A method for unmanned aerial vehicle parking inspection and photographing

Abstract

The application discloses a kind of unmanned aerial vehicle parking inspection photographing method, comprising the following steps: unmanned aerial vehicle gathers parking lot scene image, calibrates parking lot boundary line;Unmanned aerial vehicle flies along the parking lot boundary line, collects parking lot image and is associated with GPS track, analyzes and extracts parking lot position information;Unmanned aerial vehicle executes autonomous charging decision according to remaining power and parking space occupation state;After completing charging or emergency landing, unmanned aerial vehicle enters the target area of inspection, obtains several subareas and establishes the state record table of each subarea;After completing the task of inspection, unmanned aerial vehicle lands and charges, and according to the inspection result of each subarea, the state record table is updated, and the inspection report is generated and uploaded to the cloud.

Description

Technical Field

[0001] This invention belongs to the field of drone parking lot inspection technology, and particularly relates to a drone parking inspection and photography method. Background Technology

[0002] With the rapid development of drone technology, drones have shown broad application prospects in fields such as intelligent parking management, automated parking assistance, and inspection and monitoring. Traditional parking management systems typically rely on fixed cameras for parking space monitoring, which suffers from blind spots, high deployment costs, and maintenance difficulties. Drones, with their advantages of high mobility, wide field of view, and flexible deployment, offer a new technological approach to solving these problems.

[0003] Currently, there are several technological solutions for applying drones to parking lot management. For example, some solutions use drones for aerial photography and image recognition technology to obtain the occupancy status of parking spaces; others use drones for parking lot inspections, achieving area monitoring through patrol photography. However, existing technologies still have the following shortcomings: First, there is a lack of autonomous boundary recognition capability for drones. Existing solutions typically require manual pre-setting of parking lot boundaries or reliance on pre-stored maps, making it impossible to automatically identify boundaries in unknown parking lot environments. This results in drones being unable to autonomously plan their flight paths and exhibiting poor adaptability.

[0004] Second, the accuracy of parking space information extraction is insufficient and it is disconnected from location information. Existing technologies mostly use a single image recognition method to extract parking spaces without correlating and analyzing the image information with GPS trajectories. This results in inaccurate mapping between parking space location information and real geographic coordinates, making it impossible to achieve accurate positioning and navigation subsequently.

[0005] Third, there is a lack of intelligent autonomous charging decision-making mechanisms. Drones face power constraints during missions, and most existing solutions rely on manual intervention or simple power threshold judgments. They fail to incorporate factors such as parking space occupancy status, charging station location, and area overlap into the decision-making process, resulting in low charging efficiency and even drones losing contact due to depleted power.

[0006] Fourth, the inspection task and parking space status management are disconnected. In existing technologies, parking space management and area inspection are usually performed as independent tasks, failing to form a data loop. The vehicle identification results obtained from the inspection cannot be dynamically synchronized with the parking space status information database, resulting in inspection reports lacking fine-grained zoning information, making it difficult to support subsequent intelligent scheduling and decision optimization.

[0007] Fifth, there is a lack of a sub-region-based status recording mechanism. The existing solution treats the inspection area as a whole without finely dividing the area and recording its status, making it difficult to refine the inspection results down to specific parking spaces or local areas, resulting in low information utilization.

[0008] In summary, existing drone parking and inspection methods suffer from problems such as reliance on preset boundary recognition, poor parking space positioning accuracy, insufficient intelligence in charging decisions, separation of inspection and parking space management, and lack of refined area division. There is an urgent need to propose a technical solution that can achieve autonomous boundary recognition by drones, high-precision parking space positioning, intelligent charging decisions, collaborative inspection and parking space management, and refined area division. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides a drone parking inspection and photography method. Through core technologies such as image color difference analysis, GPS trajectory association, autonomous charging decision-making, area division, and status recording, it solves the technical problems in the prior art, such as boundary recognition relying on presets, poor parking space positioning accuracy, insufficient intelligence in charging decision-making, separation of inspection and parking space management, and lack of fine-grained area division. It has significant technological progress and practical value.

[0010] The technical solution of this invention is implemented as follows: A method for taking photos of vehicles parked by drone includes the following steps: S1. The drone collects images of the parking lot scene through a camera, and identifies and marks the parking lot boundary line based on image color difference analysis; S2. The drone flies along the marked parking lot boundary line, collects parking lot images and associates them with GPS trajectory, and extracts the location and outline information of parking spaces through analysis and processing; S3. The drone makes autonomous charging decisions based on the remaining battery power and the parking space occupancy status, including: establishing a parking space information database, confirming the actual status of parking spaces through inspection and photography, and selecting charging points based on overlapping areas. S4. After the drone completes charging or makes an emergency landing, it enters the target area for inspection and obtains several sub-areas through cruise photography, image stitching, vehicle recognition and area division, and establishes a status record table for each sub-area. S5. After the inspection mission is completed, the drone lands to charge and updates the status record table based on the inspection results of each sub-area, generating an inspection report and uploading it to the cloud.

