Satellite video urban traffic flow estimation method for complex scene occlusion

Abstract

The application relates to a satellite video urban traffic flow estimation method for a complex shielding scene, which comprises the following steps: interframe registration and target region cutting of satellite video pretreatment; motion vehicle detection based on background modeling and road information extraction; tracking of the motion vehicle and reconstruction of the vehicle in the shielding area; traffic flow parameter estimation based on the detected and tracked vehicle. When the satellite video traffic flow estimation is performed, the application considers the shielded vehicle in the area, reconstructs the shielded vehicle through the motion vehicle tracking mode, and effectively improves the accuracy of the urban traffic flow estimation of the video satellite. Through the multidisciplinary intersection of remote sensing science, computer science and traffic transportation science, the application can truly and accurately simulate and reverse the urban traffic volume, and has important significance for formulating effective traffic policies and improving the urban traffic management level.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of satellite remote sensing image processing, intelligent transportation, computer vision, etc., and specifically relates to a satellite video urban traffic flow estimation method for complex, occluded scenarios. Background Technology

[0002] Video satellites, a novel method of Earth observation that has emerged in recent years, offer ultra-high spatial resolution and continuous imaging capabilities. They enable the recording of real-time dynamic information about the Earth's surface in a clearer and more intuitive way, creating opportunities for the macroscopic, visual, and dynamic extraction and display of urban traffic flow information. This has made them a hot topic in the fields of video remote sensing and intelligent transportation. In their paper, "Vehicle Detection and Traffic Density Monitoring from Very High Resolution Satellite Video Data. 2015 IEEE International Geoscience and Remote Sensing Symposium. 2015, 1881-1884," Kopsiaftis et al. used Skysat satellite video and an average value modeling method to detect moving vehicles and study traffic volume and density in the Las Vegas area. Ahmadi et al., in their paper "Moving Vehicle Detection, Tracking and Traffic Parameter Estimation from a Satellite Video: A Perspective on a Smarter City," *International Journal of Remote Sensing*, 2019, 40(22): 8379-8394, used a combination of median modeling and morphological methods to detect moving vehicles and study traffic flow parameters based on video footage of the Las Vegas area taken by Skysat satellite. Domestically, Luo Yile et al., in their paper "Research on Traffic Flow Parameter Extraction from Satellite Video Based on Optical Flow Method," *Computer Engineering and Applications*, 2018, 54(10): 204-207, used optical flow to extract traffic flow parameters based on satellite video of the Boston area provided by UrtheCast. These methods all detect traffic flow in open scenes and do not consider the impact of occlusion factors such as tall buildings, overpasses, and shadows on the estimation of urban traffic flow parameters, resulting in distortions in the estimation of traffic volume, speed, and density. Summary of the Invention

[0003] This invention addresses the shortcomings of existing technologies by proposing a satellite video-based urban traffic flow estimation method for complex, occluded scenarios. The aim is to mitigate the impact of occlusion in satellite video on ground traffic flow statistics. This method reconstructs vehicle targets within occluded areas using target tracking technology. By considering both vehicle targets detected in open areas and those within occluded areas, the accuracy of traffic flow parameter estimation is improved.

[0004] To achieve the above-mentioned objectives, the technical solution of this invention is as follows:

[0005] This invention discloses a satellite video method for estimating urban traffic flow in complex, occluded scenarios, comprising:

[0006] Preprocessing of satellite video yields preprocessed video images;

[0007] Extract the road range from the preprocessed video image;

[0008] Based on the road range, moving vehicles are detected in the preprocessed video images to obtain moving vehicle detection results;

[0009] Based on the vehicle detection results and the road range, the moving vehicle is tracked to obtain its tracking trajectory.

[0010] Based on the detection results of moving vehicles and the tracking trajectory of moving vehicles, the number of vehicles in the target area is counted, and the number of vehicles in the target area is obtained.

