AIoT device control method and system under cloud edge collaborative architecture

Abstract

The invention provides an AIoT device control method and system under a cloud edge collaborative architecture, and relates to the technical field of intelligent control, and the method comprises the steps: carrying out the clustering division of a target road according to a night brightness monitoring record, and laying a brightness sensing layer for each independent road segment; monitoring a plurality of real-time ambient brightness; the data sensing layer is used for monitoring and obtaining road traffic flow and vehicle images; predicting the vehicle lamp brightness in the edge calculation layer; and according to the plurality of real-time environment brightness and the predicted vehicle lamp brightness, optimizing the night standard driving brightness to obtain a plurality of adaptive driving brightness, and dynamically regulating and controlling the street lamp system of the target road. The technical problem that driving safety is affected due to excessive illumination or insufficient illumination of the street lamp due to the fact that AIoT street lamp control cannot dynamically adjust the brightness according to real-time traffic and environmental conditions in the prior art can be solved, and the energy utilization efficiency and the driving safety are improved by dynamically regulating and controlling the street lamp according to actual requirements of a road section.

Description

Technical Field

[0001] The present application relates to the field of intelligent control technology, and specifically to an AIoT device control method and system under a cloud-edge collaborative architecture. Background Art

[0002] In urban construction, AIoT devices are widely used in traffic management, environmental monitoring, smart street light control and other fields. As a typical application of AIoT, smart street lights are equipped with sensors, cameras, communication modules and other equipment to sense environmental information, traffic flow, etc. in real time, and dynamically adjust according to these data to achieve energy saving, improve road safety and optimize management efficiency. Traditional smart street light control usually adjusts the brightness of street lights based on fixed time periods or ambient light intensity sensors, but it cannot take into account dynamic factors such as traffic volume, weather changes, and traffic patterns on the road section, resulting in insufficient brightness of street lights during heavy traffic or in bad weather, affecting driving safety; and during periods of low traffic volume or high ambient brightness, it may cause excessive lighting and waste energy.

[0003] To sum up, there are technical problems in the prior art that AIoT street lamp control cannot dynamically adjust the brightness according to real-time traffic and environmental conditions, resulting in excessive or insufficient lighting of street lamps, further affecting energy utilization efficiency and driving safety. Summary of the invention

[0004] The purpose of this application is to provide an AIoT device control method and system under a cloud-edge collaborative architecture, so as to solve the technical problem in the prior art that the AIoT street lamp control cannot dynamically adjust the brightness according to real-time traffic and environmental conditions, resulting in excessive or insufficient lighting of the street lamps, further affecting energy utilization efficiency and driving safety.

[0005] In order to achieve the above objectives, the present application provides an AIoT device control method and system under a cloud-edge collaborative architecture.

[0006] In the first aspect, the present application provides an AIoT device control method under a cloud-edge collaborative architecture, which is implemented by an AIoT device control system under the cloud-edge collaborative architecture, wherein the AIoT device control method under the cloud-edge collaborative architecture includes: clustering the target road according to the night brightness monitoring records of the target road, determining multiple independent sections, and deploying a brightness perception layer for each independent section; monitoring and obtaining multiple real-time ambient brightness of the multiple independent sections through the brightness perception layer; monitoring and obtaining road traffic flow and vehicle images through the data perception layer at a predetermined position, wherein the predetermined position also has an edge computing layer embedded; at the edge computing layer, obtaining predicted headlight brightness based on the road traffic flow and vehicle image analysis; optimizing the standard driving brightness at night based on the multiple real-time ambient brightness and predicted headlight brightness, obtaining multiple adaptive driving brightness, and dynamically adjusting the street light system of the target road.

[0007] Optionally, if no passing vehicle is detected at the predetermined position, the street light system of the target road is dynamically adjusted according to a predetermined minimum brightness, wherein the predetermined position is in front of the target road and is separated from the target road by a predetermined distance.

[0008] Optionally, the brightness perception layer and the data perception layer are communicatively connected to the edge computing layer, wherein the brightness perception layer is embedded with a brightness sensor, and the data perception layer is embedded with a surveillance camera.

[0009] Optionally, a nighttime brightness monitoring record of a target road is obtained, wherein the nighttime brightness monitoring record includes a plurality of nighttime brightness monitoring data, each piece of nighttime brightness monitoring data includes ambient brightness data at a plurality of predetermined monitoring coordinates, and adjacent predetermined monitoring coordinates are spaced at a predetermined length; the plurality of nighttime brightness monitoring data are divided according to a plurality of predetermined moonlight intensity intervals to determine a plurality of sample brightness monitoring data sets; the target roads are clustered and divided according to the plurality of sample brightness monitoring data sets to determine a plurality of road section division results; the plurality of road section division results are sequentially fused and clustered to obtain a plurality of independent road sections.

[0010] Optionally, the target road is divided according to the predetermined length to determine multiple sub-segments, and the first sub-segment and the second sub-segment are selected in sequence; a first sample brightness monitoring data set is randomly selected, and the first brightness data set and the second brightness data set of the first sub-segment and the second sub-segment are respectively extracted, and deviation calculation is performed to obtain a first brightness deviation set; if the first brightness deviation sets are all less than a predetermined deviation scalar, the first sub-segment and the second sub-segment are merged and set as a first aggregated segment, wherein the brightness data of the aggregated segment is the average of the brightness data of several sub-segments; if there is data greater than or equal to the predetermined deviation scalar in the first brightness deviation set, the first sub-segment is set as the first aggregated segment, and with the second sub-segment as the starting point, iterative clustering is performed until all sub-segments are traversed, and a first segment division result is obtained, which is added to the multiple segment division results.

[0011] Optionally, multiple first aggregated road segments from the multiple road segment division results are obtained in sequence, wherein each first aggregated road segment includes a number of first sub-segments; the frequencies of occurrence of the first sub-segments in the multiple first aggregated road segments are counted respectively, and the first sub-segments with a frequency greater than a predetermined threshold are added to the first independent road segment, and the first sub-segments with a frequency less than or equal to the predetermined threshold are added to the second independent road segment; taking the second independent road segment as the starting point, continue to perform fusion clustering to obtain multiple independent road segments.

[0012] Optionally, at the edge computing layer, a vehicle identification plug-in is used to perform model identification based on the vehicle image to obtain the vehicle model, wherein the vehicle identification plug-in is constructed based on a convolutional neural network, trained at the cloud computing layer, and transferred to the edge computing layer after convergence; the headlight information is matched in the database based on the vehicle model to obtain the vehicle high beam brightness and the vehicle low beam brightness; at the edge computing layer, a headlight state prediction plug-in is used to predict and output multiple headlight state probabilities of the multiple independent road sections according to the current time node and the road traffic volume, wherein the headlight state probability includes a low beam probability and a high beam probability; multiple predicted headlight brightnesses are calculated based on the vehicle high beam brightness, the vehicle low beam brightness, multiple low beam probabilities and multiple high beam probabilities.

