A data analysis-based road lighting energy-saving optimization control method

Abstract

The present application relates to the technical field of intelligent lighting control, and more particularly to a road lighting energy-saving optimization control method based on data analysis, which comprises the following steps: dividing the overhead image of a street lamp into blocks, comparing and analyzing the color proportion characteristics, and combining the overall morphology of adjacent image blocks to perform dynamic trajectory matching of complete image blocks in adjacent overhead images, identifying dynamic pixel points in each overhead image, obtaining the surrounding road activity intensity of each adjacent street lamp in the current period, and then predicting the surrounding road activity intensity of the target street lamp in combination with the surrounding road activity intensity of the adjacent street lamp, so as to predict the flow of people and vehicles around the target street lamp in advance, thereby adjusting the brightness of the street lamp in advance, and effectively solving the problem of lag in traditional monitoring and adjustment.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting control technology, and in particular to a method for optimizing energy-saving control of road lighting based on data analysis. Background Technology

[0002] Street lighting is a crucial component of urban infrastructure. Traditional street lighting systems typically operate with fixed activation times and brightness levels. However, due to the long operating hours and wide coverage of these systems, this traditional approach results in significant energy consumption. As green energy conservation gains increasing importance, street lighting systems are gradually shifting from traditional timed control to intelligent control based on real-time monitoring data. This involves adjusting streetlight operating parameters in real time according to actual lighting needs, reducing energy consumption while ensuring road safety and improving energy efficiency.

[0003] Existing methods for energy-saving control of road lighting typically set fixed adjustment rules based on monitoring data such as ambient illuminance, vehicle flow, and pedestrian flow. These methods often acquire vehicle and pedestrian flow data separately using radar and infrared sensors. However, in complex urban road environments, vehicle and pedestrian activities are often highly coupled, leading to potential inaccuracies in sensor data and making it difficult to consistently reflect real road conditions. Furthermore, existing methods often adjust based on real-time monitoring data from a single location, resulting in a certain degree of lag and difficulty adapting to rapidly changing road conditions, thus impacting the travel experience for both vehicles and pedestrians.

[0004] Therefore, how to dynamically adjust the road lighting system to improve the timeliness and accuracy of road lighting adjustment has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a road lighting energy-saving optimization control method based on data analysis to solve the problem of how to dynamically adjust the road lighting system in order to improve the timeliness and accuracy of road lighting adjustment.

[0006] This invention provides a data analysis-based method for optimizing and controlling energy-saving road lighting, which includes the following steps: Identify the adjacent streetlights of the target streetlight and acquire multiple frames of overhead images of each adjacent streetlight within the current time period; For any adjacent street light, perform color difference analysis between adjacent pixels on each frame of the overhead image of the adjacent street light to divide each frame of the overhead image into multiple image blocks; by extracting the color feature vector of each image block, merge the image blocks in each frame of the overhead image to obtain the complete image block in the corresponding overhead image. Dynamic trajectory matching is performed on complete image blocks between adjacent overhead images to obtain multiple complete image block sets. Motion trajectory feature analysis is performed on each complete image block set to determine the dynamic pixels in each frame of overhead images. Based on the dynamic pixels of all overhead images, the activity intensity of the surrounding road of any adjacent street light in the current time period is calculated. The correlation between the surrounding road activity intensity of each adjacent street light and the target street light during the target historical period is analyzed to obtain the degree of influence of each adjacent street light on the target street light. Combining the surrounding road activity intensity of all adjacent street lights in the current period and the degree of influence of each adjacent street light on the target street light, the surrounding road activity intensity of the target street light in the future period is predicted to obtain the corresponding predicted value. The lighting intensity of the target street light is dynamically adjusted according to the predicted value.

[0007] Preferably, the step of performing color difference analysis between adjacent pixels on each frame of the overhead image of any adjacent street lamp to divide each frame of the overhead image into multiple image blocks includes: For any overhead image, calculate the absolute value of the color difference between each pair of adjacent pixels in the R, G, and B channels, respectively, and obtain the sum of the absolute values ​​of the color difference between the corresponding two adjacent pixels as the degree of color difference; Arrange the color difference between any two adjacent pixels in any aerial image in ascending order and construct a color difference change curve. Obtain the color difference degree corresponding to the inflection point of the color difference change curve as the color difference degree threshold. Divide adjacent pixels with color difference degree less than or equal to the color difference degree threshold into the same connected component. Divide any aerial image into multiple image blocks according to the divided connected components.

[0008] Preferably, the step of extracting the color feature vector of each image patch and merging the image patches in each frame of the overhead image to obtain the complete image patch in the corresponding overhead image includes: For any image block, based on the R-channel value, G-channel value, and B-channel value of each pixel in the image block, calculate the average R-channel value, average G-channel value, and average B-channel value. Using the sum of the average R-channel value, average G-channel value, and average B-channel value as the denominator, and using the average R-channel value, average G-channel value, and average B-channel value as the numerator, respectively, obtain the corresponding R-channel value ratio, G-channel value ratio, and B-channel value ratio, which together form the color feature vector of the image block. For any two adjacent image blocks in any overhead image, based on the color feature vectors of the two adjacent image blocks, calculate the absolute value of the difference in the proportion of R channel values, the absolute value of the difference in the proportion of G channel values, and the absolute value of the difference in the proportion of B channel values, respectively, and obtain the cumulative value of the absolute value of the difference. Use the negative of the cumulative value as the independent variable of an exponential function with the natural constant as the base to obtain the color feature similarity between the two adjacent image blocks. Obtain the color feature similarity between every two adjacent image blocks in all overhead images, cluster all color feature similarities to obtain the cluster centers of the two clusters, and take the minimum value of the color feature similarity corresponding to the cluster centers as the color feature similarity threshold; if the color feature similarity between any two adjacent image blocks is greater than or equal to the color feature similarity threshold, then merge any two adjacent image blocks to obtain a merged image block. Based on transitive logical relationships, all merged image blocks in any given overhead image are transitively merged to obtain multiple complete image blocks in the given overhead image.