[0011] Based on the number of vehicles within the target area and the road range, the regional traffic density is obtained;

[0012] Calculate the traffic speed within the target area based on the tracking trajectory of the moving vehicle and the regional traffic density;

[0013] Based on the regional traffic density and traffic speed, calculate the traffic volume at a certain location or section of the road.

[0014] In the above traffic flow estimation method, the preprocessing of satellite video specifically includes: performing inter-frame registration and target region cropping on the satellite video, wherein the inter-frame registration adopts SIFT, ORB or AKAZA registration algorithm.

[0015] In the above traffic flow estimation method, the extraction of road ranges from the images is performed using D-LinkNet and RoadNet convolutional neural network algorithms.

[0016] In the above traffic flow estimation method, the moving vehicle detection specifically involves using the ViBe background modeling method in conjunction with the extracted road range to detect moving vehicles.

[0017] In the above traffic flow estimation method, the specific method for tracking moving vehicles to obtain their tracking trajectories is as follows:

[0018] Step S41: Construct the bounding rectangle of target j as (x j ,y j ,r j ,c j ), where x j With y j Let r be the position of the top-left corner of the target's bounding rectangle. j With c j These are the length and width of the circumscribed rectangle, respectively.

[0019] Step S42: Construct tracking trajectory information M;

[0020] Step S43: Based on the target j and the tracking trajectory information M, determine whether the target j is in a masked state. If yes, proceed to step S44; otherwise, proceed to step S45.

[0021] Step S44: Determine whether the target is lost. If so, the trajectory status is retained for a delay. The spatial position of the target in the lost frame is reconstructed based on the tracked trajectory information M. Proceed to step S45. If not, proceed to step S46.

[0022] Step S45: When the target reappears after crossing the obscured area, determine whether the target matches the delayed trajectory. If yes, proceed to step S46; otherwise, proceed to step S44.

[0023] Step S46: Update the trajectory status.

[0024] In the above traffic flow estimation method, the construction of the tracking trajectory information M specifically involves:

[0025] M = (ID, KF, AGE, TAC, CIC)

[0026] Among them, ID is the unique identifier of the trajectory, and the same ID in different frames represents the same target; KF represents the target's state information, recording the target's position and velocity information in the current frame; AGE represents the total number of frames since the trajectory was first detected; TAC represents the number of frames in which the trajectory has been recorded, used to characterize whether the trajectory is continuous between frames; CIC represents the frame difference since the trajectory was last recorded, used to characterize the time the target has been lost.

[0027] In the above traffic flow estimation method, the spatial position of the target in the lost frame is reconstructed based on the tracked trajectory information M. Specifically, the lost frame of the target is obtained using AGE, TAC, and CIC, and the target position relationship recorded by the state parameter KF is combined to reconstruct the spatial position of the target in the lost frame.

[0028] In the above traffic flow estimation method, the step of calculating the number of vehicles in the target area based on the detection results of moving vehicles and their tracking trajectories is as follows:

[0029] C = C1 + C2

[0030]

[0031]

[0032] Where C represents the number of vehicles detected within the region, C1 represents the number of vehicles within the unmasked region, and C2 represents the number of vehicles reconstructed within the masked region; [-t,t] represents the time interval between the target entering and leaving the masked region; n is the minimum threshold for determining that the target is a real target rather than a transient false target on the frame; AGE i CIC i This indicates that the target is discontinuous across time-series frames; TAC i =0 indicates that the trajectory was not recorded in frame i; TAC i+t ≠0 and TAC i-t ≠0 indicates that the target exists within t frames before and after the i-th frame.

[0033] In the above traffic flow estimation method, the step of obtaining the regional traffic density based on the number of vehicles in the target area and the road range specifically involves:

[0034] K i =N i / L

[0035] Where, N i The number of vehicles detected in the target area on the i-th frame is converted into the number of standard vehicles; L is the length of the road in the target area, and K is the number of standard vehicles. i Traffic density.

