Traffic fusion data acquisition processing method and data acquisition system

Abstract

The invention relates to a traffic fusion data acquisition and processing method and a data acquisition system. The method comprises the following steps: acquiring multi-dimensional multi-path signal data acquired by an intelligent lamp post integration system; carrying out data preprocessing on the collected original data; for the preprocessed data, an orthogonal rational wavelet group filter bank is matched with a sensor array and a network to carry out multipath separation on multipath signal data to form multi-dimensional wavelet filter array data; and performing data fusion on the separated multi-path signal data, and performing optimization in the range of the whole network by adopting a wavelet probabilistic neural network model to obtain a numerical value closest to a real value. Compared with the prior art, the method has the advantages that the problem of multi-path separation is solved, multi-dimensional multi-path signals are fused and optimized in the range of the whole network, data accuracy is improved, and urban managers are effectively helped to detect traffic conditions and deal with possible problems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle-road cooperation data processing, and in particular to a traffic fusion data acquisition and processing method and a data acquisition system. BACKGROUND

[0002] With the acceleration of urbanization and the rapid increase in the number of cars, traffic congestion, accidents and other problems are becoming increasingly prominent. In order to cope with these challenges, a scheme based on vehicle-road cooperation technology is adopted, through information exchange between vehicles and road infrastructure, to realize the optimization and safety management of traffic flow.

[0003] In the vehicle-road cooperation system, the intelligent lamp pole is an important part of the road infrastructure, not only with the traditional lighting function, but also through the integration of various sensors, communication equipment and data processing units, to realize the real-time acquisition, transmission and processing of traffic information. However, the existing intelligent lamp poles still have deficiencies in function integration and data fusion, such as scattered sensor data, serious information island phenomenon, limited data processing capacity, etc., which leads to the fact that their role in the vehicle-road cooperation system cannot be fully played. SUMMARY

[0004] The purpose of the present application is to overcome the deficiencies of the existing intelligent lamp poles in function integration and data fusion and to provide a traffic fusion data acquisition and processing method and a data acquisition system.

[0005] The purpose of the present application can be achieved by the following technical solutions:

[0006] As a first aspect of the present application, a traffic fusion data acquisition and processing method is provided, the method steps comprising:

[0007] acquiring multi-dimensional multi-path signal data collected by an intelligent lamp pole integrated system;

[0008] performing data preprocessing on the collected raw data;

[0009] For the preprocessed data, a multi-path separation is performed on the multi-path signal data using an orthogonal rational wavelet filter bank in cooperation with a sensor array and a network, to form a multi-dimensional wavelet filter array data;

[0010] After data fusion of the separated multi-path signal data, a wavelet probabilistic neural network model is used to optimize within the entire network and obtain a value closest to the true value.

[0011] As a preferred technical solution, the multi-path separation of the multi-path signal data is performed according to the following steps:

[0012] According to the characteristics of the collected signals, a mother wavelet function is selected, and then a corresponding orthogonal rational wavelet filter bank is designed according to the requirements of orthogonality and rationality;

[0013] Based on the designed orthogonal rational wavelet filter bank, multi-dimensional and multi-path signal data collected by the intelligent lamp pole integrated system are subjected to wavelet decomposition, features in different scales and directions are extracted, and reserved wavelet coefficients are selected according to application requirements;

[0014] Multi-scale features are extracted from the wavelet coefficients, and directional features are extracted by using the directionality of the wavelet transform.

[0015] As a preferred technical solution, the principle for selecting the mother wavelet function comprises: frequency characteristics of the signal, smoothness of the signal, boundary characteristics of the signal, computational complexity, and boundary characteristics of the signal.

[0016] As a preferred technical solution, the design steps of the filter bank are as follows:

[0017] According to the characteristics of the signal and the analysis requirements, a suitable wavelet order is selected; wavelet polynomial zero points are solved; according to the solved zero points, low-pass filter and high-pass filter coefficients of the filter bank are calculated; and orthogonality and completeness of the filter bank are verified.