[0009] Preferably, the dynamic trajectory matching of complete image patches between adjacent overhead images to obtain multiple sets of complete image patches includes: Taking the i-th complete image patch in the first frame of the overhead image as the starting point of trajectory analysis, calculate the color feature similarity between the i-th complete image patch and each complete image patch in the second frame of the overhead image. Take the complete image patches in the second frame of the overhead image with a color feature similarity greater than or equal to the color feature similarity threshold as candidate complete image patches of the i-th complete image patch. If the number of candidate complete image patches of the i-th complete image patch is 0, then check whether there are candidate complete image patches of the i-th complete image patch in each subsequent frame of the overhead image. If none of them exist, then the i-th complete image patch is treated as a separate set of complete image patches. If the number of candidate complete image blocks for the i-th complete image block is not 0, then calculate the edge similarity between the i-th complete image block and each of the candidate complete image blocks, and obtain the candidate complete image blocks whose edge similarity is greater than or equal to the preset edge similarity threshold, which are denoted as matching complete image blocks. If the number of matching complete image blocks is 1, then the matching complete image blocks are used as the continuous adjacent trajectory points of the i-th complete image block. If the number of matching complete image blocks is greater than 1, then according to the edge similarity corresponding to each matching complete image block, the matching complete image block with the maximum edge similarity is used as the continuous adjacent trajectory point of the i-th complete image block. If there are multiple matching complete image blocks with the maximum edge similarity, then calculate the distance between the centroid coordinates of each matching complete image block with the maximum edge similarity and the centroid coordinates of the i-th complete image block, and select the matching complete image block with the minimum distance as the continuous adjacent trajectory point of the i-th complete image block. According to the method for obtaining the continuous adjacent trajectory points of the i-th complete image block, the continuous adjacent trajectory points of the i-th complete image block are obtained in the third frame of the overhead image, and so on, to obtain a set of complete image blocks with the i-th complete image block as the starting point of trajectory analysis. After obtaining the set of complete image blocks with each complete image block in the first frame of the overhead image as the starting point of trajectory analysis, the non-contiguous adjacent trajectory points in the second frame of the overhead image are used as the starting point of trajectory analysis. Following the method for obtaining the set of complete image blocks with the i-th complete image block as the starting point of trajectory analysis, the set of complete image blocks with the non-contiguous adjacent trajectory points in the second frame of the overhead image as the starting point of trajectory analysis is obtained respectively. This process is repeated to divide the complete image blocks of all overhead images into multiple sets of complete image blocks.

[0010] Preferably, calculating the color feature similarity between the i-th complete image patch and each complete image patch in the second frame overhead image includes: For any complete image block in the second frame of the overhead image, the maximum color feature similarity is taken as the color feature similarity between any image block in the first complete image block and each image block in the i-th complete image block, based on the color feature similarity between any image block in the first complete image block and each image block in the i-th complete image block. The mean value of the color feature similarity between each image block in any complete image block and the i-th complete image block is calculated as the color feature similarity between any complete image block and the i-th complete image block.

[0011] Preferably, calculating the edge similarity between the i-th complete image patch and each of the candidate complete image patches includes: For any candidate complete image block, based on the distance between every two pixels in the candidate complete image block, obtain the two pixels corresponding to the maximum distance as the pixel pair to be rotated, connect the pixel pairs to be rotated with a straight line to obtain a connecting line to be rotated and the center point on the connecting line to be rotated, which is denoted as the center point to be rotated. Based on the distance between every two pixels in the i-th complete image block, the two pixels corresponding to the maximum distance are obtained as the reference pixel pair. The reference pixel pairs are connected by a straight line to obtain a reference connecting line and the center point on the reference line, which is denoted as the reference center point. With the i-th complete image block fixed, the center point to be rotated coincides with the reference center point. By rotating any candidate complete image block, the reference connecting line and the connecting line to be rotated coincide, resulting in two coincidence methods. The Hausdorff distance between the edge of the i-th complete image block and the edge of any candidate complete image block is obtained under each coincidence method. The negative of the minimum Hausdorff distance is taken as the independent variable of an exponential function with the natural constant as the base, to obtain the edge similarity between the i-th complete image block and any candidate complete image block.

[0012] Preferably, the step of performing motion trajectory feature analysis on each of the complete image patch sets to determine the dynamic pixels in each frame of overhead image includes: For any complete image patch set, if the number of complete image patches in the set is 1, then the pixels in the complete image patches in the set are marked as dynamic pixels. If the number of complete image blocks in any complete image block set is greater than 1, then obtain the centroid coordinates of each complete image block in the set in the corresponding overhead image, calculate the centroid distance between complete image blocks in adjacent overhead images, and obtain the average centroid distance. If the average centroid distance is greater than or equal to the preset static centroid distance, then the pixels in each complete image block in any complete image block set are marked as dynamic pixels; otherwise, the pixels in each complete image block in any complete image block set are marked as static pixels. Dynamic pixel marking is performed on each complete image block set to obtain the dynamic pixels in each frame of overhead image.

[0013] Preferably, the step of calculating the surrounding road activity intensity of any adjacent street light in the current time period based on the dynamic pixels of all overhead images includes: Based on the number of dynamic pixels in each frame of the overhead image, the proportion of dynamic pixels in each frame of the overhead image is obtained to obtain the average proportion of dynamic pixels. The average proportion of dynamic pixels is then normalized to obtain the activity intensity of the surrounding roads of any adjacent street light in the current time period.

[0014] Preferably, the step of analyzing the correlation between the intensity of surrounding road activity between each of the adjacent streetlights and the target streetlight during the target historical time period, to obtain the degree of influence of each of the adjacent streetlights on the target streetlight, includes: For any adjacent street light, obtain the surrounding road activity intensity sequence of the adjacent street light in the target historical time period and the surrounding road activity intensity sequence of the target street light in the target historical time period. Using the normalized cross-correlation function, calculate the maximum correlation coefficient and the delay time of the maximum correlation coefficient between the surrounding road activity intensity sequence of the adjacent street light in the target historical time period and the surrounding road activity intensity sequence of the target street light in the target historical time period. Based on the maximum correlation coefficient and the delay time of the maximum correlation coefficient between any adjacent street light and the target street light in multiple historical time periods belonging to the same historical time period as the target historical time period, calculate the average value of the maximum correlation coefficient and the average value of the delay time. use The function obtains the sign value corresponding to the average value of the delay time, normalizes the sign value to obtain the normalized sign value, and uses the product between the normalized sign value and the average value of the maximum correlation coefficient as the degree of influence of any adjacent street light on the target street light.

[0015] Preferably, the step of combining the surrounding road activity intensity of all adjacent streetlights in the current time period and the influence degree of each adjacent streetlight on the target streetlight to predict the surrounding road activity intensity of the target streetlight in future time periods, and obtaining the corresponding predicted value, includes: Using the degree of influence of each adjacent street light on the target street light as a weight, the activity intensity of the surrounding roads of all adjacent street lights in the current time period is weighted and summed to obtain the activity intensity adjustment value; the activity intensity of the surrounding roads of the target street light in the current time period is obtained, and the sum of the activity intensity of the surrounding roads of the target street light in the current time period and the activity intensity adjustment value is used as the predicted value of the activity intensity of the surrounding roads of the target street light in future time periods.