[0036] In the traffic flow estimation method described above, the traffic speed within the target area is calculated based on the tracking trajectory of moving vehicles and the regional traffic density, specifically as follows:

[0037]

[0038] v j =d / Δt

[0039] Δt=n*t s

[0040] Where m is the number of vehicles in the target area during the observation time, n is the number of frames, and t is the number of vehicles in the observation time. s d is the frame rate; d is the distance traveled by vehicle j within the observation time; and Δt is the observation time interval.

[0041] In the above traffic flow estimation method, the calculation of traffic volume at a specific location or cross-section on the road based on traffic density and traffic speed specifically involves:

[0042] Q = v × K

[0043] Where Q is traffic volume, v is speed, and K is traffic density.

[0044] The advantages of this invention over the prior art are as follows:

[0045] (1) This invention, through the analysis of the trajectory formed by moving vehicle tracking technology, can effectively remove false moving targets and reconstruct vehicle targets within the obscured area, which is a technological innovation of this invention. By including vehicles within the obscured area and comprehensively calculating vehicle targets detected in both open and obscured areas, the accuracy of satellite video in estimating traffic flow parameters is effectively improved, which is not considered in other current satellite video traffic flow parameter estimation methods.

[0046] (2) This invention utilizes emerging video satellite remote sensing data to combine computer vision moving target detection and tracking methods with traffic models. Through the interdisciplinary integration of remote sensing science, computer science, and transportation science, it can realistically and accurately simulate and invert urban traffic volume, which is of great significance for formulating effective traffic policies and improving the level of urban traffic management. Attached Figure Description

[0047] Figure 1 This is a flowchart of a satellite video urban traffic flow estimation method for complex, obscured scenarios, according to the present invention.

[0048] Figure 2 This is a flowchart illustrating a satellite video method for estimating urban traffic flow in complex, obscured scenarios, according to the present invention.

[0049] Figure 3 Image data map for traffic flow estimation based on Jilin-1 satellite video, which is one embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of road extraction results according to one embodiment of the present invention;

[0051] Figure 5This is a schematic diagram illustrating the detection results of a moving vehicle according to one embodiment of the present invention;

[0052] Figure 6 This is a schematic diagram illustrating the tracking results of a moving vehicle according to one embodiment of the present invention. Detailed Implementation

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0054] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The embodiments cannot be described in detail here, but the embodiments of the present invention are not limited to the following embodiments.

[0055] This invention discloses a satellite video method for estimating urban traffic flow in complex, occluded scenarios, comprising:

[0056] Step 1: Preprocess the satellite video to obtain preprocessed video images. Preprocessing specifically involves inter-frame registration and target region cropping, i.e., using the previous frame as a reference to align the subsequent frame. In the aligned video images, the target region for traffic flow detection is cropped out. Inter-frame registration uses SIFT, ORB, or AKAZA registration algorithms.

[0057] Step 2: In the preprocessed video images, the D-LinkNet and RoadNet convolutional neural network algorithms are used to extract the road range in the images; urban roads are the basis for extracting traffic volume parameters of the target area.

[0058] Step 3: Based on the road range, the ViBe background modeling method is used in the preprocessed video image to detect moving vehicles in combination with the extracted road range, and the moving vehicle detection results are obtained. A background model is constructed based on the spatiotemporal neighborhood random sampling method. The current frame is compared with the background model. By setting a certain threshold and combining road information and morphological processing methods, the vehicle target and the background are segmented.

[0059] Step 4: Based on the moving vehicle detection results and road range, track the moving vehicles using Kalman filtering and the Hungarian algorithm to obtain their tracking trajectories. Perform trajectory creation, state update, deletion, and delay retention operations based on the matching results. Under unoccluded conditions, the target can usually be tracked stably; however, under occlusion conditions, the target may be lost. In this state, the trajectory enters delay retention. When the target reappears after crossing the occlusion area, it is matched with the retained trajectory again. If a successful match is achieved, the trajectory's state is updated. The specific method is as follows:

[0060] Step S41: Construct the bounding rectangle of target j as (x j ,y j ,r j ,c j ), where x j With y j Let r be the position of the top-left corner of the target's bounding rectangle. j With c j These are the length and width of the circumscribed rectangle, respectively.

