timestamps of the datasets to align with a unified timeline. Methods such as interpolation, resampling, or time shifting are used to perform time calibration, so that the calibrated datasets have a consistent time reference. After time calibration, the time alignment degree is verified by comparing the difference between the timestamps before and after calibration and calculating the alignment error. Based on the above steps, a fusion time alignment branch is established.

[0036] After the time alignment is completed, spatial registration is performed. First, a virtual spatial coordinate system is constructed. This coordinate system can be two-dimensional or three-dimensional depending on the spatial characteristics of the dataset. Then, the positions of the multiple regional datasets are scanned and registered. The spatial position information of the different datasets is aligned to the unified virtual spatial coordinate system using methods such as coordinate transformation, coordinate conversion, or spatial interpolation to generate a set of registered coordinates.

[0037] Next, based on the registered coordinate sets, the registered coordinate sets between different datasets are compared by calculating the Euclidean distance, Manhattan distance, or other appropriate measurement style between the coordinates, and the spatial offset between them is calculated. Statistical analysis is performed on the calculated spatial offsets to calculate statistics such as the mean, maximum, minimum, and standard deviation of the offsets to understand the distribution and degree of the offset. An appropriate spatial consistency threshold is set based on the analysis needs and data characteristics. The calculated spatial offsets are compared with the set threshold to evaluate the consistency in the spatial dimension of the multiple types of regional datasets. If the offset is equal to or less than the threshold, the spatial consistency is considered to meet the requirements; if not, the spatial registration method or parameters are readjusted.

[0038] After the spatial consistency verification is complete, a fusion spatial consistency branch is established to realize automatic spatial consistency and fusion of multiple regional datasets. When establishing the fusion spatial consistency branch, a fusion spatial consistency branch framework is established based on the spatial registration and verification process. The selected spatial registration algorithm (e.g., algorithms for feature extraction, registration transformation calculation, coordinate mapping, etc.) is integrated into the fusion spatial consistency branch. A scripting language such as Python is used to automatically load multiple regional datasets, extract features from the multiple regional datasets, perform spatial registration, and finally verify the spatial consistency and output the matching results, thereby realizing an automated process for the fusion spatial consistency branch. During the execution process, the spatial registration parameters and algorithms are continuously optimized based on the spatial consistency verification results to improve matching accuracy and efficiency. This establishes the fusion spatial consistency branch.

[0039] A data fusion channel is constructed based on the fused time alignment branch and the fused space alignment branch. In this channel, there is a connection order between the fused time alignment branch and the fused space alignment branch. For example, time alignment is performed first, followed by space alignment. First, the fused time alignment branch performs time calibration of the data to ensure consistency in the time dimension. After that, the time-aligned data is passed to the fused space alignment branch, where spatial registration is performed to ensure consistency in the space dimension. After processing by these two branches, the data has a unified time and space reference, providing a basis for subsequent data fusion and analysis.

[0040] The time and space alignment order is not fixed, but can be adjusted according to the specific application scenario and data characteristics.

[0041] Additionally, the method may further comprise: Sending the plurality of regional data sets to the fusion time alignment branch to generate time alignment information for the plurality of regional data sets; Sending the plurality of types of regional data sets to the fusion spatial alignment branch to generate spatial alignment information of the plurality of types of regional data sets; creating a data fusion strategy based on the time-aligned information and the spatial-aligned information; According to the data fusion strategy, weight the plurality of types of regional data sets to obtain a distribution weight set, where the distribution weight set corresponds to the plurality of types of regional data sets one-to-one, and the sum of the distribution weight set is equal to 1; The method includes performing initial data fusion on the plurality of types of regional datasets based on the distribution weight set in accordance with the data fusion strategy to generate the initial fused dataset.

[0042] In an embodiment of the present application, multiple types of regional datasets to be processed are sent to an established fusion time alignment branch. In the fusion time alignment branch, the datasets undergo steps such as timestamp standardization, time deviation detection, and time calibration, so that the time dimensions of different datasets are aligned into a unified timeline. After the time alignment process, this branch generates time alignment information, which includes the calibrated timestamps, time alignment error statistics, and time alignment degree evaluation results.