[0016] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: This invention effectively reduces interference from changes in lighting and shadows on image matching by dividing the overhead image of a streetlight into blocks, comparing and analyzing color proportions, and matching the overall shape of adjacent image blocks. This improves the accuracy of dynamic part recognition and makes the assessed road activity intensity more reflective of the real situation of traffic and pedestrian flow. Furthermore, by combining the road activity intensity around adjacent streetlights with the prediction of the road activity intensity around the target streetlight, it is possible to predict the flow of people and vehicles towards the target streetlight in advance, thereby adjusting the streetlight brightness in advance. This effectively solves the problem of lagging adjustment in traditional monitoring and adjustment, and achieves energy conservation and consumption reduction while ensuring road traffic safety. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a road lighting energy-saving optimization control method based on data analysis provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of adjacent streetlights of a target streetlight provided in an embodiment of the present invention; Figure 3 This is a schematic diagram showing the rotational overlap of the i-th complete image block and any candidate complete image block in an embodiment of the present invention. Detailed Implementation

[0019] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0020] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0021] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0022] See Figure 1This is a flowchart of a road lighting energy-saving optimization control method based on data analysis provided in Embodiment 1 of the present invention, as shown below. Figure 1 As shown, the method may include: Step S101: Determine the adjacent streetlights of the target streetlight and obtain an overhead image of each adjacent streetlight during the current time period.

[0023] Road environments are often complex, with mixed pedestrian and vehicular traffic, such as near intersections. Traditional road lighting control methods primarily use radar and infrared sensors to monitor pedestrian and vehicular traffic separately before assessing the required lighting intensity. However, in complex road conditions, pedestrian and vehicular traffic can interfere with each other, leading to inaccurate monitoring data and affecting the precision of lighting intensity adjustments. Furthermore, traditional road lighting control methods often adjust lighting only after pedestrian and vehicular traffic has passed, resulting in a certain lag. When vehicles pass quickly, the adjustment speed may not meet traffic demands, posing a road safety risk. Therefore, in this embodiment of the invention, road activity intensity is analyzed by combining changes in continuously collected images, and the road activity intensity at different streetlight locations is predicted based on the relationship between road activity intensity changes, thereby improving the timeliness and accuracy of road lighting adjustments.

[0024] Using a single streetlight as the target streetlight and a 5-second image acquisition period, the road activity intensity for the next 5 seconds is predicted using images acquired within that 5-second period. Since the prediction of the target streetlight also needs to consider the traffic conditions of adjacent streetlights, and the distance between adjacent streetlights is typically between 30m and 40m, while urban road vehicle speeds are usually around 40km / h, and vehicles typically only cross one streetlight interval within 5 seconds, the prediction of the target streetlight's road activity intensity only considers its adjacent streetlights, as follows: Figure 2 As shown, A represents the target street light, and a, b, c, d, and e represent the adjacent street lights of the target street light.

[0025] Before analyzing the target streetlights, multiple frames of overhead images of the target streetlight and each adjacent streetlight were acquired within the current time period (the most recent 5 seconds). For acquiring consecutive frames of overhead images of each streetlight within 5 seconds: a camera was positioned near the top of the streetlight pole, close to the lamp arm. The camera position should ensure that pedestrians and vehicles are clearly visible in the image, not too small, and avoid obstruction by large vehicles such as buses. The height of the camera above the ground was between 8 and 10 meters, which could be adjusted according to the actual streetlight height and road conditions. The camera was used at a 45° overhead angle to capture the largest possible target area while minimizing mutual obstruction by the targets (pedestrians and vehicles). Based on the normal traffic speed of pedestrians and vehicles on urban roads, an overhead image was acquired every 0.5 seconds to more clearly reflect dynamic changes in the image.

[0026] This allows us to obtain multiple overhead images of the target streetlight and each adjacent streetlight within the current time period.

[0027] Step S102: For any adjacent street light, perform color difference analysis between adjacent pixels in each frame of the overhead image of any adjacent street light to divide each frame of the overhead image into multiple image blocks; by extracting the color feature vector of each image block, merge the image blocks in each frame of the overhead image to obtain the complete image block in the corresponding overhead image.

[0028] The main purpose of adjusting the lighting intensity of streetlights is to ensure that pedestrians and vehicles have good visibility. When there are no pedestrians or vehicles passing by, the lighting intensity should be appropriately reduced to reduce the energy consumption of streetlights. That is, when there are dynamic changes in the collected overhead images, it indicates that there may be pedestrians and vehicles passing by in the current road section. Then, by analyzing the dynamic changes in the continuously collected overhead images, the intensity of road activity near each streetlight can be evaluated, that is, the intensity of surrounding road activity. The greater the intensity of surrounding road activity, the stronger the lighting intensity is required.

[0029] Since the analysis method for the intensity of surrounding road activity is consistent for each street light in the current time period, taking one adjacent street light as an example, denoted as any adjacent street light, the area that changes in the aerial image of any adjacent street light in the current time period may be vehicles and pedestrians. However, judging directly based on the color change of a single pixel cannot distinguish between the actual moving part and the background part that changes due to interference from the moving part. Therefore, it is necessary to divide each frame of the aerial image of any adjacent street light in the current time period into image blocks, and by analyzing similar image blocks in consecutive aerial images, track the dynamic trajectory of each region to identify the dynamically changing region.

[0030] Because the sharpness of aerial images varies under different weather conditions such as rain and fog, the difference in color between objects may vary at different times. Therefore, it is necessary to determine the color difference segmentation threshold for different regions based on the sharpness of the aerial images within the current time period. Then, each frame of the aerial image is divided into multiple image blocks using the color difference segmentation threshold. The specific process is as follows: For any overhead image, calculate the absolute value of the color difference between every two adjacent pixels in the R, G, and B channels, respectively. The sum of these absolute values ​​is used as the degree of color difference. "Adjacent" refers to pixels within an eight-neighborhood. The formula for calculating the degree of color difference between the u-th and v-th pixels in any overhead image is as follows: ; in, This represents the degree of color difference between the u-th pixel and the v-th pixel in the p-th frame of the overhead image, where the u-th pixel and the v-th pixel are adjacent pixels within their eight-neighborhood. This represents the T color channel value of the u-th pixel in the p-th frame of the overhead image (here, an RGB image is used, so the T color channel value corresponds to the R channel value, G channel value, and B channel value, respectively). This represents the T color channel value of the u-th pixel in the p-th frame of the overhead image. This represents the absolute value of the difference between the u-th pixel and the v-th pixel in the same color channel. The larger the absolute value of this difference, the greater the degree of color difference.

[0031] Similarly, the color difference between any two adjacent pixels in any overhead image is obtained. Since the difference between the same color is usually small, and although image sharpness can cause changes in the difference between different colors, the difference between the same color is usually less affected. Therefore, the color difference between any two adjacent pixels in any overhead image is arranged in ascending order, and a color difference change curve is constructed. The horizontal axis of the color difference change curve is the sort number, and the vertical axis is the color difference degree. The inflection point of the color difference change curve is obtained using the Kneedle algorithm. The color difference degree corresponding to the inflection point is used as the color difference degree threshold to distinguish between the same color and different colors. Then, adjacent pixels with a color difference degree less than or equal to the color difference degree threshold are divided into the same connected component. Based on the divided connected components, any overhead image is divided into multiple image blocks. It should be noted that, in order to avoid interference from noisy pixels, if there is a single pixel after generating the connected component, the average color of the three channels of the other pixels in the eight neighborhood of the single pixel is used as the color of the single pixel, and then it is assigned to the connected component to eliminate noisy pixels.