[0018] As a preferred technical solution, the data fusion of the separated multi-path signal data is specifically as follows:

[0019] The corresponding wavelet coefficients of each layer of signals collected by different sensors are directly added to obtain a new set of wavelet coefficients C new , and the fusion signal is reconstructed from the new wavelet coefficients;

[0020] Fisher values of wavelet coefficients of two kinds of signals are calculated respectively, and the wavelet coefficients with the highest Fisher value in each layer are selected as the wavelet coefficients with the best separability to reconstruct the fusion signal:

[0021]

[0022] Wherein, F represents the Fisher value;

[0023] Based on the Fisher value weighting and wavelet reconstruction fusion rule, Fisher values of wavelet coefficients of two kinds of signals are calculated, and weights W allocated to wavelet coefficients in each layer are calculated:

[0024]

[0025] New wavelet coefficients are obtained by using the fusion rule:

[0026] C new =W1C1+W2C2

[0027] The fusion signal is obtained through wavelet reconstruction.

[0028] As a preferred technical solution, the wavelet probabilistic neural network model is built according to data characteristics,

[0029] The input parameters of the wavelet probabilistic neural network model include: input original signals or data sets; a wavelet function, used for wavelet transform of the input data and selection of the number of wavelet decomposition layers;

[0030] The output parameters of the wavelet probabilistic neural network model include: a classification result output by the model, corresponding to a category to which the input data belongs and a probability of each category, indicating a possibility of the input data belonging to each category;

[0031] For the prediction result of the signal fusion and optimization combined model, a final prediction result is obtained through a voting mechanism.

[0032] As a preferred technical solution, the wavelet probabilistic neural network model structure includes the following parts:

[0033] A wavelet transform layer: performing wavelet decomposition on the input data to generate multi-scale and multi-directional wavelet coefficients;

[0034] A feature extraction layer: extracting important features from the wavelet coefficients to form a feature vector;

[0035] A probabilistic neural network layer includes:

[0036] An input layer: receiving the feature vector extracted from the wavelet transform layer;

[0037] A pattern layer: calculating the Euclidean distance between the input features and the training samples, and calculating the similarity according to a Gaussian kernel function;

[0038] A summation layer: summing up the similarity of each category to obtain the total similarity of each category;

[0039] An output layer: calculating the probability of the input data belonging to each category according to the total similarity of each category, and outputting a classification result.

[0040] As a preferred technical solution, the data acquisition and processing method acquires, transmits and stores data as follows:

[0041] Network transmission and storage: compressing the data using the sparsity of wavelet transform, and encoding the compressed data for network transmission;

[0042] Data recovery: performing inverse transform on the received wavelet coefficients to recover the original signal, and performing enhancement processing on the recovered signal.

[0043] As a second aspect of the present application, a traffic fusion data acquisition system based on intelligent lamp pole is provided, comprising a vehicle-mounted device, an intelligent lamp pole and a central control platform;

[0044] The vehicle-mounted device comprises a vehicle-mounted MEC unit and a vehicle-mounted Bluetooth device arranged on the vehicle, and is used for interacting with the intelligent lamp pole.

[0045] The intelligent lamp pole is built by means of pole combination and box combination, and comprises a multi-sensor data acquisition device, a perception camera, a road side unit (RSU) and a router, which are used for interacting with the vehicle-mounted device and the central control platform.

[0046] The central control platform comprises a data fusion unit and a human-computer interaction interface, and the central control platform interacts with the intelligent lamp pole, and the data fusion unit executes the traffic fusion data acquisition and processing method as described above.

[0047] As a preferred technical solution, the data fusion unit is based on a data fusion method of orthogonal rational wavelet family, and the collected original data is preprocessed; then a filter bank of orthogonal rational wavelet family is constructed in cooperation with a sensor array and a network to form a multi-dimensional wavelet filter array data, and a wavelet probabilistic neural network model is built to fuse and optimize the multi-dimensional multi-path signals in the entire network range to obtain a value closest to the true value.

[0048] Compared with the prior art, the present application has the following beneficial effects:

[0049] 1) The data fusion method based on the orthogonal rational wavelet family is used to construct a filter bank of orthogonal rational wavelet family in cooperation with a sensor array and a network to form a multi-dimensional wavelet filter array data, so as to solve the problem of multi-path separation.

[0050] 2) The wavelet probabilistic neural network model is built to fuse and optimize the multi-dimensional multi-path signals in the entire network range, to perform de-redundancy processing on the data, reduce the network traffic, reduce the network energy consumption and improve the energy utilization rate.

[0051] 3) The intelligent lamp pole built by means of pole combination and box combination integrates the multi-sensor data acquisition device, the perception camera, the road side unit (RSU) and the router in the intelligent lamp pole, and realizes the transmission of multiple data parameters through the interaction with the vehicle and the central platform. Through the design of pole combination and box combination, the intelligent lamp pole can effectively integrate the traffic poles, monitoring poles, road signs and other infrastructures on the road, reduce the number of installations, save the urban road space and improve the urban landscape aesthetics. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a traffic fusion data acquisition and processing method flowchart of the present application.