[0043] Multiple regional datasets are sent to the fusion spatial alignment branch, where the datasets undergo steps such as feature extraction, spatial registration, and spatial alignment verification to ensure that the spatial locations of different datasets correspond accurately. After the spatial alignment process is completed, this branch generates spatial alignment information, which includes the registered coordinate set, spatial offset statistics, and spatial alignment evaluation results.

[0044] The data fusion strategy is developed based on Bayesian estimation. The prior distribution of the parameters of each data source is defined based on historical data or other reliable information. The likelihood function corresponding to each data source is calculated based on the time consistency information and the space consistency information. Using Bayes' theorem, the posterior distribution of the parameters is calculated based on the prior distribution and the likelihood function. For multiple data sources, the posterior distributions corresponding to the data sources are fused by weighted average or product fusion. The fusion result is extracted from the posterior distribution after fusion.

[0045] Before weighting begins, preprocessing is first performed on the multiple regional datasets. Weighting principles are determined based on the data fusion strategy. These principles take into account factors such as data reliability, relevance, importance, and quality. For example, datasets with high temporal and spatial consistency are assigned higher weights, while datasets containing a lot of noise and outliers are given lower weights. According to the weighting principles, the weights for each dataset are calculated using methods such as analysis of variance and correlation analysis. After the above steps, a distribution weight set is generated. Each weight in the set corresponds to a specific multiple regional dataset, and the sum of these weights is equal to 1.

[0046] During the initial data fusion, the values ​​of each dataset are weighted based on a set of distribution weights. Specifically, for each data point or feature, its value is multiplied by the weight of the corresponding dataset, and then all weighted values ​​are summed to obtain the fused value. After the fusion calculations in the above steps, the initial fused dataset is obtained.

[0047] Furthermore, step S300 in the method provided by the embodiment of the present application performing data cleansing on the initial fusion dataset, and executing a plurality of data processing instructions based on the cleansing result, wherein the plurality of data processing instructions include missing value processing and outlier processing; generating a cleansed dataset through the missing value processing and the outlier processing; The method further includes integrating the cleansed datasets, performing data reduction on the integrated cleansed datasets, verifying the cleansed dataset based on the reduction result, updating the initial fusion dataset, and outputting it as the target fusion dataset.

[0048] In the present embodiment, data cleansing is first performed on the initial fusion dataset to correct errors and inconsistencies in the initial fusion dataset and improve data quality. Then, multiple data processing instructions are executed based on the cleansing results. Missing values ​​in the dataset are imputed using the mean, median, mode, or a specific value. Outliers in the dataset are identified using statistical methods, such as Z-score, and the outliers are removed and replaced with the mean or median according to the actual situation. After missing value processing and outlier processing, a cleansed dataset is obtained.

[0049] When integrating cleansed datasets, the first step is to unify the data formats, ensuring that all datasets are consistent in terms of field names, data types, data formats, etc. Second, conflicting data is handled, i.e., when conflicts exist for the same field in different datasets, the value of one data source is retained through voting or weighting methods. Finally, duplicate data is removed during the integration process, ensuring that the integrated dataset does not contain redundant information.

[0050] When performing data reduction, the most meaningful features are selected for subsequent analysis based on the importance and relevance of the data, and features that do not significantly affect the analysis results are removed. Dimensionality reduction techniques, such as principal component analysis and linear discriminant analysis, are used to convert high-dimensional feature spaces into low-dimensional feature spaces. If the data set is too large, sampling techniques such as random sampling and stratified sampling can be used to select some representative data, reducing the complexity of data processing and analysis.

[0051] The cleansed dataset is then verified based on the reduction results to check the completeness of the data and ensure that the reduced dataset does not lose any important information or features. Next, statistical methods are used to compare the datasets before and after reduction to evaluate the impact of the reduction process on the data analysis results. Finally, based on actual needs, it is confirmed whether the reduced dataset still meets the analysis needs. After verifying the above steps, if the quality and accuracy of the reduced cleansed dataset are confirmed to meet the requirements, it is updated as the initial fusion dataset and output as the target fusion dataset.