[0032] Typically, the light intensity is strongest under streetlights, and it gradually decreases with increasing distance from the streetlight. This means that the light intensity in aerial images taken from different locations may not be entirely consistent. Furthermore, the color of vehicles or pedestrians may vary depending on the light intensity they encounter. Additionally, reflections and shadows can cause the same object to be segmented into different image blocks in different aerial images. Therefore, matching solely based on the shape and color of image blocks may fail due to these factors. In this embodiment of the invention, it is necessary to first merge the image blocks in each frame of the aerial image to obtain the complete structure of the corresponding object as much as possible, facilitating dynamic trajectory matching between different aerial images.

[0033] Although changes in light intensity will cause changes in the R, G, and B channel values, their primary color characteristics usually do not change significantly. For example, a red car appears red under different light intensities, so its R channel value is relatively large compared to the other two color channel values. Therefore, firstly, based on the three channel values ​​(R channel value, G channel value, and B channel value) of each pixel, the color feature vector of each image block is extracted: For any image block, based on the R channel value, G channel value, and B channel value of each pixel in that image block, the average R channel value, average G channel value, and average B channel value are calculated. Using the sum of the average R channel value, average G channel value, and average B channel value as the denominator, and using the average R channel value, average G channel value, and average B channel value as the numerator, respectively, the corresponding R channel value ratio, G channel value ratio, and B channel value ratio are obtained, forming the color feature vector of that image block.

[0034] In one embodiment, the formula for calculating the color feature vector of any image patch is: ; ; ; in, Let w represent the color feature vector of the w-th image patch in the p-th frame of the overhead image. , and These represent the average R-channel value, average G-channel value, and average B-channel value of the w-th image patch in the p-th frame of the overhead image, respectively. If an image patch is black, i.e. , and If all values ​​are 0, then the color ratio for each channel should also be 0, i.e., set to 0. .

[0035] After extracting the color feature vector of each image block in any aerial image, the image blocks in any aerial image are merged based on the similarity of their color feature vectors. First, for any two adjacent image blocks in any aerial image (meaning adjacent within an eight-neighborhood), the absolute values ​​of the differences in the R-channel, G-channel, and B-channel values ​​are calculated based on their color feature vectors. The sum of these absolute values ​​is then obtained. The negative of this sum is used as the independent variable of an exponential function with the natural constant as its base to obtain the color feature similarity between the two adjacent image blocks. Let the color feature similarity between the w-th and x-th image blocks in the p-th frame aerial image be: ; in, This represents the color feature similarity between the w-th image patch and the x-th image patch in the p-th frame of the overhead image. This represents an element value in the color feature vector of the x-th image patch in the p-th frame of the overhead image. This represents an element value in the color feature vector of the w-th image patch in the p-th frame of the overhead image. It represents an exponential function with the natural constant e as the base, and || represents the absolute value symbol. This represents the absolute value of the difference between the element values ​​of the w-th and x-th image patches in the same color channel of the p-th frame overhead image. The smaller the absolute value of the difference between element values, the higher the color feature similarity between the w-th image patch and the x-th image patch.

[0036] Following the aforementioned method for obtaining color feature similarity, the color feature similarity between every two adjacent image blocks in all overhead images can be obtained. Then, the K-means clustering algorithm is used to cluster all color feature similarities, setting K=2 to obtain the cluster centers of the two clusters. The cluster with higher color feature similarity corresponding to the cluster center contains image blocks whose colors are compatible. Therefore, the minimum value among the color feature similarities corresponding to the cluster centers is taken as the color feature similarity threshold Y. If the color feature similarity between any two adjacent image blocks is greater than or equal to the color feature similarity threshold, it indicates that the two image blocks have similar dominant colors, differing only due to lighting and shadow effects. Therefore, any two adjacent image blocks are merged to obtain a merged image block. Conversely, if the color feature similarity is not merged, these two image blocks are treated as separate complete image blocks. Considering that the complete structure of an object may correspond to multiple image blocks, based on transitive logic, all merged image blocks in any overhead image are transitively merged to obtain multiple complete image blocks in any overhead image. For example: Suppose that image block 1 and image 2 can be merged into one merged image block, and image block 2 and image 3 can be merged into one merged image block, then image blocks 1, 2, and 3 will be merged into one complete image block.

[0037] Therefore, based on the method of dividing multiple complete image blocks in any aerial image, each frame of aerial image is divided into multiple complete image blocks.

[0038] Step S103: Perform dynamic trajectory matching on complete image blocks between adjacent overhead images to obtain multiple complete image block sets. Perform motion trajectory feature analysis on each complete image block set to determine the dynamic pixels in each frame of overhead images. Based on the dynamic pixels of all overhead images, calculate the surrounding road activity intensity of any adjacent street light in the current time period.

[0039] Since each complete image patch in an overhead view corresponds to the complete structure of an object, and to analyze the pedestrian and vehicular traffic flow of any adjacent streetlight within the current time period, considering that pedestrian and vehicular traffic are dynamic objects, dynamic trajectory matching is performed on the complete image patches between adjacent overhead views. This results in the complete image patches in all overhead views being divided into multiple sets, each set representing the dynamic trajectory of an object. The specific method for dynamic trajectory matching of complete image patches between adjacent overhead views is as follows:

[0040] Based on the acquisition order of aerial images of any adjacent streetlights within the current time period, taking the i-th complete image block in the first frame of the aerial image as the starting point for trajectory analysis, the color feature similarity between the i-th complete image block and each complete image block in the second frame of the aerial image is calculated: For any complete image block in the second frame of the aerial image, based on the color feature similarity between any image block in any complete image block and each image block in the i-th complete image block, the maximum color feature similarity is taken as the color feature similarity between any image block and the i-th complete image block; the mean of the color feature similarity between each image block in any complete image block and the i-th complete image block is calculated as the color feature similarity between any complete image block and the i-th complete image block.

[0041] In one embodiment, taking the h-th complete image patch in the second frame of the overhead view as an example, the formula for calculating the color feature similarity between the h-th complete image patch and the i-th complete image patch is as follows: ; in, This indicates the second frame of the overhead image. The color feature similarity between the h-th complete image patch and the i-th complete image patch. This indicates the number of image patches contained in the h-th complete image patch in the second overhead view. This represents the number of image patches contained in the i-th complete image patch. This represents the color feature similarity between the x-th image patch in the i-th complete image patch and the w-th image patch in the h-th complete image patch in the second frame overhead image. This represents the maximum value of the color feature similarity between the w-th image block in the h-th complete image block of the second frame overhead image and each image block in the i-th complete image block. This maximum value is used to represent the color feature similarity between the w-th image block in the h-th complete image block of the second frame overhead image and the i-th complete image block.