[0013] Optionally, at the cloud computing layer, based on the historical monitoring data of the first independent road section, a sample time node set and a sample traffic flow set are collected, and the proportion of vehicles using high and low beams at different sample time nodes and sample traffic flows are marked to obtain a sample low beam probability set and a sample high beam probability set; a feedforward neural network is trained using the sample time node set, the sample traffic flow set, the sample low beam probability set and the sample high beam probability set until convergence to obtain a first headlight state prediction branch; multiple headlight state prediction branches of multiple independent road sections are analyzed in turn, the headlight state prediction plug-in is mapped and constructed, and is delegated to the edge computing layer.

[0014] Optionally, if the low beam probability or the high beam probability is greater than a predetermined probability threshold, the low beam probability or the high beam probability is marked as 1, wherein the predetermined probability threshold is 75%; a weighted calculation is performed based on the vehicle high beam brightness, the vehicle low beam brightness, multiple low beam probabilities and multiple high beam probabilities to output multiple predicted headlight brightnesses.

[0015] In the second aspect, the present application also provides an AIoT device control system under a cloud-edge collaborative architecture, which is used to execute the AIoT device control method under the cloud-edge collaborative architecture as described in the first aspect, wherein the AIoT device control system under the cloud-edge collaborative architecture includes: a clustering module, which is used to cluster the target road according to the night brightness monitoring records of the target road, determine multiple independent road sections, and deploy a brightness perception layer for each independent road section; a brightness monitoring module, which is used to monitor and obtain multiple real-time ambient brightness of the multiple independent road sections through the brightness perception layer; a road condition monitoring module, which is used to monitor and obtain road traffic flow and vehicle images through the data perception layer at a predetermined position, wherein the predetermined position also has an edge computing layer embedded; a headlight brightness prediction module, which is used to obtain multiple predicted headlight brightnesses according to the road traffic flow and vehicle image analysis at the edge computing layer; a brightness control module, which is used to optimize the standard driving brightness at night according to the multiple real-time ambient brightnesses and multiple predicted headlight brightnesses, obtain multiple adaptive driving brightnesses, and dynamically control the street light system of the target road.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: According to the nighttime brightness monitoring records of the target road, the target road is clustered and divided to determine multiple independent sections, and a brightness perception layer is arranged for each independent section; multiple real-time environmental brightness of the multiple independent sections is obtained through the brightness perception layer monitoring; the road traffic volume and vehicle image are obtained through the data perception layer monitoring at the predetermined position, wherein the predetermined position is also embedded with an edge computing layer; at the edge computing layer, the predicted headlight brightness is obtained according to the road traffic volume and vehicle image analysis; according to the multiple real-time environmental brightness and predicted headlight brightness, the nighttime standard driving brightness is optimized to obtain multiple adapted driving brightness, and the streetlight system of the target road is dynamically regulated. That is, according to the nighttime brightness monitoring records, the target road is divided into multiple independent sections, and the brightness perception layers are arranged respectively to obtain the real-time environmental brightness of each section, and the headlight brightness is predicted according to the road traffic volume and vehicle image, and the streetlight system is dynamically regulated by the real-time environmental brightness and predicted headlight brightness, which not only avoids excessive lighting, but also ensures sufficient lighting, and enhances driving safety while improving energy efficiency.

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0019] Figure 1 This is a flowchart of the AIoT device control method under the cloud-edge collaborative architecture of this application; Figure 2 This is a structural diagram of the AIoT device control system under the cloud-edge collaborative architecture of this application.

[0020] Explanation of the accompanying drawings: clustering division module 11, brightness monitoring module 12, road condition monitoring module 13, vehicle light brightness prediction module 14, brightness control module 15. DETAILED DESCRIPTION

[0021] This application solves the technical problem in the prior art that the AIoT street light control cannot dynamically adjust the brightness according to real-time traffic and environmental conditions, resulting in excessive or insufficient lighting of street lights, further affecting energy efficiency and driving safety, by providing an AIoT device control method and system under a cloud-edge collaborative architecture. The target road is divided into multiple independent sections based on nighttime brightness monitoring records, and brightness perception layers are deployed separately to obtain the real-time ambient brightness of each section. The headlight brightness is predicted based on road traffic volume and vehicle images, and the street light system is dynamically adjusted based on real-time ambient brightness and predicted headlight brightness, which not only avoids excessive lighting but also ensures sufficient lighting, thereby improving energy efficiency and enhancing driving safety.

[0022] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0023] For example, please refer to the attached Figure 1 The present application provides an AIoT device control method under a cloud-edge collaborative architecture, wherein the AIoT device control method under the cloud-edge collaborative architecture is applied to an AIoT device control system under the cloud-edge collaborative architecture, and the AIoT device control method under the cloud-edge collaborative architecture specifically includes the following steps: S100: Clustering the target road according to the nighttime brightness monitoring record of the target road, determining a plurality of independent road sections, and laying out a brightness perception layer for each independent road section.

[0024] Furthermore, the present application S100 includes: Acquire a nighttime brightness monitoring record of a target road, wherein the nighttime brightness monitoring record comprises a plurality of nighttime brightness monitoring data, each piece of nighttime brightness monitoring data comprises ambient brightness data under a plurality of predetermined monitoring coordinates, and adjacent predetermined monitoring coordinates are spaced at a predetermined length; divide the plurality of nighttime brightness monitoring data according to a plurality of predetermined moonlight intensity intervals to determine a plurality of sample brightness monitoring data sets; cluster the target roads according to the plurality of sample brightness monitoring data sets to determine a plurality of road section division results; and sequentially fuse and cluster the plurality of road section division results to obtain a plurality of independent road sections.

[0025] Furthermore, the present application also includes the following steps: The target road is divided according to the predetermined length to determine multiple sub-segments, and the first sub-segment and the second sub-segment are selected in sequence; a first sample brightness monitoring data set is randomly selected, and the first brightness data set and the second brightness data set of the first sub-segment and the second sub-segment are respectively extracted, and a deviation is calculated to obtain a first brightness deviation set; if the first brightness deviation sets are all less than a predetermined deviation scalar, the first sub-segment and the second sub-segment are merged to set as a first aggregated segment, wherein the brightness data of the aggregated segment is the average of the brightness data of several sub-segments; if there is data greater than or equal to the predetermined deviation scalar in the first brightness deviation set, the first sub-segment is set as the first aggregated segment, and iterative clustering is performed with the second sub-segment as the starting point until all sub-segments are traversed, and a first segment division result is obtained, which is added to the multiple segment division results.

[0026] Furthermore, the present application also includes the following steps: Sequentially obtain multiple first aggregated road segments from the multiple road segment division results, wherein each first aggregated road segment includes a number of first sub-road segments; respectively count the frequencies of occurrence of the first sub-road segments in the multiple first aggregated road segments, and add the first sub-road segments with a frequency greater than a predetermined threshold to the first independent road segment, and add the first sub-road segments with a frequency less than or equal to the predetermined threshold to the second independent road segment; and continue to perform fusion clustering with the second independent road segment as a starting point to obtain multiple independent road segments.