[0011] Furthermore, S1 specifically includes: The drone takes pictures of the parking lot scene using a camera and extracts the difference between the maximum and minimum values ​​of the red, green, and blue chromaticity values ​​in the image. : ; in, These represent the chromaticity values ​​of each pixel in the red, green, and blue channels, respectively; for Threshold filtering is performed, assuming the total number of pixels in the image is... The area threshold is If the following conditions are met: ; in, Represents pixel coordinates. The preset color difference threshold, For indicator functions; If the current image is determined to contain the parking lot boundary, the image is filtered using the difference of Gaussian operator to enhance the boundary features; otherwise, the image is reacquired. Edge recognition is performed on the filtered image, and the detected contours are filtered based on the position of the parking lot boundary lines. If only two boundary lines remain after filtering, the area is determined to be a parking lot area, and the parking lot boundary lines are marked.

[0012] Furthermore, in step S2, the drone flies along the boundary of the parking lot and takes pictures of the parking lot using a camera. The obtained images are numbered sequentially and recorded on a GPS track. Then, a neural network model is used to analyze the parking space information, specifically including: S2.1 Extract and process the GPS trajectory to filter out the boundary areas; S2.2 Based on the selected boundary regions, the image is divided into several segmentation blocks according to the segmentation ratio. The azimuth sequence of each segmentation block on the GPS trajectory is calculated. By clustering analysis and relation value calculation of the azimuth sequence, corner candidate regions and edge candidate regions in the segmentation block are identified. The optimal parking space is selected according to the sum of the relation values ​​of the corner candidate regions and the edge candidate regions. At the same time, a unique identification code is generated for each parking space and its GPS coordinates are recorded.

[0013] Furthermore, in S2.1, the specific steps for filtering out the boundary region are as follows: S2.11. Based on the calibrated parking lot boundary lines, extract the corresponding boundary lines from the image and calculate the adjacent GPS points on the GPS trajectory. and slope : ; in, and The first The and the first The planar coordinates of the GPS points; Based on the slope Determine the tangent direction of the GPS track at that point, and mark the image area on the GPS track whose tangent direction is consistent with the direction of the calibrated parking lot boundary line as a boundary. Traverse all GPS points to obtain all boundary areas. S2.12 If the obtained boundary line contains only one boundary line segment, then directly use the boundary line segment as the parking lot boundary area and output it to S2.2; if multiple boundary line segments are obtained, then each boundary line segment is used as a boundary candidate set, and each boundary line segment in the set is selected in turn as the processing object and output to S2.2 for subsequent processing.

[0014] Furthermore, in S2.2, the specific steps for identifying parking spaces based on boundary areas are as follows: S2.21 Receive the boundary region or boundary candidate set output by S2.1, divide the image into several blocks according to the segmentation ratio, and calculate the azimuth angle of each segmented block to its adjacent points on the GPS track. And save it to get the sequence. Cluster these sequences sequentially to obtain a set of parking spaces. If there are overlapping parking spaces, the overlapping areas will be removed from the original set. S2.22. For each segment within a parking space, calculate its average azimuth angle. : ; in, This indicates the number of azimuth angles within the segmented block. Indicates the first In the segmented block, the first One azimuth angle; S2.23. Compare the azimuth sequence of each segment with the average azimuth sequence, and calculate the average distance between the two sequences. : ; The average distance is used as the relation value of the corresponding segmentation block. ,Right now ; S2.24. Sum the relationship values ​​between the line segments on the edge of each segmented region and the parking space segmentation blocks to obtain the sum of the relationship values ​​between each parking space segmentation block and its corresponding block. ; S2.25, Set the relation value Greater than the preset corner threshold The segmented blocks are identified as corner candidate areas for parking spaces, and the relation values ​​are... Less than or equal to The segmented blocks are determined as candidate edge regions for parking spaces; For each parking space in the polygonal parking space candidate set, sum the relationship values ​​between each edge and the corner candidate regions of the parking space, and take the parking space candidate region with the highest relationship value as the parking space output. If the sum of the relationship values ​​of all candidate regions is... All are below the preset screening threshold If so, then the parking lot will be abandoned.