[0042] Similarly, the color feature similarity between each complete image patch in the second frame of the overhead image and the i-th complete image patch is obtained. Complete image patches in the second frame of the overhead image with a color feature similarity greater than or equal to the color feature similarity threshold Y are considered as candidate complete image patches for the i-th complete image patch. If the color feature similarity between a complete image patch in the second frame of the overhead image and the i-th complete image patch is not greater than or equal to Y, that is, if the number of candidate complete image patches corresponding to the i-th complete image patch in the second frame of the overhead image is 0, then it is considered that the i-th complete image patch cannot be matched with a complete image belonging to the same object structure in the second frame of the overhead image. The i-th complete image block indicates that the vehicle or pedestrian may have left the image range, or a stationary object may be obscured by a moving vehicle or pedestrian, making it impossible to match. Therefore, the i-th complete image block needs to be matched sequentially with the complete image blocks in each subsequent overhead image (the overhead image after the second overhead image). If the color feature similarity between the complete image block in each subsequent overhead image and the i-th complete image block is not greater than or equal to Y, it means that the i-th complete image block belongs to a dynamic object and has left the current image range. Therefore, the i-th complete image block is treated as a separate complete image block set.

[0043] If the number of candidate complete image patches corresponding to the i-th complete image patch in the second frame of the overhead image is not zero, then considering that there may be vehicle turning in the first and second frames of the overhead image, making it impossible to directly match the i-th complete image patch with the candidate complete image patches, the matching of the i-th complete image patch with the candidate complete image patches is completed by calculating the edge similarity between the i-th complete image patch and each of the candidate complete image patches: For any candidate complete image block in the second frame of the overhead view, based on the distance between every two pixels in the candidate complete image block, the two pixels corresponding to the maximum distance are obtained as the pixel pair to be rotated. These pixel pairs are then connected by a straight line to obtain a connecting line to be rotated and its center point, denoted as the center point to be rotated. Similarly, based on the distance between every two pixels in the i-th complete image block, the two pixels corresponding to the maximum distance are obtained as the reference pixel pair. These reference pixel pairs are then connected by a straight line to obtain a reference connecting line and its center point, denoted as the reference center point. The i-th complete image block is fixed in place, and the center point to be rotated and the reference center point are aligned. By rotating the candidate complete image block, the reference connecting line and the connecting line to be rotated are aligned, thus obtaining two alignment methods, as shown in the reference. Figure 3 , Figure 3In this diagram, A represents the i-th complete image block, B represents any candidate complete image block, 1 and 2 represent the reference pixel pair of the i-th complete image block, 3 and 4 represent the pixel pair to be rotated of any candidate complete image block, 5 represents the center point of the line connecting the reference pixel pairs of the i-th complete image block, and 6 represents the center point of the line connecting the pixel pairs to be rotated of any candidate complete image block. Overlap method 1 is as follows: and With the same direction, the second overlapping method is: and They are in the same direction.

[0044] It should be noted that the metrics used for analysis based on pixel coordinates, such as the distance between pixels and the centroid coordinates of the complete image patch, are all obtained from a two-dimensional image coordinate system constructed with the lower left corner of the overhead image as the origin.

[0045] The Hausdorff distance between the edge of the i-th complete image patch and the edge of any candidate complete image patch is obtained under each of the aforementioned overlap methods. The negative of the smallest Hausdorff distance is taken as the independent variable of an exponential function with the natural constant as the base, and the edge similarity between the i-th complete image patch and any candidate complete image patch is obtained.

[0046] The formula for calculating the edge similarity between the i-th complete image patch and the w-th candidate complete image patch is as follows: ; in, This represents the edge similarity between the i-th complete image patch and the w-th candidate complete image patch. and Let represent the Hausdorff distances between the edge of the i-th complete image patch and the edge of the w-th candidate complete image patch under the two overlap methods. Since the edge lengths of the two complete image patches may not be consistent due to factors such as occlusion, the Hausdorff distance is used here. This represents the minimum function; the smaller the Hausdorff distance, the more similar the edges. This represents an exponential function with the natural constant as its base.

[0047] Similarly, the edge similarity between the i-th complete image patch and each candidate complete image patch is obtained. The candidate complete image patch with an edge similarity greater than or equal to the preset edge similarity threshold is obtained and denoted as the matched complete image patch. The preset edge similarity threshold is obtained as follows: the edge similarity between each complete image patch in the first frame of the overhead image and its candidate complete image patch in the second frame of the overhead image is obtained. The K-means clustering method is used to cluster all edge similarities. K=2 is set to obtain the cluster with the larger edge similarity corresponding to the cluster center. The smallest edge similarity in the cluster is obtained as the edge similarity threshold. If the number of matching complete image patches is 1, then the matching complete image patch is taken as the continuous adjacent trajectory point of the i-th complete image patch. If the number of matching complete image patches is greater than 1, then according to the edge similarity corresponding to each matching complete image patch, the matching complete image patch with the maximum edge similarity is taken as the continuous adjacent trajectory point of the i-th complete image patch. If there are multiple matching complete image patches with the maximum edge similarity, then considering that the movement distance of vehicles and pedestrians is limited in a short period of time, the distance between the centroid coordinates of the matching complete image patch with the maximum edge similarity and the centroid coordinates of the i-th complete image patch is calculated, and the matching complete image patch with the minimum distance is selected as the continuous adjacent trajectory point of the i-th complete image patch.

[0048] At this point, the dynamic trajectory matching of the i-th complete image block in the second frame of the overhead image is completed. Then, following the method for obtaining the continuous adjacent trajectory points of the i-th complete image block, the continuous adjacent trajectory points of the i-th complete image block are obtained in the third frame of the overhead image. This process is repeated to obtain a set of complete image blocks with the i-th complete image block as the starting point for trajectory analysis. After obtaining the set of complete image blocks with each complete image block in the first frame of the overhead image as the starting point for trajectory analysis, the non-continuous adjacent trajectory points in the second frame of the overhead image are used as the starting point for trajectory analysis. Following the method for obtaining the set of complete image blocks with the i-th complete image block as the starting point for trajectory analysis, the sets of complete image blocks with the non-continuous adjacent trajectory points in the second frame of the overhead image as the starting point for trajectory analysis are obtained respectively. This process is repeated to divide the complete image blocks of all the overhead images into multiple sets of complete image blocks.