[0053] Figure 2 A flow framework schematic diagram for collecting multi-dimensional data and performing multi-path separation and fusion according to the present application;

[0054] Figure 3 A framework diagram of a data collection system according to the present application. DETAILED DESCRIPTION

[0055] The present application will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments are implemented on the premise of the technical solutions of the present application, and detailed implementation manners and specific operation processes are given, but the protection scope of the present application is not limited to the following embodiments.

[0056] Embodiment 1

[0057] As one of the embodiments of the present application, the embodiment provides a traffic fusion data collection and processing method, a data fusion method based on an orthogonal rational wavelet family, which removes noise, abnormal values and repeated values by cleaning the collected original data, ensures the accuracy and reliability of the data, and then constructs an orthogonal rational wavelet filter bank in cooperation with a sensor array and a network to form a multi-dimensional wavelet filter array data, solves the problem of multi-path separation, and optimizes the multi-dimensional multi-path signal in the entire network to obtain a value closest to the true value. The specific step process is as shown in Figure 1

[0058] S1, data collection, selecting a sensor type, since the intelligent lamp pole integrated system will be applied to an intersection, selecting a perception camera to obtain intuitive picture videos, vehicle shape features and other information, using a road side unit RSU and a router to realize the interconnection of devices, vehicles and a central platform, and performing low-latency reliable communication. Based on the sensor, multi-dimensional multi-path signal data such as time, space, user behavior, sensor data, etc. is obtained.

[0059] S2, data preprocessing, removing noise, processing missing values and abnormal values from the collected data, and aligning data from different sources according to timestamps or other key indicators. At the same time, a standard is preset, and the data is standardized or normalized to facilitate subsequent processing.

[0060] S3, using an orthogonal rational wavelet filter bank in cooperation with a sensor array and a network to form a multi-dimensional wavelet filter array data to realize multi-path separation:

[0061] Wavelet filter bank design, selecting a suitable mother wavelet function according to the signal characteristics, and then designing a corresponding filter bank according to the orthogonality and rationality requirements.

[0062] ​The principles for selecting the mother wavelet function include 1) frequency characteristics of the signal, 2) smoothness of the signal, 3) boundary characteristics of the signal, 4) computational complexity, and 5) boundary characteristics of the signal. For example, for a relatively smooth signal, a wavelet with a small vanishing moment is selected, such as a high-order Daubechies wavelet (DB wavelet); for a non-smooth signal, a wavelet with a large vanishing moment is selected, such as a low-order Daubechies wavelet (DB wavelet); for a signal with boundary effects, a wavelet with good boundary processing is selected, such as a Symlets wavelet; for a real-time processing requirement, a wavelet with high computational efficiency is selected, such as a Haar wavelet. For traffic flow signals, which are high-frequency abrupt signals and may suddenly change, the Daubechies wavelet (DB wavelet) has good time-domain localization capability and can effectively capture the abrupt characteristics of the signal. For traffic noise signals, which are stationary signals, the Symlets wavelet has good smoothness and is suitable for stationary signal analysis.

[0063] Specific design of the filter bank:

[0064] Determining the wavelet order: according to the signal characteristics and analysis requirements, a suitable wavelet order is selected. For example, the 4th order Daubechies wavelet (db4) is selected;

[0065] Solving the wavelet polynomial zero: using Matlab, the zero of the wavelet polynomial is solved. For example, the Matlab function roots is used to solve it;

[0066] Calculating the filter bank coefficients: according to the solved zero, the low-pass filter and high-pass filter coefficients of the filter bank are calculated. The Matlab function wfilters can be used to obtain the filter bank coefficients.

[0067] Verify the orthogonality and completeness of the filter bank: through theoretical derivation or numerical verification, ensure that the filter bank meets the requirements of orthogonality and completeness.

[0068] Multi-dimensional wavelet filtering: multi-dimensional data collected by the sensor is decomposed by wavelet, and features of different scales and directions are extracted. According to the application requirements, the retained wavelet coefficients are selected.

[0069] Data processing and separation: multi-scale features are extracted from wavelet coefficients, and directional features are extracted using the directionality of wavelet transform.