[0052] Furthermore, the method comprises: Obtaining a target analysis result, a background analysis result, and a noise analysis result based on the target fusion dataset; inputting the target analysis result, the background analysis result, and the noise analysis result into a decision unit, and obtaining a discrimination calibration analysis result based on the decision unit, wherein the discrimination calibration analysis result includes the target analysis result or / and the background analysis result or / and the noise analysis result; The method further includes generating supervised labels using the discriminant calibration analysis results, and identifying anomalous points in the target fusion dataset based on the supervised labels.

[0053] In this embodiment, feature extraction is first performed on the target fusion dataset to extract features related to light vehicles, including the shape, size, speed, and movement trajectory of the light vehicles. Next, classification and identification are performed using the extracted features to select data corresponding to light vehicles. The classification and identification results are integrated to form target analysis results, which include information such as the location, quantity, and movement status of the light vehicles.

[0054] The background analysis result refers to other environmental information unrelated to the target. Environment-related features, such as road conditions, buildings, vegetation, etc., are extracted from the target fusion dataset. The extracted environment features are integrated to form the background analysis result.

[0055] The noise analysis result is an analysis of abnormal or erroneous data in the target fusion dataset. This noise can be caused by sensor failure or data transmission errors. Statistical methods are used to detect abnormal values ​​and erroneous data in the target fusion dataset. The detected abnormal values ​​are classified and identified to determine whether they are noise and analyze their causes. The results of noise classification and identification are combined to form the noise analysis result.

[0056] The target analysis result, background analysis result, and noise analysis result are input to a classifier, which then comprehensively analyzes and judges these results based on preset rules to obtain a discrimination / calibration analysis result. In this process, the classifier calibrates the target analysis result based on the background analysis result or removes erroneous data based on the noise analysis result. The final discrimination / calibration analysis result includes the calibrated target analysis result, background analysis result, and / or noise-removed data.

[0057] To generate a surveillance label, the system first analyzes the identification and calibration analysis results to identify key data contained in the results, such as the type of light vehicle, its location, speed, and movement trajectory. Based on the analyzed identification and calibration analysis results, the system generates a corresponding surveillance label. The surveillance label is an abstract representation of the vehicle's status, which can reflect the vehicle's real-time status. For example, a label containing information such as its unique identifier, location, speed, and status can be generated for each light vehicle.

[0058] After the monitoring labels are obtained, anomalous points are identified in the target fusion dataset based on the monitoring labels. An anomalous point refers to a data point that does not match the normal operating state of the light vehicle, and is caused by a sensor failure, a data transmission error, or other causes. When identifying an anomalous point, a judgment is made based on the information in the monitoring labels and a pre-set rule. For example, a speed threshold is set, and if the speed of the light vehicle exceeds that threshold, it is considered to be an anomalous state, and the corresponding data point is identified as an anomalous point.

[0059] Furthermore, the method comprises: Perform training based on a training operator to obtain the classifier, where the training operator includes a plurality of sets of training samples, where each set of training samples includes a predetermined target sample, a background sample, a noise sample and a test sample; The method further includes obtaining a discrimination error accuracy based on the determiner, and activating the determiner to analyze the target fusion dataset if the discrimination error accuracy is smaller than a preset error accuracy.

[0060] In an embodiment of the present application, the training operator includes multiple sets of training samples, each set of training samples including a predetermined target sample, a background sample, a noise sample, and a test sample. The predetermined target sample is a light vehicle target identified and extracted from the target fusion dataset. The predetermined background sample represents background information unrelated to the target in the target fusion dataset. The predetermined noise sample simulates abnormal or erroneous data that may be included in the target fusion dataset, including noise data caused by sensor failure, data transmission errors, etc. The test sample is used to evaluate the performance of the classifier during the training process.

[0061] Before training begins, an appropriate model structure is selected. In this application, a convolutional neural network (CNN) model is selected, and the model is repeatedly trained using samples in the training operator. In each iteration, a set of samples is randomly selected from the training operator and input into the model. The difference between the model output and the actual label, i.e., the loss function, is calculated. Then, an optimization algorithm such as gradient descent is used to update the model parameters to minimize the loss function. This process is repeated multiple times until the model's performance meets the preset requirements or the preset number of iterations is reached. After training is completed, a classifier is obtained.

[0062] The classifier is then applied to an independent validation sample set to obtain the classification error accuracy. The validation sample set must contain data with known labels, and the classification error accuracy refers to the ratio of the number of samples correctly classified by the classifier to the total number of samples. If the classification error accuracy falls below a preset error accuracy threshold, the classifier's performance is deemed to meet the requirements, and it can be launched to analyze the target fusion dataset.