[0027] Specifically, the target road is monitored at night by means of a photosensitive sensor, a camera or other light sensing equipment to obtain a nighttime brightness monitoring record. The nighttime brightness monitoring record includes multiple nighttime brightness monitoring data, each of which contains multiple ambient brightness values ​​at predetermined monitoring coordinates. The ambient brightness data is the brightness value measured by the sensor at the predetermined coordinate position, usually in lux, which represents the luminous flux received per unit area. The predetermined monitoring coordinates are measurement points predetermined when the sensor or equipment is installed, and are used to collect environmental data. The predetermined length between adjacent predetermined monitoring coordinates means that there is a predetermined length between two adjacent predetermined monitoring coordinates, and this predetermined length is the distance between two adjacent monitoring points.

[0028] The predetermined moonlight intensity interval refers to multiple brightness intervals divided according to different moonlight brightness conditions (usually affected by weather, moon phase, season, etc.). Different moonlight intensities will affect the brightness on the ground, so the influence of moonlight must be considered when monitoring brightness at night. The moonlight intensity interval divides different moonlight intensities into several intervals. Common intervals may include low intensity, medium intensity, and high intensity. For example, low intensity may be 0 to 50 lux, medium intensity is 51 to 100 lux, and high intensity is 101 to 150 lux.

[0029] According to multiple predetermined moonlight intensity intervals, multiple nighttime brightness monitoring data are divided to obtain multiple sample brightness monitoring data sets. Each sample brightness monitoring data set represents brightness data within a specific moonlight intensity interval. For example, the moonlight intensity is divided from 0 to 100 lux into three intervals: low, medium, and high. The moonlight intensity is 50 lux in some periods, and these data can be classified into the medium-intensity moonlight interval; for periods with weak moonlight intensity, they are classified into the low-intensity moonlight interval.

[0030] The target road is divided into multiple sub-segments according to a predetermined length. The predetermined length is a length pre-set according to factors such as the overall length of the road, traffic demand, lighting requirements, etc., which may be 100 meters, 200 meters or 500 meters, etc. The setting of the predetermined length needs to balance the accuracy of the segment division and the complexity of the control. According to the total length and the predetermined length of the target road, the target road is divided into multiple sub-segments. Among the multiple sub-segments divided, the first sub-segment and the second sub-segment are selected in sequence. That is to say, starting from the starting point, the sections are selected one by one according to the actual order of the road. For example, when the preset length is 200 meters, the first sub-segment is the position from the starting point of the target road to the first 200 meters, and the second sub-segment is the position from the 200th meter to the 400th meter.

[0031] A first sample brightness monitoring data set is randomly selected from multiple sample brightness monitoring data sets, which includes brightness monitoring data under a certain moonlight intensity. A first brightness data set corresponding to the first sub-section and a second brightness data set corresponding to the second sub-section are matched from the first sample brightness monitoring data set, and deviation calculation is performed on the two data sets to obtain a first brightness deviation set. Deviation calculation is used to measure the difference in brightness data between two sub-sections, and is performed by comparing the brightness values ​​of the two data sets.

[0032] The first brightness deviation set indicates whether the brightness data of the first sub-segment and the second sub-segment are consistent or close. If the deviation between the brightness values ​​of the two sub-segments is small, it means that the brightness conditions of the two sub-segments are similar and can be merged into one segment for unified regulation. If the deviation is large, it means that the brightness of the two sub-segments varies greatly and may need to be managed separately. For example, the brightness data of the first sub-segment is: [20, 25, 22, 18, 23] (unit: lux), and the brightness data of the second sub-segment is: [35, 38, 32, 36, 37]. Then the difference in brightness data between the first sub-segment and the second sub-segment is the deviation, which is [15, 13, 10, 18, 14].

[0033] The predetermined deviation scalar is a preset threshold used to determine whether the brightness of two sub-segments is similar enough. If the deviation is less than the scalar, it is considered that the brightness data of the two sub-segments are similar and can be merged into one segment. If all values ​​in the deviation set are less than the predetermined deviation scalar, it is considered that the brightness of the two sub-segments is very similar and can be merged into an aggregated segment. The brightness data of this aggregated segment is obtained by averaging the brightness data of several aggregated sub-segments.

[0034] If there is data greater than or equal to the predetermined deviation scalar in the first brightness deviation set, that is, at least one deviation value is greater than or equal to the predetermined deviation scalar, it will not be merged, but the current first sub-segment will be kept as the aggregated segment, and the clustering judgment will continue to be performed on the next sub-segment. The clustering operation will continue on the remaining sub-segments. First, the current sub-segment will be retained as the basis of the aggregated segment, and then the deviation will be calculated with the next sub-segment, and it will be determined whether to merge. This process will continue until all sub-segments have been traversed and clustered.

[0035] The brightness difference between each two sub-segments is compared step by step in order to determine whether clustering can be performed. If the brightness difference between two sub-segments meets the conditions, they are merged. If the deviation is greater than the predetermined scalar, the previous segment is set as the aggregation segment, and the subsequent sub-segments are used as new starting points, and this process is repeated until all sub-segments have been traversed. Through iterative clustering, the clustering division result of the first sample brightness monitoring data set is finally obtained, that is, the first segment division result, which shows the aggregation grouping of multiple sub-segments in the target road. Each aggregation segment contains multiple sub-segments, and the brightness differences of these sub-segments are small, which are suitable for adjustment under unified control.

[0036] The above steps are performed on other sample brightness monitoring data sets in the multiple sample brightness monitoring data sets to obtain multiple road section division results after the multiple sample brightness monitoring data sets are clustered and divided.

[0037] Multiple first aggregated road sections are obtained in sequence from multiple road section division results. The sections are composed of multiple sub-sections. The sub-sections within each aggregated road section have small brightness differences, and the complexity of street light adjustment can be reduced by merging sub-sections. Each first aggregated road section includes one or more first sub-sections. The frequency of each sub-section is counted. The frequency refers to the number of times a sub-section appears in different aggregated road sections. It helps to understand which sub-sections appear frequently in multiple aggregated road sections, which means that these sections may be more representative and need to be given more attention. For example, if sub-section A appears multiple times in multiple aggregated road sections, it may be a traffic-intensive area or a main road section, and therefore should be given priority control.

[0038] The frequency of the first sub-segment appearing in each first aggregated segment is counted respectively. If the frequency of the first sub-segment appearing in the first aggregated segment is greater than a predetermined threshold, the corresponding segment division results are clustered into the first independent segment. Otherwise, the corresponding segment division results of the first sub-segment with a lower frequency of appearing in each first aggregated segment are clustered into the second independent segment. The predetermined threshold is a criterion for determining which sub-segments are more important. Using the second independent segment as the starting point, the above steps are repeated for clustering until multiple independent segments are obtained.

[0039] The target road is clustered and divided into multiple independent sections, and a brightness perception layer is laid out for each independent section. According to the characteristics of the section, a suitable brightness sensor is selected and installed at key locations of each independent section, such as the starting point, end point, intersection, etc. The brightness data of each independent section has a high consistency, so that similar control strategies can be used for these sections.