[0015] Furthermore, S3 specifically includes: S3.1 Based on the parking space information obtained in S2, generate a unique identification code for each parking space, and record its GPS coordinates, charging pile location and occupancy status to establish a parking space status information database. S3.2, Preset charging decision threshold The drone monitors the remaining battery power in real time. ; S3.3, if The drone then takes pictures and samples according to the planned inspection path, uses a telephoto lens to obtain images of the parking space area, classifies and identifies the actual occupancy status of the parking spaces through image classification, and selects the best unoccupied parking spaces based on the S3.1 information database. After landing, it identifies the corresponding identification code, pairs the charging pile with the parking space, and starts charging. like The drone will then fly directly to the pre-planned emergency landing point. S3.4 After charging or making an emergency landing, the drone updates the occupancy status in the parking space status information database and uses the current landing point as the takeoff point for subsequent inspection tasks.

[0016] Furthermore, S4 specifically includes: S4.1, Preset minimum power requirement for inspection After the drone completes S3, it reads the current remaining battery power. ; S4.2, if Then take off from the current landing point and set the flight inspection path according to the inspection target area; S4.3 The drone cruises and takes pictures according to the planned path, and uses a pixel-based image stitching algorithm to stitch together each captured area map to generate a complete panoramic image of the road surface. S4.4 Use a neural network model trained from the dataset to identify vehicles in panoramic images and extract obstacle points; S4.5 Extract the identified obstacle points, generate a grid using Voronoi Diagrams, label the area occupied by the vehicle, and divide the overall identification area into several sub-regions using the edge information-based region division method, and establish an independent region identifier and status record table for each sub-region. like If the battery level is low, it will continue to charge at the current charging point until the battery reaches its maximum capacity. Then, execute steps S4.2 through S4.5.

[0017] Furthermore, S4.3 specifically includes: S4.31. For each area map to be stitched, sort the data of points in the GPS track that contain that area map in ascending order, and calculate the slope of each data point. Mark the coordinates of each point as The distance between two adjacent points is calculated. : ; in, and These represent the numbers on the GPS track. The point and the first The coordinates of a point in the preset projection direction; S4.32. For each point, calculate its distance to the previous point and the average distance to all its neighboring points. proportion : ; ; in, This represents the total number of points on the GPS track. This represents the average distance between all adjacent points. Indicates the first The ratio of segment distance to average distance; like ,in If the preset segmentation weight threshold is not met, the current point will be used as the dividing point; otherwise, the current point will be used as an undivided point. S4.33. Filter the obtained boundary points and take the point with the farthest distance as the first boundary point. If the distance from the boundary point to the farthest point is less than the distance between the nearest points between the boundary points, delete the boundary points between adjacent boundary points. Then calculate the distance between each boundary point and the farthest point, and take the point with the farthest distance as the new boundary point. Repeat this process until all boundary points meet the requirements. S4.34. Based on the selected boundary points, the GPS track is divided into several continuous intervals. Each interval is composed of two adjacent boundary points and all the unbounded points between them. Each boundary point is connected to its nearest previous point, and two adjacent boundary points are connected to form several rectangular areas. Each rectangular area corresponds to a splicing sub-area. S4.35. Use a feature-based block method to find the non-deformable regions within each stitching sub-region, and stitch the images of each stitching sub-region at their corresponding original positions to generate a panoramic image.

[0018] Furthermore, S4.5 specifically includes: S4.51. For obstacle areas, use Voronoi Diagrams to obtain the dividing lines of the corresponding areas, use edge detection to extract the boundary lines of the obstacle areas, and intersect the dividing lines in the areas with the boundary lines to obtain the coordinates of the intersection points. S4.52. For each intersection point, starting from that point, traverse the boundary lines adjacent to it. Let the Euclidean distance between the two points be... ,like ,in If the preset boundary merging distance threshold is not met, then delete one of the boundary lines until all intersection points have been traversed, and use the remaining boundary lines as edge candidate regions; S4.53, for each edge candidate region, treat it as a polygon in the Voronoi Diagram, and let the i-th edge candidate region be... The side length is The Euclidean distance from the midpoint of one side to the midpoint of the adjacent side is . Then the proportion corresponding to this side length for: ; in, Indicates the first The length of the strip, Let represent the Euclidean distance between the midpoint of an edge and the midpoint of its adjacent edge; normalize the sum of all weights to obtain the normalized weight. : ; in, This represents the sum of the weights corresponding to all edges; right Sort the edges in ascending order. For the edges corresponding to the lengths of the first K edges, find the midpoint and connect the midpoint to the center point of the candidate edge region. Pair the closed region obtained by the connection with the parking space, where K is the preset number of candidate edges. S4.54. Based on the pairing results, the closed areas that are successfully paired with parking spaces are marked as vehicle-occupied areas, and the closed areas that are not paired with any parking spaces are marked as other obstacle areas. Based on the spatial relationship of each closed area, the overall identification area is divided into several sub-areas. Each sub-area contains at least one parking space and its surrounding environment information. An independent area identification and status record table is established for each sub-area.