[0061] Step S42: Construct the tracking trajectory information M, specifically as follows:

[0062] M = (ID, KF, AGE, TAC, CIC)

[0063] Here, ID is the unique identifier of the trajectory, and the same ID in different frames represents the same target; KF represents the target's state information, recording the target's position and velocity information in the current frame; AGE represents the total number of frames since the trajectory was first detected; TAC represents the number of frames the trajectory has been recorded, used to characterize whether the trajectory is continuous between frames; and CIC represents the frame difference since the trajectory was last recorded, used to characterize the time when the target was lost. The tracked trajectory information is expressed as (ID, KF, AGE, TAC, CIC), which records the precise time information of the detection and loss of the same ID target. Using AGE, TAC, and CIC, the lost frames of the target can be obtained. Combined with the target position relationship recorded by the state parameter KF, the spatial position of the target in the lost frames can be reconstructed. It should be noted that this method can not only reconstruct targets in occluded areas, but also has a certain ability to handle target loss caused by the target's fluctuating brightness and trajectory anomalies caused by false moving targets.

[0064] Step S43: Based on the target j and the tracking trajectory information M, determine whether the target j is in a masked state. If yes, proceed to step S44; otherwise, proceed to step S45.

[0065] Step S44: Determine whether the target is lost. If so, the trajectory status is retained for a delay. The spatial position of the target in the lost frame is reconstructed based on the tracked trajectory information M. Proceed to step S45. If not, proceed to step S46.

[0066] The spatial position of the target in the lost frame is reconstructed based on the tracked trajectory information M. Specifically, the lost frame of the target is obtained using AGE, TAC, and CIC, and the target position relationship recorded by the state parameter KF is combined to reconstruct the spatial position of the target in the lost frame.

[0067] Step S45: When the target reappears after crossing the obscured area, determine whether the target matches the delayed trajectory. If yes, proceed to step S46; otherwise, proceed to step S44.

[0068] Step S46: Update the trajectory status.

[0069] Step 5: Utilize tracking trajectories to remove false moving targets and reconstruct targets in occluded areas. Based on the detected moving vehicles and the reconstructed vehicles, and using the moving vehicle detection results and tracking trajectories, count the number of vehicles within the target area to obtain the total number of vehicles in the target area. Specifically:

[0070] C = C1 + C2

[0071]

[0072]

[0073] Where C represents the number of vehicles detected within the region, C1 represents the number of vehicles within the unmasked region, and C1 is calculated by counting the number of valid IDs in the corresponding frame, where valid means that the target has been continuously detected in the previous n consecutive frames; C2 represents the number of reconstructed vehicles within the masked region, and C2 is calculated by counting the number of intermittently lost trajectories within a certain time range between consecutive frames; [-t, t] represents the time interval between the target entering and leaving the masked region; n represents the minimum threshold for determining that the target is a real target rather than a transient false target in the frame; AGE i CIC i This indicates that the target is discontinuous across time-series frames, meaning there may be occlusion present; TAC i =0 indicates that the trajectory was not recorded in frame i, meaning the target may be lost in frame i; TAC i+t ≠0 and TAC i-t ≠0 means that the target appears in both frames t before and after the i-th frame. If the above requirement is met, it can be confirmed that the target was lost in the current frame.

[0074] Step 6: Based on the number of vehicles in the target area and the road network, obtain the regional traffic density, specifically:

[0075] K i =N i / L

[0076] Where, N i The number of vehicles detected in the target area on the i-th frame is converted into the number of standard vehicles; L is the length of the road in the target area, and K is the number of standard vehicles. i Traffic density. The conversion factors for small cars (3.5m-7m), medium-sized cars (7m-10m), and large cars (over 10m) are 1, 2, and 3, respectively.