[0040] Through step-by-step clustering, frequency statistics and fusion clustering, important road sections are identified. Accurate road section control is achieved through cluster division and the layout of brightness perception layer. Each independent road section will be adjusted according to specific brightness data to ensure that the street light system can adapt to different traffic flows and ambient lighting conditions.

[0041] S200: Acquire multiple real-time ambient brightnesses of the multiple independent road sections through brightness perception layer monitoring.

[0042] Specifically, multiple independent road sections are monitored in real time through the brightness perception layer to obtain multiple real-time ambient brightness corresponding to the multiple independent road sections, reflecting the light intensity of different areas of the road. For example, areas with heavy traffic may have higher brightness due to the effects of vehicle lights and street lights, while sections with less traffic or less busy roads may have lower brightness. Through the monitoring of the brightness perception layer, the ambient brightness data of each independent road section is obtained in real time, thereby achieving more accurate street light control, so that the street light system can be dynamically adjusted according to the actual brightness requirements of each road section, avoiding over-lighting or under-lighting in traditional street light systems. Especially in dense traffic or special weather conditions, the lighting brightness can be increased in a timely manner to improve road safety; in areas with relatively smooth traffic, energy can be saved by reducing the brightness.

[0043] S300: Monitoring and acquiring road traffic flow and vehicle images through a data perception layer at a predetermined location, wherein the predetermined location also has an edge computing layer embedded therein.

[0044] The brightness perception layer and the data perception layer are communicatively connected with the edge computing layer, wherein the brightness perception layer is embedded with a brightness sensor, and the data perception layer is embedded with a surveillance camera.

[0045] Specifically, the data perception layer at the predetermined location is used for monitoring to obtain the road traffic flow and vehicle images at the predetermined location. The predetermined location is a selected location for monitoring the passing vehicles, such as a city intersection, a busy road section, etc., where the data perception layer, edge computing layer, brightness perception layer, etc. are deployed. Traffic flow refers to the number of vehicles passing through a certain location per unit time. Vehicle images are vehicle image data obtained through surveillance cameras, which are usually used for license plate recognition, traffic condition analysis, etc. Through traffic flow and vehicle images, the traffic conditions can be understood in real time, and the brightness adjustment of street lights can be optimized.

[0046] The brightness perception layer is a layer dedicated to monitoring road ambient lighting. It is embedded with brightness sensors that can measure the ambient brightness (in lux) of various areas on the road in real time. The brightness data collected by the sensor will help determine whether the brightness of the street lights needs to be adjusted to ensure that the road lighting matches the actual environmental needs.

[0047] The data perception layer is a layer dedicated to monitoring traffic flow and vehicle behavior. It is embedded with surveillance cameras, which provide necessary traffic flow data by capturing real-time information such as traffic conditions, number of vehicles and speed on the road. Through image analysis technology, these cameras can not only identify vehicles, but also analyze traffic conditions on the road, such as congestion and speed, and further provide decision-making basis for AIoT devices.

[0048] The edge computing layer refers to the technology that processes and calculates data near the source of data collection (such as the perception layer on the road), quickly processing data from the brightness perception layer and the data perception layer without transmitting the data to a remote cloud server for processing. Edge computing can greatly reduce response time and bandwidth requirements. The data processed includes traffic flow, vehicle images, ambient brightness, etc., and the brightness of street lights is dynamically adjusted based on this data to provide real-time traffic response.

[0049] The brightness perception layer and data perception layer monitor the target road through the embedded brightness sensor and surveillance camera respectively. The monitoring data is transmitted to the edge computing layer, which processes and analyzes the data in real time to determine whether the street lights need to be adjusted. For example, if the traffic volume is high or the brightness in a certain area is insufficient, the edge computing layer will immediately give instructions based on this information to adjust the brightness of the street lights. Moreover, since data processing occurs at the edge computing layer, it can respond to changes quickly, avoiding problems caused by transmission delays.

[0050] S400: At the edge computing layer, a plurality of predicted headlight brightnesses are obtained based on the road traffic flow and vehicle image analysis.

[0051] Further, the present application S400 includes: At the edge computing layer, a vehicle identification plug-in is used to perform model identification based on the vehicle image to obtain the vehicle model, wherein the vehicle identification plug-in is constructed based on a convolutional neural network, trained at the cloud computing layer, and transferred to the edge computing layer after convergence; based on the vehicle model, the headlight information is matched in the database to obtain the vehicle high beam brightness and the vehicle low beam brightness; at the edge computing layer, a headlight state prediction plug-in is used to respectively predict and output multiple headlight state probabilities of the multiple independent road sections according to the current time node and the road traffic volume, wherein the headlight state probability includes a low beam probability and a high beam probability; based on the vehicle high beam brightness, the vehicle low beam brightness, multiple low beam probabilities and multiple high beam probabilities, multiple predicted headlight brightnesses are calculated.

[0052] Specifically, the road traffic volume and vehicle images monitored by the data perception layer are transmitted to the edge computing layer. In the cloud computing layer, the vehicle recognition plug-in is trained and transferred to the edge computing layer to determine the vehicle model based on the vehicle image. The vehicle recognition plug-in is a computer vision-based tool that can identify the vehicle model by processing the vehicle image.

[0053] The vehicle recognition plug-in is a model built on a convolutional neural network that can identify the model of a vehicle by analyzing vehicle images. It extracts key information from vehicle images obtained by cameras or sensors to identify the type or brand of the vehicle, such as Audi A6 or Toyota Corolla. Convolutional neural networks automatically extract features (such as edges, shapes, colors, etc.) in images through convolution operations and can identify objects in images.

[0054] Obtain a variety of vehicle models and their corresponding vehicle images from various angles from public transportation databases, images collected by surveillance cameras, and automobile manufacturers, and mark the specific vehicle model on each vehicle image. The images in the dataset should include photos of vehicles at different time periods (especially at night) to ensure that there are sufficient lighting changes in the images (such as the influence of high beams, low beams, street lighting, etc.). The images should contain a variety of weather conditions (such as rainy days, foggy days, etc.) and different traffic environments (such as dense traffic, different speeds, etc.). For example, obtain an existing bus image dataset and expand the data in combination with actual surveillance videos. Label the vehicle model for each image to ensure the accuracy of the labeling. In night scenes, you can simulate different night conditions by adjusting the brightness, contrast, and color saturation of the image to enhance the model's ability to recognize vehicles under different lighting conditions.

[0055] Design a convolutional neural network architecture suitable for vehicle recognition tasks, including multiple convolutional layers and pooling layers. The input layer is responsible for receiving vehicle images, usually RGB three-channel images, and the image size can be set to a fixed size, such as 224x224 pixels; the convolutional layer is responsible for extracting local features in the image, such as appearance, front and rear design, body contour, wheel design, texture, color, etc. The activation function is used to increase the nonlinear ability of the model. ReLU is often used as an activation function in convolutional neural networks. The pooling layer (such as the maximum pooling layer) is used to reduce the size of the feature map, reduce the amount of calculation, and enhance the robustness of the model. After the convolutional layer extracts the features, the fully connected layer maps the features to the target category (such as a specific vehicle model). Finally, the model category of the vehicle is output.