[0063] If the classification error accuracy of the classifier meets the requirement, the classifier is activated to analyze the target fusion dataset.

[0064] In summary, the embodiments of the present application have at least the following technical effects: The present application configures a sensor group based on multiple sensors, collects data based on a target range using the sensor group, generates multiple types of area datasets, transmits the multiple types of area datasets to a data fusion channel to generate an initial fusion dataset, synchronizes the initial fusion dataset with a data pre-processing unit for pre-processing, updates the initial fusion dataset, generates a target fusion dataset, uses a feature extraction unit to scan the target fusion dataset, extracts features of the target light vehicle, generates a target feature information set, configures a target identification unit based on the target feature information set, and uses the target identification unit to intelligently identify the target fusion dataset. The present invention solves the technical problem of data redundancy and collision between different sensors, which makes data processing complicated in the prior art, and achieves the technical effect of improving the accuracy and stability of identification precision by fusing data from different sensors.

[0065] Example 2 Based on the same inventive idea as the light vehicle identification method based on the cooperation of multiple sensors in the above embodiment, as shown in FIG. 2 , the present application provides a light vehicle identification system based on the cooperation of multiple sensors, where the system in the embodiment of the present application is based on the same inventive idea as the method embodiment, and the system includes the following modules: The multiple types of area data set generation module 11 constructs a sensor group based on multiple sensors, collects data based on a target range using the sensor group, and generates multiple types of area data sets. The initial fusion data set generation module 12 transmits the multiple types of regional data sets to a data fusion channel to generate an initial fusion data set. The target fusion dataset generation module 13 synchronizes the initial fusion dataset with the data pre-processing unit to perform pre-processing, update the initial fusion dataset, and generate a target fusion dataset. The target feature information set generation module 14 uses a feature extraction unit to scan the target fusion dataset, extract features of the target light vehicle, and generate a target feature information set, where there is a correspondence between the target feature information set and the target fusion dataset. The intelligent identification module 15 constructs a target identification unit based on the target feature information set, and uses the target identification unit to intelligently identify the target fusion data set.

[0066] Furthermore, the system Based on the monitoring needs information, the plurality of sensors are arranged within a target range, and the sensors in the same area are associated with each other through communication, and based on the associated information, the sensor group is constructed; collecting data from a target area with the sensor group to generate a plurality of area data sets; The method further includes performing a cluster analysis on the plurality of regional data sets to generate the plurality of types of regional data sets.

[0067] Furthermore, the system Standardizing timestamps for the plurality of types of area data sets, detecting time deviations of these area data sets based on the standardized timestamps, and generating time deviation data; Perform time calibration according to the time deviation data, verify the time consistency according to the calibration result, and establish a fusion time consistency branch; constructing a virtual spatial coordinate system, scanning the plurality of types of area data sets to perform positional registration, and generating a registration coordinate set; verifying spatial consistency based on the registration coordinate set and establishing a fusion spatial consistency branch; The method further includes constructing the data fusion channel based on the fused time-aligned branch and the fused space-aligned branch, where the fused time-aligned branch and the fused space-aligned branch have a connection order.

[0068] Furthermore, the system Sending the plurality of regional data sets to the fusion time alignment branch to generate time alignment information for the plurality of regional data sets; Sending the plurality of types of regional data sets to the fusion spatial alignment branch to generate spatial alignment information of the plurality of types of regional data sets; creating a data fusion strategy based on the time-aligned information and the spatial-aligned information; According to the data fusion strategy, weight the plurality of types of regional data sets to obtain a distribution weight set, where the distribution weight set corresponds one-to-one with the plurality of types of regional data sets, and the sum of the distribution weight set is equal to 1; The method further includes performing initial data fusion on the plurality of types of regional datasets based on the distribution weight set according to the data fusion strategy to generate the initial fused dataset.

[0069] Furthermore, the system performing data cleansing on the initial fusion dataset, and executing a plurality of data processing instructions based on the cleansing result, wherein the plurality of data processing instructions include missing value processing and outlier processing; generating a cleansed dataset through the missing value processing and the outlier processing; The method further includes integrating the cleansed datasets, performing data reduction on the integrated cleansed datasets, verifying the cleansed dataset based on the reduction result, updating the initial fusion dataset, and outputting it as the target fusion dataset.