[0049] A complete set of image patches can represent that it corresponds to the same object, but it is also necessary to determine whether the object represented by each complete set of image patches is a dynamic object. Specifically, motion trajectory feature analysis is performed on each complete set of image patches to determine the dynamic pixels in each frame of overhead image. The determination method is as follows:

[0050] For any set of complete image patches, if the number of complete image patches in the set is 1, it indicates that the corresponding object is a dynamic object. This dynamic object exists only in one frame of the overhead image because it has moved out of the range of any adjacent street lamp. Therefore, the pixels in the complete image patches in the set are marked as dynamic pixels. If the number of complete image patches in the set is greater than 1, the centroid coordinates of each complete image patch in the set are obtained in the corresponding overhead image. The centroid distance between the complete image patches in adjacent overhead images is calculated to obtain the average centroid distance, which is used to characterize the probability that any set of complete image patches is a dynamic object.

[0051] Since the position of a stationary object typically does not change across multiple overhead images, the centroid position of any complete image patch set will hardly change. Therefore, based on the probability (average centroid distance) that all complete image patch sets are dynamic objects, the Ostu method is used to obtain a threshold for the average centroid distance as the static centroid distance Q. If the average centroid distance of any complete image patch set is greater than or equal to Q, then the object represented by any complete image patch set is considered dynamic, and the pixels in each complete image patch of any complete image patch set are marked as dynamic pixels. Conversely, if the average centroid distance is less than or equal to Q, the pixels in each complete image patch of any complete image patch set are marked as static pixels.

[0052] Similarly, each complete image patch set is traversed, and dynamic pixel markers are applied to each complete image patch set to obtain the dynamic pixels in each frame of the overhead image. Then, based on the dynamic pixels of all overhead images, the activity intensity of the surrounding road for any adjacent streetlight in the current time period is calculated: based on the number of dynamic pixels in each frame of the overhead image, the proportion of dynamic pixels in each frame of the overhead image is obtained to obtain the average proportion of dynamic pixels. This average proportion of dynamic pixels is then normalized to obtain the activity intensity of the surrounding road for any adjacent streetlight in the current time period.

[0053] The formula for calculating the activity intensity of the surrounding roads for any adjacent streetlight during the current time period is as follows: ; in, This indicates the activity intensity of the surrounding roads for any adjacent streetlight during the current time period. This indicates the number of overhead images of any adjacent streetlight within the current time period. This represents the number of dynamic pixels in the p-th overhead image of any adjacent street light within the current time period. This represents the number of pixels in the p-th overhead image of any adjacent street light within the current time period. This represents the percentage of dynamic pixels in the p-th overhead image of any adjacent streetlight within the current time period. A higher percentage of dynamic pixels indicates a higher intensity of road activity around any adjacent streetlight. This represents the maximum and minimum value normalization function.

[0054] Similarly, it is possible to obtain the activity intensity of the surrounding roads of the target street light during the current time period, as well as the activity intensity of the surrounding roads of each adjacent street light during the current time period.

[0055] Step S104: Analyze the correlation between the surrounding road activity intensity of each adjacent street light and the target street light in the target historical time period to obtain the degree of influence of each adjacent street light on the target street light; combine the surrounding road activity intensity of all adjacent street lights in the current time period and the degree of influence of each adjacent street light on the target street light to predict the surrounding road activity intensity of the target street light in the future time period to obtain the corresponding predicted value; and dynamically adjust the lighting intensity of the target street light according to the predicted value.

[0056] To analyze the propagation relationship between pedestrian and vehicle traffic at different street light locations, it is also necessary to combine historical data analysis. Since people's travel purposes differ at different times (such as weekdays and non-weekdays), the flow direction of pedestrians and vehicles may also be different. Therefore, the historical data within the most recent three days that are in the same week as the current time period is used as the historical data to be analyzed. Specifically, one day including the current time period is used as the target historical period, and the most recent three days that are in the same week as the current time period are used as multiple historical periods belonging to the same period as the target historical period. For example, if the current time period is Wednesday, then Wednesday is used as the target historical period, and the previous Wednesday, the Wednesday before that, and the Wednesday before that are used as multiple historical periods belonging to the same period as the target historical period.

[0057] Based on the method for obtaining the surrounding road activity intensity of each streetlight in the current time period, the surrounding road activity intensity of each streetlight in each time period within the target historical time period is obtained, thus forming a sequence of surrounding road activity intensity for each streetlight within the target historical time period. This allows us to obtain the surrounding road activity intensity sequences of each adjacent streetlight and the target streetlight within the target historical time period. Furthermore, we analyze the correlation between the surrounding road activity intensity of each adjacent streetlight and the target streetlight within the target historical time period to determine the degree of influence of each adjacent streetlight on the target streetlight.

[0058] For any adjacent street light, using a normalized cross-correlation function, calculate the maximum correlation coefficient X and the delay time of the maximum correlation coefficient between the surrounding road activity intensity sequence of the adjacent street light in the target historical time period and the surrounding road activity intensity sequence of the target street light in the target historical time period. Similarly, the maximum correlation coefficient and the delay time of the maximum correlation coefficient between any adjacent street light and the target street light in the surrounding road activity intensity sequence within each historical time period are obtained. Then, the average value of the maximum correlation coefficient and the average value of the delay time are calculated as the degree of correlation between the surrounding road activity intensity of any adjacent street light and the target street light. and delay time .

[0059] It should be noted that, based on the normal traffic speed of pedestrians and vehicles and the distance between streetlights, the delay time is set within the range of [-30s, 30s], and the interval of the delay time is consistent with the image acquisition time interval (0.5s). It is stipulated here that when the activity intensity sequence of the surrounding road corresponding to the target streetlight is delayed relative to the adjacent streetlight, its delay time is a positive number.

[0060] To avoid the situation where the main movement of vehicles and pedestrians is from the target streetlight towards an adjacent streetlight, and this is misinterpreted as the adjacent streetlight affecting the target streetlight, it is necessary to incorporate delay time. When the degree of correlation The larger the value, and the longer the delay time. When the value is positive, it indicates a greater influence of adjacent streetlights on the surrounding road activity intensity of the target streetlight within the current time period. Therefore, the influence of each adjacent streetlight on the road activity intensity near the streetlight to be predicted (target streetlight) is calculated: for any adjacent streetlight, using... The function obtains the average delay time of any two adjacent streetlights. The corresponding sign value is normalized to obtain a normalized sign value. The normalized sign value is then averaged with the maximum correlation coefficient. The product between them represents the degree of influence of any adjacent street light on the target street light.

[0061] In one embodiment, the formula for calculating the influence of the a-th adjacent street light on the target street light is: ; in, This indicates the degree of influence of the a-th adjacent street light on the target street light. This represents the average of the maximum correlation coefficients between the a-th and a-th adjacent streetlights. This represents the average delay time of the a-th adjacent street light. Represents a sign function, when hour, ,when hour, ,when hour, , This represents the maximum and minimum value normalization function. After normalization, The values ​​are 1, 0.5, and 0.