[0077] Step 7: Analyze the tracking trajectory of the moving vehicle. Based on the vehicle's position in the preceding and following frames, determine the vehicle's travel distance, taking into account the inter-frame time interval. Then, based on the tracking trajectory of the moving vehicle and the regional traffic density, calculate the traffic speed within the target area. Specifically:

[0078]

[0079] v j =d / Δt

[0080] Δt=n*t s

[0081] Where m is the number of vehicles in the target area during the observation time, n is the number of frames, and t is the number of vehicles in the observation time. s d is the frame rate; d is the distance traveled by vehicle j within the observation time; and Δt is the observation time interval.

[0082] Step 8: Based on regional traffic density and traffic speed, calculate the traffic volume at a specific location or cross-section on the road, specifically as follows:

[0083] Q = v × K

[0084] Where Q is traffic volume, v is speed, and K is traffic density.

[0085] Example:

[0086] To address the impact of occlusion phenomena such as tall buildings, overpasses, and shadows in satellite video on ground traffic flow estimation, this implementation method fully utilizes the dynamic changes of vehicle targets between satellite video frames. Through moving target detection and tracking, it reconstructs the traffic flow within the occluded area, and on this basis, realizes the statistics of traffic density, traffic speed, and traffic volume. The following will use video imagery of the Atlanta area of ​​the United States taken by Jilin-1 Video 03 satellite of Changguang Satellite Technology Co., Ltd. in May 2017 as an example to illustrate the specific implementation method of this invention.

[0087] Figure 1 , Figure 2 The following describes the general execution flow of each step in the satellite video urban traffic flow estimation method for complex, obscured scenarios in this embodiment, as well as the specific execution steps and processes within each step.

[0088] like Figure 1 and Figure 2 As shown, this satellite video method for estimating urban traffic flow in complex, occluded scenarios comprises four processes: data preprocessing, moving vehicle detection, moving vehicle tracking and vehicle reconstruction in occluded areas, and traffic flow parameter estimation. The implementation steps of this method are as follows:

[0089] Step 1: Perform preprocessing on the satellite video, including inter-frame registration and target region cropping. This involves using the previous frame as a reference to align the subsequent frame. From the aligned video image, the target region for traffic flow detection is cropped out. This region is selected as the highway area near Atlanta Airport, such as... Figure 3 As shown.

[0090] Step 2: In the preprocessed video image, extract the road area based on the D-LinkNet convolutional neural network algorithm. The road extraction results are as follows: Figure 4 As shown.

[0091] Step 3: Detect moving vehicles in the preprocessed video imagery. A background sample set is constructed based on 20 randomly sampled samples from the spatiotemporal neighborhood. The current frame is compared with the background model one by one, and a threshold of 20 is set as the segmentation identifier for potential moving vehicles and the background. Combined with road information and morphological processing methods, vehicle targets are segmented from the background. In this embodiment, the morphological processing method for ground vehicles is as follows: in Jilin-1 satellite imagery, the length is typically between 3-32 pixels; the width is typically between 1-6 pixels; the area is typically less than 192 pixels; and the aspect ratio is typically less than 6. The final result of moving vehicle detection is as follows: Figure 5 As shown.

[0092] Step 4: Based on the detection of moving vehicles, perform moving target tracking. Construct the bounding rectangle of target j as its feature representation, denoted as (x... j ,y j ,r j ,c j ),x j With y j The value r represents the position of the top-left corner of the target's bounding rectangle. j With c jThese represent the length and width of the bounding rectangle, respectively. The trajectory is expressed as (ID, KF, AGE, TAC, CIC), where ID is the unique identifier of the trajectory, and the same ID in different frames represents the same target; KF represents the target's state information, recording the target's position and velocity information in the current frame; AGE represents the total number of frames since the trajectory was first detected; TAC represents the number of frames the trajectory has been recorded, used to characterize whether the trajectory is continuous between frames; and CIC represents the frame difference since the trajectory was last recorded, used to characterize the time the target has been lost.