[0070] Furthermore, the system Obtaining a target analysis result, a background analysis result, and a noise analysis result based on the target fusion dataset; inputting the target analysis result, the background analysis result, and the noise analysis result into a decision unit, and obtaining a discrimination calibration analysis result based on the decision unit, wherein the discrimination calibration analysis result includes the target analysis result or / and the background analysis result or / and the noise analysis result; The method further includes generating supervised labels using the discriminant calibration analysis results, and identifying anomalous points in the target fusion dataset based on the supervised labels.

[0071] Furthermore, the system Perform training based on a training operator to obtain the classifier, where the training operator includes a plurality of sets of training samples, each set of training samples including a predetermined target sample, a background sample, a noise sample and a test sample; The method further includes obtaining a discrimination error accuracy based on the determiner, and activating the determiner to analyze the target fusion dataset if the discrimination error accuracy is smaller than a preset error accuracy.

[0072] It should be noted that the order of the embodiments of the present application described above is for illustrative purposes only and does not imply superiority or inferiority of the embodiments. Also, the above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some circumstances, the actions or steps recited in the claims may be performed in an order different from that shown in the embodiments and still achieve the desired results. Furthermore, processes depicted in the figures may achieve the desired results outside the particular order or sequential order shown. In some embodiments, multitasking or parallel processing may also be possible or advantageous.

[0073] The above are only preferred embodiments of the present invention, and do not limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall fall within the protection scope of this application.

[0074] The specification and drawings are merely illustrative of the present application and are considered to cover any modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technology, the present application shall include these modifications and variations. [Explanation of symbols]

[0075] 11. Multiple-area dataset generation module 12 Initial fusion dataset generation module 13 Target Fusion Dataset Generation Module 14 Target feature information set generation module 15 Intelligent Identification Module S100, S200, S300, S400, S500 steps

Claims

1. A light vehicle identification method based on cooperation of multiple sensors, comprising: a region data set generation step of constructing a sensor group based on a plurality of sensors, collecting data based on a target range using the sensor group, and generating a plurality of types of region data sets; a fusion dataset generation step of transmitting the plurality of types of regional datasets to a data fusion channel to generate an initial fusion dataset; a target fusion dataset generation step of synchronizing the initial fusion dataset with a data pre-processing unit to perform pre-processing on the initial fusion dataset to update the initial fusion dataset and generate a target fusion dataset; a target feature information set generating step of using a feature extraction unit to scan the target fusion dataset to extract features of a target light vehicle and generate a target feature information set, where there is a correspondence between the target feature information set and the target fusion dataset; and constructing a target identification unit based on the target feature information set, and intelligently identifying the target fusion data set by the target identification unit; The target fusion dataset generation step includes: performing data cleansing on the initial fusion dataset, and executing a plurality of data processing instructions based on the cleansing result, wherein the plurality of data processing instructions include missing value processing and outlier processing; generating a cleansed dataset through the missing value processing and the outlier processing; aggregating the cleansed datasets, performing data reduction on the combined cleansed datasets, validating the cleansed dataset based on the reduction results, and updating the initial fusion dataset to output as the target fusion dataset.

2. obtaining a target analysis result, a background analysis result, and a noise analysis result based on the target fusion dataset; inputting the target analysis result, the background analysis result, and the noise analysis result into a decision unit, and obtaining a discrimination calibration analysis result based on the decision unit, wherein the discrimination calibration analysis result includes the target analysis result, or / and the background analysis result, or / and the noise analysis result; The method of claim 1 , further comprising generating supervised labels using the discriminant calibration analysis results, and identifying anomalous points in the target fusion dataset based on the supervised labels.

3. Perform training based on a training operator to obtain the classifier, where the training operator includes a plurality of sets of training samples, each set of training samples including a predetermined target sample, a background sample, a noise sample and a test sample; The method of claim 2 , further comprising: obtaining a classification error accuracy based on the determiner; and activating the determiner to analyze a target fusion dataset when the classification error accuracy is smaller than a preset error accuracy.

Images