[0070] S4, data fusion: integrating spatial and temporal data, constructing spatio-temporal features, and extracting and constructing useful features to enhance the prediction ability of the model. At the same time, data fusion is performed on multi-path signals, and multi-dimensional wavelet coefficients of different sensors are fused to form a unified feature description:

[0071] The corresponding wavelet coefficients of each layer of signals collected by different sensors are directly added to obtain a new set of wavelet coefficients:

[0072] C new = C1 + C2

[0073] Then the fusion signal is reconstructed by the new wavelet coefficients, where C represents the wavelet coefficients.

[0074] The Fisher values of the wavelet coefficients of the two signals are calculated respectively, and the wavelet coefficients with the highest Fisher value in each layer are selected as the best discriminant wavelet coefficients:

[0075]

[0076] The wavelet coefficients of the fusion signal are reconstructed, where F represents the Fisher value.

[0077] Based on the Fisher value weighting and wavelet reconstruction fusion rule, the Fisher values of the wavelet coefficients of the two signals are first calculated, and then the weight W assigned to each layer wavelet coefficient is calculated by the following formula:

[0078]

[0079] The new wavelet coefficients are obtained using the fusion rule:

[0080] C new = W1C1 + W2C2

[0081] Finally, the fusion signal is obtained by wavelet reconstruction. The wavelet transform fusion example is shown in Figure 2 ;

[0082] Fisher criterion is a statistical parameter that projects high-dimensional parameters into one-dimensional space to measure the discriminant properties of sample classes. Fisher value represents the separability of two classes of samples, and the larger the Fisher value, the stronger the separability.

[0083] S5, model construction and optimization, according to the data characteristics, build a wavelet probabilistic neural network model, first use the training data set to train the model, then evaluate the model performance through cross-validation method to avoid overfitting. Subsequently, use grid search, random search or Bayesian optimization method to adjust the model hyperparameters to optimize the model performance.

[0084] The input parameters of the wavelet probabilistic neural network model include:

[0085] Raw data: input raw signal or data set, which can be time series data, image data or other forms of multi-dimensional data.

[0086] Wavelet function: select appropriate wavelet function (such as Daubechies, Haar, etc.) to perform wavelet transform on the input data.

[0087] Decomposition level: Select the number of wavelet decomposition levels.

[0088] Output parameters include:

[0089] Classification label: The classification result output by the model, corresponding to the category to which the input data belongs.

[0090] Probability distribution: Output the probability of each category, indicating the likelihood of the input data belonging to each category.

[0091] Wavelet probabilistic neural network model structure includes the following main parts:

[0092] 1) Wavelet transform layer: Wavelet decomposition of input data, generating multi-scale, multi-directional wavelet coefficients.

[0093] 2) Feature extraction layer: Extract important features from wavelet coefficients to form feature vectors.

[0094] 3) Probabilistic neural network layer contains:

[0095] 3.1) Input layer: Receive feature vectors extracted from wavelet transform layer.

[0096] 3.2) Pattern layer: Calculate the Euclidean distance between input features and training samples, and calculate the similarity according to the Gaussian kernel function.

[0097] 3.3) Summation layer: Sum the similarity of each category to get the total similarity of each category.

[0098] 4) Output layer: Calculate the probability of input data belonging to each category according to the total similarity of each category, and output the classification result.

[0099] Signal fusion and optimization, combine the prediction results of the model, get the final prediction result through the voting mechanism.

[0100] Result verification and evaluation, use accuracy, mean square error, R square value and other indicators to evaluate the prediction effect of the model, analyze the difference between the prediction result and the true value, find out the error source. Finally, adjust and correct the model according to the error analysis result.

[0101] S6 deploy the final optimized model to the application scenario to process real-time data.

[0102] In the method of the present application, the transmission and storage of sensor collected data is as follows:

[0103] Network transmission and storage, use the sparsity of wavelet transform to compress data, and encode the compressed data for network transmission.

[0104] Data transmission, selecting appropriate transmission protocols to ensure data integrity and transmission efficiency, and storing received data for subsequent processing and analysis.

[0105] Data recovery, inverse transform of received wavelet coefficients, recovery of original signal, and enhancement processing of recovered signal to improve signal-to-noise ratio.