[0056] The entire dataset is divided into a training set and a validation set. Usually the training set accounts for 80% of the dataset and the validation set accounts for 20%. The validation set is used to monitor the performance of the model during the training process to prevent overfitting. The constructed convolutional neural network architecture is trained using the training set. The output value is calculated through forward propagation and compared with the actual label to calculate the loss. The cross entropy loss function is used to measure the gap between the model prediction and the actual label. The training goal is to minimize the loss function and improve the accuracy of the model. During the training process, the model is evaluated on the validation set after each epoch of training to calculate its accuracy or other performance indicators. The gradient is calculated and the weights of the model are updated through the back propagation algorithm. This process is repeated for multiple epochs until the loss function converges and the model shows good accuracy on both the training set and the validation set. During the training process, the parameters of the model are continuously adjusted according to the loss calculated by the loss function, such as increasing the number of network layers, using a more complex convolutional neural network structure, and performing more data augmentation. After the training is completed, the model is finally evaluated using the test set. The classification effect of the model on different vehicle models is evaluated by calculating indicators such as accuracy, precision, recall, and F1 score. When the classification accuracy of the model reaches 95%, the training is stopped and the model obtained at this time is used as a vehicle recognition plug-in.

[0057] This process is usually trained in the cloud computing layer, using a large amount of labeled vehicle image data. Through these images, the convolutional neural network can gradually learn how to identify various features of the vehicle. The goal of training is to optimize the weights and parameters in the network until the accuracy of the model reaches the predetermined standard. When the convolutional neural network model converges, that is, the training result reaches a satisfactory accuracy, the trained vehicle recognition plug-in is transferred from the cloud computing layer to the edge computing layer. The plug-in transferred to the edge computing layer will no longer rely on the remote server, but directly process the real-time vehicle image data collected from the camera locally. For example, suppose a surveillance camera on a road section captures an image of a car. After being processed by the vehicle recognition plug-in in the edge computing layer, the model of the vehicle (such as Audi A6) is quickly identified. The execution of this model in the edge computing layer avoids the delay of transmitting image data to the remote server and improves the response speed.

[0058] According to the identified vehicle model, the database is matched to obtain the corresponding headlight information, including the vehicle's high beam brightness and low beam brightness. The database is a structured data storage system that contains relevant information about the vehicle and stores detailed information corresponding to each vehicle model, including headlight type, high beam brightness, low beam brightness and other data. This information is usually defined in the vehicle specifications provided by the vehicle manufacturer, and the brightness value may be different for different vehicle models and lighting systems. For example, the vehicle lighting information is shown in Table 1: Table 1 Vehicle lighting information table

[0059] Surveillance cameras usually capture high-quality videos and provide real-time traffic data. Video data captured from surveillance cameras needs to be analyzed to extract useful information, especially the number of vehicles on the road. A video stream consists of consecutive frames, each of which is a static image that captures the road conditions at a certain moment. In video analysis, the video stream is decomposed into individual image frames, and then each frame is processed. In each frame, the vehicle needs to be identified and marked by the target detection algorithm (such as the YOLO algorithm), and the position of the vehicle is located in each frame by learning vehicles of different shapes, colors, and features. By detecting the vehicles in each frame, the traffic volume is calculated. The traffic volume refers to the number of vehicles passing through a specific road section per unit time. The number of detected vehicles is the number of vehicles in the frame. The vehicle flow in a certain period of time can be calculated by the number of vehicles in several consecutive frames. Based on the vehicle detection results in each frame, the traffic volume per unit time (such as per minute or per hour) is estimated. For example, if 200 vehicles pass through the road section every minute, the output traffic volume is 200 vehicles / minute.

[0060] At the edge computing layer, the built headlight status prediction plug-in is used to predict the headlight status probability of each independent road section based on the current time node and the traffic volume on the road, that is, the probability of the vehicle using low beam and high beam at the current time node. Since the traffic volume and time node change in real time, the prediction plug-in in the edge computing layer can continuously update the prediction results based on new data.

[0061] The brightness of the headlights on multiple road sections is calculated based on the high-beam and low-beam brightness obtained from the database, and the low-beam and high-beam probabilities obtained from the headlight status prediction plug-in. Based on the high-beam and low-beam brightness of the vehicle and the predicted usage probability, if the predicted usage probability of the low-beam or high-beam is greater than 75%, the predicted headlight brightness is determined according to the low-beam or high-beam. Otherwise, it is weighted according to the usage probability of the low-beam and high-beam. The headlight brightness is predicted through road traffic flow and vehicle image analysis. Based on the predicted headlight usage pattern, the street light brightness is reduced during periods of low traffic or high high-beam usage, thereby saving energy.

[0062] Furthermore, the present application also includes the following steps: At the cloud computing layer, based on the historical monitoring data of the first independent road section, a sample time node set and a sample traffic flow set are collected, and the proportion of vehicles using high and low beams at different sample time nodes and sample traffic flows are marked to obtain a sample low beam probability set and a sample high beam probability set; a feedforward neural network is trained using the sample time node set, the sample traffic flow set, the sample low beam probability set and the sample high beam probability set until convergence to obtain a first headlight state prediction branch; multiple headlight state prediction branches of multiple independent road sections are analyzed in turn, the headlight state prediction plug-in is mapped and constructed, and is delegated to the edge computing layer.

[0063] Specifically, at the cloud computing layer, based on the historical monitoring data of the first independent road section, including the traffic flow data, time node data and the status of vehicles using high beam and low beam lights on the road section. The sample time node set is the time node data extracted from the historical monitoring data, and each time node represents a specific moment or time period. The sample traffic flow set includes traffic flow data corresponding to multiple sample time nodes, indicating the number of vehicles passing through the road section within the time period. Traffic flow is related to the use pattern of vehicle lights. When the traffic flow is different, the proportion of vehicles using high beam and low beam lights may be different.

[0064] A feedforward neural network is trained using a sample time node set, a sample traffic flow set, a sample low beam probability set, and a sample high beam probability set. The data set is preprocessed to remove or correct errors and outliers in the data set to ensure data quality. The data set is standardized to prevent different dimensions from affecting the model training effect. The data set is divided into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to adjust the model parameters, and the test set is used to evaluate the final performance of the model.

[0065] A feedforward neural network is a classic type of artificial neural network in which information flows in one direction, from the input layer to the output layer, without feedback. The number of nodes in the input layer is equal to the number of features in the sample time node set and the sample traffic flow set, and the number of nodes in the output layer is equal to the dimension of the light state probability. Before starting training, the weights and biases of the network need to be initialized. Usually, a small random number is used for initialization. During training, the input data is forward propagated through the neural network. The output of each layer will be used as the input of the next layer. A loss function is selected to quantify the difference between the predicted value and the actual value, such as cross entropy loss. The error calculated by the loss function is back-propagated through the network to update the weights and biases of the network. An optimization algorithm (such as gradient descent) is used to adjust the weights of the network to minimize the loss function. The process of forward propagation and back-propagation is repeated, and the weights of the network are updated at each iteration until the model performance reaches a certain threshold or the training reaches a preset number of iterations. Use a validation set to evaluate the performance of the model and check whether overfitting or underfitting occurs. Based on the performance of the validation set, adjust the hyperparameters of the network, such as learning rate, hidden layer size, number of layers, etc. When the performance of the model on the validation set is satisfactory, the training is stopped and the first vehicle light state prediction branch is obtained.