[0062] Furthermore, when predicting the surrounding road activity intensity of the target streetlight in the next time period, since vehicles and pedestrians near the target streetlight in the current time period may not all leave the current time period's range in the next time period (i.e., the future time period), the surrounding road activity intensity of the target streetlight in the future time period is predicted based on the propagation relationship of the surrounding road activity intensity between adjacent streetlights and the target streetlight. Specifically, by combining the surrounding road activity intensity of all adjacent streetlights in the current time period and the influence degree of each adjacent streetlight on the target streetlight, the surrounding road activity intensity of the target streetlight in the future time period is predicted, yielding the corresponding predicted value:

[0063] Using the degree of influence of each adjacent street light on the target street light as a weight, the activity intensity of the surrounding roads of all adjacent street lights in the current time period is weighted and summed to obtain the activity intensity adjustment value; according to the method for obtaining the activity intensity of the surrounding roads of adjacent street lights in the current time period, the activity intensity of the surrounding roads of the target street light in the current time period is obtained, and the sum of the activity intensity of the surrounding roads of the target street light in the current time period and the activity intensity adjustment value is used as the predicted value of the activity intensity of the surrounding roads of the target street light in future time periods.

[0064] The formula for calculating the predicted intensity of surrounding road activity around the target streetlight in future time periods is as follows: ; in, This represents the predicted intensity of road activity around the target streetlight in the future. This indicates the intensity of traffic activity on the surrounding roads during the current time period. This indicates the total number of adjacent streetlights to the target streetlight. This indicates the degree of influence of the a-th adjacent street light on the target street light. This indicates the activity intensity of the surrounding roads in the current time period for the a-th adjacent street light.

[0065] After obtaining the predicted intensity of surrounding road activity for the target streetlight in future time periods, the illumination intensity of the target streetlight can be dynamically adjusted based on the predicted values: when When it is assumed that there are few or no vehicles and pedestrians near the target streetlight, the first-level lighting intensity is adopted; when When it is considered that there are few vehicles and pedestrians near the target streetlight and the safety risk is low, a level two lighting intensity is adopted; when At that time, it was assumed that vehicles and pedestrians frequently passed near the target streetlights. To ensure traffic safety, a three-level lighting intensity was adopted. The higher the lighting intensity level, the higher the corresponding lighting brightness. For example, Level 1 uses 50% lighting brightness, Level 2 uses 80% lighting brightness, and Level 3 uses 100% lighting brightness.

[0066] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A data analysis-based method for optimizing and controlling energy-saving road lighting, characterized in that, The method includes: Identify the adjacent streetlights of the target streetlight and acquire multiple frames of overhead images of each adjacent streetlight within the current time period; For any adjacent street light, perform color difference analysis between adjacent pixels on each frame of the overhead image of the adjacent street light to divide each frame of the overhead image into multiple image blocks; by extracting the color feature vector of each image block, merge the image blocks in each frame of the overhead image to obtain the complete image block in the corresponding overhead image. Dynamic trajectory matching is performed on complete image blocks between adjacent overhead images to obtain multiple complete image block sets. Motion trajectory feature analysis is performed on each complete image block set to determine the dynamic pixels in each frame of overhead images. Based on the dynamic pixels of all overhead images, the activity intensity of the surrounding road of any adjacent street light in the current time period is calculated. The correlation between the surrounding road activity intensity of each adjacent street light and the target street light during the target historical period is analyzed to obtain the degree of influence of each adjacent street light on the target street light. Combining the surrounding road activity intensity of all adjacent street lights in the current period and the degree of influence of each adjacent street light on the target street light, the surrounding road activity intensity of the target street light in the future period is predicted to obtain the corresponding predicted value. The lighting intensity of the target street light is dynamically adjusted according to the predicted value.

2. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The step of performing color difference analysis between adjacent pixels on each frame of the overhead image of any adjacent street lamp to divide each frame of the overhead image into multiple image blocks includes: For any overhead image, calculate the absolute value of the color difference between each pair of adjacent pixels in the R, G, and B channels, respectively, and obtain the sum of the absolute values ​​of the color difference between the corresponding two adjacent pixels as the degree of color difference; Arrange the color difference between any two adjacent pixels in any aerial image in ascending order and construct a color difference change curve. Obtain the color difference degree corresponding to the inflection point of the color difference change curve as the color difference degree threshold. Divide adjacent pixels with color difference degree less than or equal to the color difference degree threshold into the same connected component. Divide any aerial image into multiple image blocks according to the divided connected components.

3. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The process involves extracting the color feature vector of each image patch, merging the image patches in each frame of the overhead view to obtain the complete image patch in the corresponding overhead view image, including: For any image block, based on the R-channel value, G-channel value, and B-channel value of each pixel in the image block, calculate the average R-channel value, average G-channel value, and average B-channel value. Using the sum of the average R-channel value, average G-channel value, and average B-channel value as the denominator, and using the average R-channel value, average G-channel value, and average B-channel value as the numerator, respectively, obtain the corresponding R-channel value ratio, G-channel value ratio, and B-channel value ratio, which together form the color feature vector of the image block. For any two adjacent image patches in any overhead image, calculate the absolute value of the difference in the proportion of R channel values, the absolute value of the difference in the proportion of G channel values, and the absolute value of the difference in the proportion of B channel values ​​between the color feature vectors of the two adjacent image patches, obtain the cumulative value of the absolute difference, and use the negative of the cumulative value as the independent variable of an exponential function with the natural constant as the base to obtain the color feature similarity between the two adjacent image patches. Obtain the color feature similarity between every two adjacent image blocks in all overhead images, cluster all color feature similarities to obtain the cluster centers of the two clusters, and take the minimum value of the color feature similarity corresponding to the cluster centers as the color feature similarity threshold; if the color feature similarity between any two adjacent image blocks is greater than or equal to the color feature similarity threshold, then merge any two adjacent image blocks to obtain a merged image block. Based on transitive logical relationships, all merged image blocks in any given overhead image are transitively merged to obtain multiple complete image blocks in the given overhead image.

4. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The dynamic trajectory matching of complete image blocks between adjacent overhead images yields multiple sets of complete image blocks, including: Taking the i-th complete image patch in the first frame of the overhead image as the starting point of trajectory analysis, calculate the color feature similarity between the i-th complete image patch and each complete image patch in the second frame of the overhead image. Take the complete image patches in the second frame of the overhead image with a color feature similarity greater than or equal to the color feature similarity threshold as candidate complete image patches of the i-th complete image patch. If the number of candidate complete image patches of the i-th complete image patch is 0, then check whether there are candidate complete image patches of the i-th complete image patch in each subsequent frame of the overhead image. If none of them exist, then the i-th complete image patch is treated as a separate set of complete image patches. If the number of candidate complete image blocks for the i-th complete image block is not 0, then calculate the edge similarity between the i-th complete image block and each of the candidate complete image blocks, and obtain the candidate complete image blocks whose edge similarity is greater than or equal to the preset edge similarity threshold, which are denoted as matching complete image blocks. If the number of matching complete image blocks is 1, then the matching complete image blocks are used as the continuous adjacent trajectory points of the i-th complete image block. If the number of matching complete image blocks is greater than 1, then according to the edge similarity corresponding to each matching complete image block, the matching complete image block with the maximum edge similarity is used as the continuous adjacent trajectory point of the i-th complete image block. If there are multiple matching complete image blocks with the maximum edge similarity, then calculate the distance between the centroid coordinates of each matching complete image block with the maximum edge similarity and the centroid coordinates of the i-th complete image block, and select the matching complete image block with the minimum distance as the continuous adjacent trajectory point of the i-th complete image block. According to the method for obtaining the continuous adjacent trajectory points of the i-th complete image block, the continuous adjacent trajectory points of the i-th complete image block are obtained in the third frame of the overhead image, and so on, to obtain a set of complete image blocks with the i-th complete image block as the starting point of trajectory analysis. After obtaining the set of complete image blocks with each complete image block in the first frame of the overhead image as the starting point of trajectory analysis, the non-contiguous adjacent trajectory points in the second frame of the overhead image are used as the starting point of trajectory analysis. Following the method for obtaining the set of complete image blocks with the i-th complete image block as the starting point of trajectory analysis, the set of complete image blocks with the non-contiguous adjacent trajectory points in the second frame of the overhead image as the starting point of trajectory analysis is obtained respectively. This process is repeated to divide the complete image blocks of all overhead images into multiple sets of complete image blocks.

5. The road lighting energy-saving optimization control method based on data analysis according to claim 4, characterized in that, The calculation of the color feature similarity between the i-th complete image patch and each complete image patch in the second frame overhead image includes: For any complete image block in the second frame of the overhead image, the maximum color feature similarity is taken as the color feature similarity between any image block in the first complete image block and each image block in the i-th complete image block, based on the color feature similarity between any image block in the first complete image block and each image block in the i-th complete image block. The mean value of the color feature similarity between each image block in any complete image block and the i-th complete image block is calculated as the color feature similarity between any complete image block and the i-th complete image block.

6. The road lighting energy-saving optimization control method based on data analysis according to claim 4, characterized in that, The calculation of the edge similarity between the i-th complete image patch and each of the candidate complete image patches includes: For any candidate complete image block, based on the distance between every two pixels in the candidate complete image block, obtain the two pixels corresponding to the maximum distance as the pixel pair to be rotated, connect the pixel pairs to be rotated with a straight line to obtain a connecting line to be rotated and the center point on the connecting line to be rotated, which is denoted as the center point to be rotated. Based on the distance between every two pixels in the i-th complete image block, the two pixels corresponding to the maximum distance are obtained as the reference pixel pair. The reference pixel pairs are connected by a straight line to obtain a reference connecting line and the center point on the reference line, which is denoted as the reference center point. With the i-th complete image block fixed, the center point to be rotated coincides with the reference center point. By rotating any candidate complete image block, the reference connecting line and the connecting line to be rotated coincide, resulting in two coincidence methods. The Hausdorff distance between the edge of the i-th complete image block and the edge of any candidate complete image block is obtained under each coincidence method. The negative of the minimum Hausdorff distance is taken as the independent variable of an exponential function with the natural constant as the base, to obtain the edge similarity between the i-th complete image block and any candidate complete image block.

7. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The step of performing motion trajectory feature analysis on each complete image patch set to determine the dynamic pixels in each frame of overhead image includes: For any complete image patch set, if the number of complete image patches in the set is 1, then the pixels in the complete image patches in the set are marked as dynamic pixels. If the number of complete image blocks in any complete image block set is greater than 1, then obtain the centroid coordinates of each complete image block in the set in the corresponding overhead image, calculate the centroid distance between complete image blocks in adjacent overhead images, and obtain the average centroid distance. If the average centroid distance is greater than or equal to the preset static centroid distance, then the pixels in each complete image block in any complete image block set are marked as dynamic pixels; otherwise, the pixels in each complete image block in any complete image block set are marked as static pixels. Dynamic pixel marking is performed on each complete image block set to obtain the dynamic pixels in each frame of overhead image.

8. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The step of calculating the surrounding road activity intensity of any adjacent street light in the current time period based on the dynamic pixels of all overhead images includes: Based on the number of dynamic pixels in each frame of the overhead image, the proportion of dynamic pixels in each frame of the overhead image is obtained to obtain the average proportion of dynamic pixels. The average proportion of dynamic pixels is then normalized using a maximum-minimum value normalization function to obtain the activity intensity of the surrounding roads of any adjacent street light in the current time period.

9. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The analysis of the correlation between the surrounding road activity intensity of each adjacent street light and the target street light within the target historical time period yields the degree of influence of each adjacent street light on the target street light, including: For any adjacent street light, obtain the surrounding road activity intensity sequence of the adjacent street light in the target historical time period and the surrounding road activity intensity sequence of the target street light in the target historical time period. Using the normalized cross-correlation function, calculate the maximum correlation coefficient and the delay time of the maximum correlation coefficient between the surrounding road activity intensity sequence of the adjacent street light in the target historical time period and the surrounding road activity intensity sequence of the target street light in the target historical time period. Based on the maximum correlation coefficient and the delay time of the maximum correlation coefficient between any adjacent street light and the target street light in multiple historical time periods belonging to the same historical time period as the target historical time period, calculate the average value of the maximum correlation coefficient and the average value of the delay time. use The function obtains the sign value corresponding to the average value of the delay time, normalizes the sign value to obtain a normalized sign value, and uses the product of the normalized sign value and the average value of the maximum correlation coefficient as the degree of influence of any adjacent street light on the target street light. ； in, This indicates the degree of influence of the a-th adjacent street light on the target street light. This represents the average of the maximum correlation coefficients between the a-th and a-th adjacent streetlights. This represents the average delay time of the a-th adjacent street light. Represents a sign function, when hour, ,when hour, ,when hour, , This represents the normalization function for maximum and minimum values.

10. The road lighting energy-saving optimization control method based on data analysis according to claim 1, characterized in that, The method combines the surrounding road activity intensity of all adjacent streetlights in the current time period with the influence of each adjacent streetlight on the target streetlight to predict the surrounding road activity intensity of the target streetlight in future time periods, obtaining the corresponding predicted value, including: Using the degree of influence of each adjacent street light on the target street light as a weight, the activity intensity of the surrounding roads of all adjacent street lights in the current time period is weighted and summed to obtain the activity intensity adjustment value; the activity intensity of the surrounding roads of the target street light in the current time period is obtained, and the sum of the activity intensity of the surrounding roads of the target street light in the current time period and the activity intensity adjustment value is used as the predicted value of the activity intensity of the surrounding roads of the target street light in future time periods.

Images