[0106] Example 2

[0107] As another embodiment of the present application, the embodiment also provides a traffic integration data collection and processing system based on integrated intelligent lamp pole, which comprises a vehicle-mounted device, an intelligent lamp pole and a central control platform, wherein:

[0108] The vehicle-mounted device is composed of a vehicle-end MEC unit and a vehicle-mounted Bluetooth device;

[0109] The intelligent lamp pole is built by the method of pole and box combination, and contains a multi-sensor data collection device, a perception camera, a roadside unit RSU and a router;

[0110] The central control platform contains a data fusion unit and a human-computer interaction interface.

[0111] The intelligent lamp pole integrated system based on data fusion also includes a positioning system, which uses the Beidou satellite navigation system.

[0112] The vehicle-end MEC unit is composed of a server, a storage device and a network device, supports edge computing and collaborative computing mode, can receive data from vehicle-mounted sensors, cameras and other devices, and perform real-time processing and analysis to extract valuable information. At the same time, it can also communicate and process cooperatively with roadside facilities to realize data exchange and resource sharing.

[0113] The vehicle-mounted Bluetooth device is composed of a Bluetooth module and a control unit, supports OBD (On-Board Diagnostics) communication mode, allows the vehicle-mounted Bluetooth device to receive and display vehicle fault information and performance data parameters. The vehicle-mounted Bluetooth device and the intelligent lamp pole can realize real-time information transmission through Bluetooth technology. This interaction allows the vehicle to send data to the lamp pole, and also receives information from the lamp pole.

[0114] The intelligent lamp pole using the pole and box combination method:

[0115] The multi-sensor data collection device is composed of a sensor interface, a data collector, a wireless connection module, a power management system, a shell and a protection structure, supports continuous collection mode, timing collection mode, trigger collection mode and remote control mode, and can perform real-time monitoring, data analysis and prediction, automatic control and remote monitoring and management.

[0116] The integrated multi-sensor device not only collects multiple target parameters, but also collects the same parameter by different sensors, thereby generating redundant data. This redundancy helps to verify the accuracy and reliability of the data. If the data of different sensors is consistent within a reasonable error range, the data is considered accurate; if the data difference is large, further checking and calibration can be performed by the subsequent data fusion unit.

[0117] The roadside unit RSU is composed of a high-gain directional beam control read-write antenna and a radio frequency controller, and supports a communication mode based on dedicated short-range communication (DSRC). DSRC is a wireless communication technology based on the IEEE 802.11p standard, designed for communication between high-speed moving vehicles and roadside facilities. The RSU communicates with the OBU through DSRC, and can obtain real-time data such as the driving state, position information, and identity information of the vehicle.

[0118] The data fusion unit is composed of a data fusion algorithm based on the orthogonal rational wavelet family. By cleaning the collected raw data, removing noise, outliers and repeated values, the accuracy and reliability of the data are ensured. Subsequently, the constructed orthogonal rational wavelet family filter bank cooperates with the sensor array and network to form a multi-dimensional wavelet filter array data, solves the problem of multi-path separation, and optimizes the multi-dimensional multi-path signal in the entire network range. The performance of the entire data fusion method is greatly improved, so as to obtain the value closest to the true value and improve the accuracy of the data.

[0119] Through the construction method of combining the rod and the box and improving the existing rod, the rod reduction rate can be improved, and the urban management cost can be reduced. The present application can present the prediction results of the traffic status and the future traffic status in the man-machine interface with friendliness and operation convenience, accurately obtain the traffic target information, improve the road traffic safety, effectively help the city managers to detect the traffic status, and cope with possible problems.

[0120] The above detailed the preferred embodiments of the present application. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment based on the prior art according to the concept of the present application shall be within the protection scope determined by the claims.

Claims

1. A traffic fusion data collection processing method, characterized in that, The method steps comprise: acquiring multi-dimensional multi-path signal data collected by the intelligent lamp pole integrated system; performing data preprocessing on the collected original data; for the preprocessed data, using an orthogonal rational wavelet filter bank to separate the multi-path signal data, and constructing a multi-dimensional wavelet filter array data; after data fusion of the separated multi-path signal data, using a wavelet probabilistic neural network model to optimize the entire network and obtain a value closest to the true value.

2. The traffic fusion data collection and processing method of claim 1, wherein, The specific steps for separating the multi-path signal data are as follows: selecting a mother wavelet function according to the characteristics of the collected signal, and then designing a corresponding orthogonal rational wavelet filter bank according to the requirements of orthogonality and rationality; based on the designed orthogonal rational wavelet filter bank, performing wavelet decomposition on the multi-dimensional multi-path signal data collected by the intelligent lamp pole integrated system, extracting features of different scales and directions, and selecting the retained wavelet coefficients according to application requirements; extracting multi-scale features from the wavelet coefficients and directional features using the directionality of wavelet transform.