[0066] Repeat the above steps for other independent road sections to obtain multiple headlight status prediction branches. Integrate the headlight status prediction branches of all independent road sections into a headlight status prediction plug-in, integrate the headlight prediction models of all road sections, and intelligently predict the headlight usage status of each road section based on real-time traffic flow, time nodes and other information. The headlight status prediction plug-in is decentralized to the edge computing layer. The edge computing layer is located close to the data source (such as surveillance cameras and sensors) and has low latency and fast response capabilities. At the edge computing layer, the traffic flow and time node data of each road section can be obtained in real time, and decisions can be made quickly through the prediction plug-in to automatically adjust the brightness of the street lights.

[0067] Furthermore, the present application also includes the following steps: If the low beam probability or the high beam probability is greater than a predetermined probability threshold, the low beam probability or the high beam probability is marked as 1, wherein the predetermined probability threshold is 75%; a weighted calculation is performed based on the vehicle high beam brightness, the vehicle low beam brightness, multiple low beam probabilities and multiple high beam probabilities to output multiple predicted headlight brightnesses.

[0068] Specifically, the predetermined probability threshold is a probability threshold set at 75% for judging whether the vehicle mainly uses low beam or high beam. That is, when the low beam probability or high beam probability is greater than or equal to 75%, it is considered that the vehicle on the road section almost completely uses low beam or high beam, which is simplified to selecting one light. At this time, the light type marked as 1 is selected and the corresponding light brightness is directly used.

[0069] If the probability of low beam or high beam is greater than a predetermined threshold, the light type is marked as 1, indicating that almost all vehicles use this light in the current period. Conversely, if it is less than a predetermined probability threshold, it is weighted according to the probability, such as 70%*high beam brightness+30%*low beam brightness, to obtain multiple predicted light brightnesses.

[0070] By dynamically adjusting the brightness of street lights based on real-time traffic flow and the probability of light usage, unnecessary lighting waste can be avoided while ensuring safety.

[0071] S500: Optimizing the nighttime standard driving brightness according to the multiple real-time environment brightnesses and the multiple predicted vehicle headlight brightnesses, obtaining multiple adaptive driving brightnesses, and dynamically regulating the streetlight system of the target road.

[0072] Specifically, the standard nighttime driving brightness is optimized based on multiple real-time ambient brightness and multiple predicted headlight brightness, thereby achieving dynamic regulation of the target road streetlight system. Real-time ambient brightness refers to the actual light intensity on the road, which is usually monitored by installed brightness sensors or surveillance cameras. Ambient brightness can vary due to a variety of factors, such as whether there are streetlights on the road, whether there is natural light (such as moonlight) on the street, and weather conditions.

[0073] The prediction of vehicle headlight brightness is based on factors such as traffic volume, time node, vehicle model, and probability of headlight use, to calculate the impact of the vehicle's headlight brightness (low beam and high beam) on road lighting at a specific time and road section.

[0074] Standard night driving brightness refers to the brightness level that ensures drivers have sufficient visibility when driving at night, and is usually set based on factors such as road type, traffic density and ambient light. The goal of standard driving brightness is to ensure that drivers can clearly see the road ahead and obstacles, while avoiding energy waste caused by excessive lighting and impact on the surrounding environment.

[0075] According to the standard driving brightness, determine the minimum required safe brightness. According to multiple real-time ambient brightness and multiple predicted headlight brightness, optimize the calculation of the standard driving brightness at night to obtain multiple adaptive driving brightness values. The actual brightness of each road section must meet the minimum standard driving brightness requirements. The street light brightness cannot exceed a preset maximum value to prevent energy waste. During periods of low traffic, the headlight brightness and street light brightness can be moderately reduced. Using these as constraints, continuously adjust the standard driving brightness at night, calculate the fitness value corresponding to each brightness, and select the one with the highest fitness value as the adaptive driving brightness, thereby obtaining multiple adaptive driving brightnesses for multiple independent sections, and dynamically adjust the street light system of the target road accordingly.

[0076] According to the calculated adaptive driving brightness, the brightness of the street lights is automatically adjusted to ensure that the lighting on the road meets the standard driving brightness and saves energy. For example, if the real-time ambient brightness and the headlight brightness are close to the standard driving brightness, the street light brightness is automatically reduced; if the traffic volume is large or the headlight brightness is low, the street light brightness is increased.

[0077] Furthermore, the present application also includes the following steps: If no passing vehicle is detected at the predetermined position, the street light system of the target road is dynamically adjusted according to the predetermined minimum brightness, wherein the predetermined position is in front of the target road and is separated from the target road by a predetermined distance.

[0078] Specifically, if no passing vehicles are detected at the predetermined location, that is, no vehicles are passing at the current moment, the test dynamically adjusts the streetlight system of the target road according to the predetermined minimum brightness, which helps to avoid wasting energy, especially during the period when no vehicles pass. The predetermined minimum brightness is a minimum lighting standard set, usually to ensure that the road still has the minimum safety lighting when no one is passing. For example, on the road at night, although there are no vehicles passing, a certain brightness will still be maintained to ensure the safety of pedestrians or other emergencies.

[0079] The predetermined position is usually set in front of the target road, which may be a certain distance from the target road, such as 300 meters away from the target road, leaving a buffer time for analysis and control, so as to monitor the traffic conditions of the road in advance and take measures in advance. By automatically adjusting to the lowest brightness when no vehicle is detected at the predetermined position, energy can be effectively saved and unnecessary lighting can be reduced.

[0080] In summary, the AIoT device control method under the cloud-edge collaborative architecture provided by this application has the following technical effects: According to the nighttime brightness monitoring records of the target road, the target road is clustered and divided to determine multiple independent sections, and a brightness perception layer is arranged for each independent section; multiple real-time environmental brightness of the multiple independent sections is obtained through the brightness perception layer monitoring; the road traffic volume and vehicle image are obtained through the data perception layer monitoring at the predetermined position, wherein the predetermined position is also embedded with an edge computing layer; at the edge computing layer, the predicted headlight brightness is obtained according to the road traffic volume and vehicle image analysis; according to the multiple real-time environmental brightness and predicted headlight brightness, the nighttime standard driving brightness is optimized to obtain multiple adapted driving brightness, and the streetlight system of the target road is dynamically regulated. That is, according to the nighttime brightness monitoring records, the target road is divided into multiple independent sections, and the brightness perception layers are arranged respectively to obtain the real-time environmental brightness of each section, and the headlight brightness is predicted according to the road traffic volume and vehicle image, and the streetlight system is dynamically regulated by the real-time environmental brightness and predicted headlight brightness, which not only avoids excessive lighting, but also ensures sufficient lighting, and enhances driving safety while improving energy efficiency.