[0093] Step 5: Track moving vehicles based on Kalman filtering and the Hungarian algorithm. Perform trajectory creation, state update, deletion, and delayed retention operations based on the matching results. Under unoccluded conditions, the target can usually be tracked stably; however, under occlusion conditions, the target may be lost. In this state, the trajectory is retained after a delay. When the target reappears after crossing the occlusion area, it is matched with the retained trajectory again. If the match is successful, the trajectory's state is updated.

[0094] Step 6: The tracked trajectory information is expressed as (ID, KF, AGE, TAC, CIC), recording the precise time information of target detection and loss for the same ID. Using AGE, TAC, and CIC, the lost frame of the target can be obtained. Combined with the target position relationship recorded by the state parameter KF, the spatial position of the target in the lost frame can be reconstructed. It should be noted that this method can not only reconstruct targets within occluded areas, but also has a certain ability to handle target loss caused by intermittent target brightness and trajectory anomalies caused by false moving targets. The tracking results for moving vehicles are as follows... Figure 6 As shown.

[0095] Step 7: By detecting and tracking moving vehicles, traffic flow parameters within the target area can be statistically analyzed. The number of vehicles detected within the area, C, includes the number of vehicles detected in the unmasked area and the number of vehicles reconstructed in the unmasked area. The statistical method is...

[0096] C = C1 + C2

[0097] Where C1 represents the traffic flow in the unmasked area, and C2 represents the reconstructed traffic flow in the masked area. C1 is calculated by counting the number of valid IDs in the corresponding frame. Valid means that the target has been continuously detected in the previous n consecutive frames.

[0098]

[0099] C2 is calculated by counting the number of discontinuously lost trajectories within a certain time range between consecutive frames, which is manifested as...

[0100]

[0101] Where [-t,t] represents the time interval between the target entering and leaving the masked area. In this embodiment, the number of vehicles detected in the unmasked area on the highway is C1=57, and the number of vehicles reconstructed in the masked area is C2=3, C=60.

[0102] Step 8: Convert the vehicle count obtained in Step 7 into standard vehicle data to calculate traffic density. The calculation method is as follows: the conversion factors for small cars (3.5m-7m), medium-sized cars (7m-10m), and large cars (over 10m) are 1, 2, and 3, respectively. The traffic density on the i-th frame is expressed using the following formula:

[0103] K i =N i / L

[0104] N i This represents the number of vehicles detected within the study road area on the i-th frame, converted to the number of standard vehicles. L represents the length of the road within the study area. In this embodiment, the number of vehicles N... i =60, the length of the main road of the expressway is L≈1.085km, and the traffic density is K. i =55.

[0105] Step 9: The average speed of the j-th car in the video image can be estimated using the following formula:

[0106] v j =d / Δt

[0107] d represents the distance traveled by vehicle j within the observation time, and Δt represents the observation time interval. Δt can be calculated using the following formula:

[0108] Δt=n*t s

[0109] n represents the number of frames, t s Indicates frame rate.

[0110] During the observation period, the traffic speed in this area can be expressed by the following formula:

[0111]

[0112] m represents the number of vehicles within the studied road area during the observation period. In this case, the traffic speed in the area is 74.45 km / h.

[0113] Step 10: Based on steps (8) and (9), calculate the standard number of vehicles at a specific location or cross-section on the road. The relationship between traffic volume, density, and speed is shown in the following formula:

[0114] Q = v × K

[0115] In this embodiment, the traffic volume Q = 4095, meaning that an average of 4095 vehicles pass through the highway area per hour.