3. The traffic fusion data collection and processing method of claim 2, wherein, The principles for selecting the mother wavelet function include: frequency characteristics of the signal, smoothness of the signal, boundary characteristics of the signal, computational complexity, and boundary characteristics of the signal.

4. The traffic fusion data collection and processing method of claim 2, wherein, The design steps of the filter bank are as follows: selecting an appropriate wavelet order according to the characteristics of the signal and analysis requirements; solving the wavelet polynomial zero point; calculating the low-pass filter and high-pass filter coefficients of the filter bank according to the solved zero point; verifying the orthogonality and completeness of the filter bank.

5. The traffic fusion data collection and processing method of claim 2, wherein, The data fusion of the separated multi-path signal data is as follows: The corresponding wavelet coefficients of each layer of signals collected by different sensors are directly added to obtain a new set of wavelet coefficients C new , and a fusion signal is reconstructed from the new set of wavelet coefficients. calculate the Fisher values of the wavelet coefficients of the two signals respectively, and select the wavelet coefficients with the highest Fisher value at each layer as the best wavelet coefficients for reconstruction: where F represents the Fisher value; based on the Fisher value weighting and wavelet reconstruction fusion rule, calculate the Fisher values of the wavelet coefficients of the two signals, and calculate the weight W allocated to each layer wavelet coefficient: use the fusion rule to obtain new wavelet coefficients: C new = W1C1 + W2C2 obtain the fusion signal through wavelet reconstruction.

6. The traffic fusion data collection and processing method of claim 1, wherein, The wavelet probabilistic neural network model is built according to the data characteristics, the input parameters of the wavelet probabilistic neural network model include: input original signal or data set; wavelet function, used for wavelet transform of the input data and selection of wavelet decomposition layer number; the output parameters of the wavelet probabilistic neural network model include: classification results output by the model, corresponding to the categories to which the input data belongs and the probability of each category, indicating the possibility of the input data belonging to each category; for the prediction results of the signal fusion and optimization combined model, the final prediction results are obtained through a voting mechanism.

7. The traffic fusion data collection and processing method of claim 1, wherein, The structure of the wavelet probabilistic neural network model includes the following parts: wavelet transform layer: performing wavelet decomposition on the input data to generate multi-scale and multi-directional wavelet coefficients; feature extraction layer: extracting important features from the wavelet coefficients to form a feature vector; the probabilistic neural network layer includes: input layer: receiving the feature vector extracted from the wavelet transform layer; Mode layer: calculate the Euclidean distance between the input features and the training samples, and calculate the similarity according to the Gaussian kernel function; Summation layer: sum the similarity of each category to obtain the total similarity of each category; Output layer: calculate the probability of input data belonging to each category according to the total similarity of each category, and output the classification result.

8. The traffic fusion data collection and processing method of claim 1, wherein, The data acquisition processing method comprises the following steps of collecting data transmission and storage: Network transmission and storage, using the sparsity of wavelet transform to compress data, and encoding the compressed data for network transmission; Data recovery, inverse transform the received wavelet coefficients to recover the original signal, and enhance the recovered signal.

9. A traffic fusion data collection system based on intelligent lamp poles, characterized in that, It includes a vehicle-mounted device, a smart lamp pole, and a central control platform. The vehicle-mounted device includes a vehicle-mounted MEC unit and a vehicle-mounted Bluetooth device deployed on the vehicle, which are used to interact with the smart lamp pole. The smart lamp pole is built by combining poles and boxes, and contains a multi-sensor data acquisition device, a perception camera, a roadside unit RSU, and a router, which are used to interact with the vehicle-mounted device and the central control platform. The central control platform contains a data fusion unit and a human-computer interaction interface, and the central control platform interacts with the smart lamp pole. 10.The traffic fusion data collection system based on intelligent lamp pole according to claim 9, wherein, The data fusion unit executes the traffic fusion data acquisition processing method according to any one of claims 1-8. The data fusion unit based on the orthogonal rational wavelet family data fusion method, by preprocessing the collected original data; subsequently, the constructed orthogonal rational wavelet filter bank cooperates with the sensor array and the network to form a multi-dimensional wavelet filter array data, and a wavelet probabilistic neural network model is built to fuse and optimize the multi-dimensional multi-path signals in the entire network to obtain the value closest to the true value.

Images