[0081] Embodiment 2, based on the same inventive concept as the AIoT device control method under the cloud-edge collaborative architecture in the aforementioned embodiment 1, this application also provides an AIoT device control system under the cloud-edge collaborative architecture, please refer to the attached Figure 2 , the AIoT device control system under the cloud-edge collaborative architecture includes: A clustering module 11 is used to cluster the target road according to the nighttime brightness monitoring records of the target road, determine multiple independent road sections, and set up a brightness perception layer for each independent road section; a brightness monitoring module 12 is used to monitor and obtain multiple real-time environmental brightness of the multiple independent road sections through the brightness perception layer; a road condition monitoring module 13 is used to monitor and obtain road traffic flow and vehicle images through the data perception layer at a predetermined position, wherein the predetermined position is also embedded with an edge computing layer; a headlight brightness prediction module 14 is used to obtain multiple predicted headlight brightnesses based on the road traffic flow and vehicle image analysis at the edge computing layer; a brightness control module 15 is used to optimize the standard driving brightness at night according to the multiple real-time environmental brightnesses and the multiple predicted headlight brightnesses, obtain multiple adaptive driving brightnesses, and dynamically control the streetlight system of the target road.

[0082] Furthermore, the clustering module 11 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: Acquire a nighttime brightness monitoring record of a target road, wherein the nighttime brightness monitoring record comprises a plurality of nighttime brightness monitoring data, each piece of nighttime brightness monitoring data comprises ambient brightness data under a plurality of predetermined monitoring coordinates, and adjacent predetermined monitoring coordinates are spaced at a predetermined length; divide the plurality of nighttime brightness monitoring data according to a plurality of predetermined moonlight intensity intervals to determine a plurality of sample brightness monitoring data sets; cluster the target roads according to the plurality of sample brightness monitoring data sets to determine a plurality of road section division results; and sequentially fuse and cluster the plurality of road section division results to obtain a plurality of independent road sections.

[0083] Furthermore, the clustering module 11 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: The target road is divided according to the predetermined length to determine multiple sub-segments, and the first sub-segment and the second sub-segment are selected in sequence; a first sample brightness monitoring data set is randomly selected, and the first brightness data set and the second brightness data set of the first sub-segment and the second sub-segment are respectively extracted, and a deviation is calculated to obtain a first brightness deviation set; if the first brightness deviation sets are all less than a predetermined deviation scalar, the first sub-segment and the second sub-segment are merged to set as a first aggregated segment, wherein the brightness data of the aggregated segment is the average of the brightness data of several sub-segments; if there is data greater than or equal to the predetermined deviation scalar in the first brightness deviation set, the first sub-segment is set as the first aggregated segment, and iterative clustering is performed with the second sub-segment as the starting point until all sub-segments are traversed, and a first segment division result is obtained, which is added to the multiple segment division results.

[0084] Furthermore, the clustering module 11 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: Sequentially obtain multiple first aggregated road segments from the multiple road segment division results, wherein each first aggregated road segment includes a number of first sub-road segments; respectively count the frequencies of occurrence of the first sub-road segments in the multiple first aggregated road segments, and add the first sub-road segments with a frequency greater than a predetermined threshold to the first independent road segment, and add the first sub-road segments with a frequency less than or equal to the predetermined threshold to the second independent road segment; and continue to perform fusion clustering with the second independent road segment as a starting point to obtain multiple independent road segments.

[0085] Furthermore, the AIoT device control system under the cloud-edge collaborative architecture also includes a second dynamic control module, and the second dynamic control module is also used to: If no passing vehicle is detected at the predetermined position, the street light system of the target road is dynamically adjusted according to the predetermined minimum brightness, wherein the predetermined position is in front of the target road and is separated from the target road by a predetermined distance.

[0086] Furthermore, the road condition monitoring module 13 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: The brightness perception layer and the data perception layer are communicatively connected with the edge computing layer, wherein the brightness perception layer is embedded with a brightness sensor, and the data perception layer is embedded with a surveillance camera.

[0087] Furthermore, the headlight brightness prediction module 14 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: At the edge computing layer, a vehicle identification plug-in is used to perform model identification based on the vehicle image to obtain the vehicle model, wherein the vehicle identification plug-in is constructed based on a convolutional neural network, trained at the cloud computing layer, and transferred to the edge computing layer after convergence; based on the vehicle model, the headlight information is matched in the database to obtain the vehicle high beam brightness and the vehicle low beam brightness; at the edge computing layer, a headlight state prediction plug-in is used to respectively predict and output multiple headlight state probabilities of the multiple independent road sections according to the current time node and the road traffic volume, wherein the headlight state probability includes a low beam probability and a high beam probability; based on the vehicle high beam brightness, the vehicle low beam brightness, multiple low beam probabilities and multiple high beam probabilities, multiple predicted headlight brightnesses are calculated.

[0088] Furthermore, the headlight brightness prediction module 14 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: At the cloud computing layer, based on the historical monitoring data of the first independent road section, a sample time node set and a sample traffic flow set are collected, and the proportion of vehicles using high and low beams at different sample time nodes and sample traffic flows are marked to obtain a sample low beam probability set and a sample high beam probability set; a feedforward neural network is trained using the sample time node set, the sample traffic flow set, the sample low beam probability set and the sample high beam probability set until convergence to obtain a first headlight state prediction branch; multiple headlight state prediction branches of multiple independent road sections are analyzed in turn, the headlight state prediction plug-in is mapped and constructed, and is delegated to the edge computing layer.

[0089] Furthermore, the headlight brightness prediction module 14 in the AIoT device control system under the cloud-edge collaborative architecture is also used for: If the low beam probability or the high beam probability is greater than a predetermined probability threshold, the low beam probability or the high beam probability is marked as 1, wherein the predetermined probability threshold is 75%; a weighted calculation is performed based on the vehicle high beam brightness, the vehicle low beam brightness, multiple low beam probabilities and multiple high beam probabilities to output multiple predicted headlight brightnesses.

[0090] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1The AIoT device control method and specific examples under the cloud-edge collaborative architecture in Example 1 are also applicable to the AIoT device control system under the cloud-edge collaborative architecture of this embodiment. Through the above detailed description of the AIoT device control method under the cloud-edge collaborative architecture, technical personnel in this field can clearly understand the AIoT device control system under the cloud-edge collaborative architecture in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here.

[0091] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0092] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. AIoT device control method under cloud-edge collaborative architecture, characterized in that: include: According to the nighttime brightness monitoring record of the target road, the target road is clustered and divided to determine a plurality of independent road sections, and a brightness perception layer is arranged for each independent road section; Acquire multiple real-time ambient brightness of the multiple independent road sections through brightness perception layer monitoring; The road traffic volume and vehicle images are monitored by a data perception layer at a predetermined location, wherein an edge computing layer is also embedded in the predetermined location; At the edge computing layer, a plurality of predicted headlight brightness are obtained according to the road traffic volume and vehicle image analysis; According to the multiple real-time ambient brightnesses and the multiple predicted headlight brightnesses, the standard driving brightness at night is optimized to obtain multiple adaptive driving brightnesses, and the streetlight system of the target road is dynamically adjusted.