Claims

1. A method for estimating urban traffic flow using satellite video in complex, occluded scenarios, characterized in that, include: Preprocessing of satellite video yields preprocessed video images; Extract the road range from the preprocessed video image; Based on the road range, moving vehicles are detected in the preprocessed video images to obtain moving vehicle detection results; Based on the vehicle detection results and road range, the moving vehicle is tracked to obtain its tracking trajectory; the tracking trajectory includes constructing tracking trajectory information M, specifically: M=（ ID,KF,AGE,TAC,CIC ） in, ID A unique identifier for the trajectory, the same across different frames. ID Representing the same goal; KF It represents the target's state information and records the target's position and velocity information in the current frame; AGE This indicates the total number of frames since the trajectory was first detected in the current frame; TAC This indicates the number of frames in which the trajectory was recorded, and is used to characterize whether the trajectory is continuous between frames; CIC This represents the frame difference between the last recorded frame and the current frame, used to characterize the time when the target was lost. Based on the detection results of moving vehicles and the tracking trajectory of moving vehicles, the number of vehicles in the target area is counted, and the number of vehicles in the target area is obtained. Based on the number of vehicles within the target area and the road range, the regional traffic density is obtained; Calculate the traffic speed within the target area based on the tracking trajectory of the moving vehicle and the regional traffic density; Based on the regional traffic density and traffic speed, calculate the traffic volume at a certain location or cross-section on the road; The method involves using the detection results of moving vehicles and their tracking trajectories to count the number of vehicles within a target area, specifically: Where C represents the number of vehicles detected in the area. This represents the number of vehicles within the unshielded area. The number of vehicles reconstructed within the masked area; [-t,t] represents the time interval between the target entering and leaving the masked area; n The minimum threshold for a target to appear in a frame to determine whether it is a real target or a transient false target; AGE i > CIC i This indicates that the target is discontinuous across time-series frames; TAC i =0 This indicates that the trajectory was not recorded in frame i; TAC i+t ≠0 and TAC i-t ≠0 This indicates that the target exists within t frames before and after the i-th frame.

2. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: The preprocessing of satellite video specifically includes: performing inter-frame registration and target region cropping on the satellite video, wherein the inter-frame registration uses SIFT, ORB or AKAZA registration algorithms.

3. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: The extraction of road extents from the images employs D-LinkNet and RoadNet convolutional neural network algorithms.

4. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: The moving vehicle detection process specifically involves using the ViBe background modeling method, combined with the extracted road area, to perform moving vehicle detection.

5. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: The method for tracking moving vehicles and obtaining their tracking trajectories is as follows: Step S41: Construct the target j The outer rectangle is ,in, x j and y j The position of the top-left corner of the target's bounding rectangle. r j and c j These are the length and width of the circumscribed rectangle, respectively. Step S42: Construct tracking trajectory information M; Step S43: Based on the target j and the tracking trajectory information M, determine whether the target j is in a masked state. If yes, proceed to step S44; otherwise, proceed to step S45. Step S44: Determine whether the target is lost. If so, the trajectory status is retained for a delay. The spatial position of the target in the lost frame is reconstructed based on the tracked trajectory information M. Proceed to step S45. If not, proceed to step S46. Step S45: When the target reappears after crossing the obscured area, determine whether the target matches the delayed trajectory. If yes, proceed to step S46; otherwise, proceed to step S44. Step S46: Update the trajectory status.

6. A satellite video urban traffic flow estimation method for complex, occluded scenarios, as described in claim 4 or 5, characterized in that: The spatial position of the target in the lost frame is reconstructed based on the tracked trajectory information M. Specifically, the lost frame of the target is obtained using AGE, TAC, and CIC, and the target position relationship recorded by the state parameter KF is combined to reconstruct the spatial position of the target in the lost frame.

7. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: The regional traffic density is obtained by combining the number of vehicles in the target area with the road range, specifically as follows: in, The number of vehicles detected in the target area on the i-th frame is converted into the number of standard vehicles; The length of the roads in the target area. K i Traffic density.

8. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: Based on the tracking trajectory of the moving vehicle and the regional traffic density, the traffic speed within the target area is calculated as follows: Where m is the number of vehicles in the target area during the observation time, and n is the number of frames. d represents the frame rate; d represents the distance traveled by vehicle j within the observation time. The observation time interval.

9. The satellite video urban traffic flow estimation method for complex, occluded scenarios as described in claim 1, characterized in that: Based on regional traffic density and traffic speed, the traffic volume at a specific location or cross-section on the road is calculated as follows: Where Q is traffic volume, v is speed, and K is traffic density.

Images