2. The AIoT device control method under the cloud-edge collaborative architecture according to claim 1 is characterized in that: If no passing vehicle is detected at the predetermined position, the street light system of the target road is dynamically adjusted according to the predetermined minimum brightness, wherein the predetermined position is in front of the target road and is separated from the target road by a predetermined distance.

3. The AIoT device control method under the cloud-edge collaborative architecture according to claim 2 is characterized in that: The brightness perception layer and the data perception layer are communicatively connected with the edge computing layer, wherein the brightness perception layer is embedded with a brightness sensor, and the data perception layer is embedded with a surveillance camera.

4. The AIoT device control method under the cloud-edge collaborative architecture according to claim 1 is characterized in that: According to the nighttime brightness monitoring records of the target road, the target road is clustered and divided to determine multiple independent road sections, including: Acquire a nighttime brightness monitoring record of a target road, wherein the nighttime brightness monitoring record includes a plurality of nighttime brightness monitoring data, each of which includes ambient brightness data at a plurality of predetermined monitoring coordinates, and adjacent predetermined monitoring coordinates are spaced at a predetermined length; Dividing the plurality of nighttime brightness monitoring data according to a plurality of predetermined moonlight intensity intervals to determine a plurality of sample brightness monitoring data sets; According to the multiple sample brightness monitoring data sets, clustering the target roads respectively to determine multiple road segment division results; The multiple road segment division results are sequentially fused and clustered to obtain multiple independent road segments.

5. The AIoT device control method under the cloud-edge collaborative architecture according to claim 4 is characterized in that: According to the multiple sample brightness monitoring data sets, the target roads are clustered and divided respectively to determine multiple road segment division results, including: Dividing the target road according to the predetermined length to determine a plurality of sub-segments, and selecting a first sub-segment and a second sub-segment in sequence; Randomly selecting a first sample brightness monitoring data set, and extracting a first brightness data set and a second brightness data set of the first sub-section and the second sub-section respectively, and performing deviation calculation to obtain a first brightness deviation set; If the first brightness deviation set is smaller than a predetermined deviation scalar, the first sub-segment and the second sub-segment are combined to form a first aggregated segment, wherein the brightness data of the aggregated segment is the average brightness data of the plurality of sub-segments; If there is data greater than or equal to a predetermined deviation scalar in the first brightness deviation set, the first sub-segment is set as the first aggregation segment, and iterative clustering is performed starting from the second sub-segment until all sub-segments are traversed, and the first segment division result is obtained and added to the multiple segment division results.

6. The AIoT device control method under the cloud-edge collaborative architecture according to claim 4 is characterized in that: The plurality of road segment division results are sequentially fused and clustered to obtain a plurality of independent road segments, including: Sequentially acquiring a plurality of first aggregated road segments from the plurality of road segment division results, wherein each first aggregated road segment includes a plurality of first sub-road segments; Counting the frequencies of occurrence of the first sub-segment in the plurality of first aggregated segments respectively, and adding the sub-segments whose frequencies are greater than a predetermined threshold to the first independent segment, and adding the sub-segments whose frequencies are less than or equal to the predetermined threshold to the second independent segment; Taking the second independent road segment as a starting point, continue to perform fusion clustering to obtain multiple independent road segments.

7. The AIoT device control method under the cloud-edge collaborative architecture according to claim 1 is characterized in that: At the edge computing layer, multiple predicted headlight brightness are obtained based on the road traffic volume and vehicle image analysis, including: At the edge computing layer, a vehicle identification plug-in is used to identify the vehicle model according to the vehicle image to obtain the vehicle model, wherein the vehicle identification plug-in is constructed based on a convolutional neural network, trained at the cloud computing layer, and transferred to the edge computing layer after convergence; Matching the vehicle light information in the database based on the vehicle model to obtain the vehicle high beam brightness and the vehicle low beam brightness; At the edge computing layer, a vehicle light state prediction plug-in is used to predict and output multiple vehicle light state probabilities of the multiple independent road sections according to the current time node and the road traffic volume, wherein the vehicle light state probability includes a low beam probability and a high beam probability; A plurality of predicted vehicle light brightnesses are calculated based on the vehicle high beam brightness, the vehicle low beam brightness, the plurality of low beam probabilities and the plurality of high beam probabilities.

8. The AIoT device control method under the cloud-edge collaborative architecture according to claim 7 is characterized in that: The construction process of the vehicle light state prediction plug-in includes: At the cloud computing layer, based on the historical monitoring data of the first independent road section, a sample time node set and a sample traffic flow set are collected, and the proportion of vehicles using high and low beams at different sample time nodes and sample traffic flows is marked to obtain a sample low beam probability set and a sample high beam probability set; Using the sample time node set, the sample traffic volume set, the sample low beam probability set and the sample high beam probability set to train a feedforward neural network until convergence, to obtain a first vehicle light state prediction branch; Analyze and obtain multiple vehicle light state prediction branches of multiple independent road sections in sequence, map and construct the vehicle light state prediction plug-in, and delegate it to the edge computing layer.

9. The AIoT device control method under the cloud-edge collaborative architecture according to claim 7 is characterized in that: According to the vehicle high beam brightness, the vehicle low beam brightness, the multiple low beam probabilities and the multiple high beam probabilities, multiple predicted vehicle light brightnesses are calculated, including: If the low beam probability or the high beam probability is greater than a predetermined probability threshold, the low beam probability or the high beam probability is marked as 1, wherein the predetermined probability threshold is 75%; A weighted calculation is performed based on the vehicle high beam brightness, the vehicle low beam brightness, a plurality of low beam probabilities and a plurality of high beam probabilities, and a plurality of predicted vehicle light brightnesses are output.

10. The AIoT device control system under the cloud-edge collaborative architecture is characterized by: Steps for implementing the AIoT device control method under the cloud-edge collaborative architecture described in any one of claims 1 to 9, wherein the AIoT device control system under the cloud-edge collaborative architecture includes: A clustering module is used to cluster the target road according to the nighttime brightness monitoring record of the target road, determine multiple independent road sections, and lay out a brightness perception layer for each independent road section; A brightness monitoring module, used for acquiring a plurality of real-time ambient brightness of the plurality of independent road sections through a brightness perception layer monitoring; A road condition monitoring module is used to monitor and obtain road traffic volume and vehicle images through a data perception layer at a predetermined location, wherein an edge computing layer is also embedded in the predetermined location; A headlight brightness prediction module, used to obtain multiple predicted headlight brightnesses based on the road traffic flow and vehicle image analysis at the edge computing layer; The brightness control module is used to optimize the standard driving brightness at night according to the multiple real-time ambient brightnesses and the multiple predicted headlight brightnesses, obtain multiple adaptive driving brightnesses, and dynamically control the streetlight system of the target road.

Images