[0019] Furthermore, S5 specifically includes: S5.1 For each parking space, record its corresponding identification code, location, whether it is occupied, and charging pile information, and plan the parking location according to the historical records in the next planning. S5.2 For the inspection area, based on the sub-areas divided in S4 and their corresponding status record tables, the information inspected by the drone is compared with the status record tables of each sub-area to obtain the inspection results of each sub-area, generate an inspection report containing the parking space occupancy status and obstacle distribution of each sub-area, and upload it to the cloud. S5.3 At the end of the inspection, the drone will land in the designated landing area to charge. If the drone does not reach the landing area within the designated time, it will be determined that the drone is damaged or has insufficient power to reach the landing point. When the drone is low on power, it will land in a nearby charging area and send an alarm message to the staff.

[0020] Compared with the prior art, the present invention achieves the following beneficial effects: First, it achieves autonomous identification and calibration of parking lot boundaries; by identifying parking lot boundaries through image color difference analysis, without the need for manual preset or reliance on pre-stored maps, the drone can autonomously complete boundary identification in unknown parking lot environments, significantly improving the system's environmental adaptability and deployment flexibility.

[0021] Secondly, it achieves accurate extraction and spatial positioning of parking space information; by associating and recording images collected by the drone flying along the boundary with GPS trajectory, the location and contour information of the parking space are extracted through analysis and processing, which solves the problem of poor positioning accuracy in traditional single image recognition, and provides an accurate spatial reference for subsequent charging decisions and inspection tasks.

[0022] Third, an intelligent autonomous charging decision-making mechanism has been established. The mechanism makes autonomous charging decisions by combining the remaining power and the parking space occupancy status. Through operations such as establishing a parking space information database, inspecting and taking photos to confirm the actual status, and judging overlapping areas, the mechanism realizes the intelligent selection of charging points, effectively avoiding the loss of drone connection due to power depletion, and improving the safety and reliability of mission execution.

[0023] Fourth, it achieves deep integration of inspection tasks and area management; after completing charging, it enters the target inspection area, and through cruise photography, image stitching, vehicle recognition and area division, it establishes an independent status record table for each sub-area, so that the inspection results can be refined to specific areas, laying a data foundation for subsequent status updates and report generation.

[0024] Fifth, a refined status recording mechanism based on sub-regions has been established. After the inspection task is completed, the status record table is updated according to the inspection results of each sub-region, and an inspection report containing the parking space occupancy status and obstacle distribution of each sub-region is generated, making the management information more refined and structured, which facilitates subsequent intelligent scheduling and decision analysis.

[0025] Sixth, a complete technical closed loop of "identification → positioning → charging → inspection → reporting" has been formed, with a clear logical progression between each step: boundary identification provides a basis for flight navigation, parking space extraction provides a location benchmark for charging decisions, charging decisions provide power assurance for inspection tasks, area division provides fine-grained status recording, and status updates provide historical data support for subsequent tasks. The entire technical solution enables autonomous, intelligent, and precise collaborative operation of drones in parking lot scenarios. Attached Figure Description

[0026] Figure 1 This is a flowchart of a drone parking inspection and photography method provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] Example like Figure 1 A method for taking photos of unmanned aerial vehicles (UAVs) during parking inspections includes the following steps: S1. The drone collects images of the parking lot scene through a camera, and identifies and marks the parking lot boundary line based on image color difference analysis; Specifically, S1 includes: The drone takes pictures of the parking lot scene using a camera and extracts the difference between the maximum and minimum values ​​of the red, green, and blue chromaticity values ​​in the image. : ; in, These represent the chromaticity values ​​of each pixel in the red, green, and blue channels, respectively; for Threshold filtering is performed, assuming the total number of pixels in the image is... The area threshold is If the following conditions are met: ; in, Represents pixel coordinates. The preset color difference threshold, For indicator functions; If the current image is determined to contain the parking lot boundary, the image is filtered using the difference of Gaussian operator to enhance the boundary features; otherwise, the image is reacquired. Edge recognition is performed on the filtered image, and the detected contours are filtered based on the position of the parking lot boundary lines. If only two boundary lines remain after filtering, the area is determined to be a parking lot area, and the parking lot boundary lines are marked.

[0029] In this embodiment, the drone is equipped with a high-definition camera and hovers at a preset height of 30 meters to take pictures of the target parking area. The acquired images are RGB color images with a resolution of 1920×1080 pixels.

[0030] The drone first extracts the red, green, and blue chromaticity values ​​of each pixel in the image and calculates the color difference. Set the color difference threshold. Within the 0-255 chromaticity range, the area ratio threshold If the image satisfies If the number of pixels in a certain area exceeds 30% of the total number of pixels, then the image is determined to contain the boundary of a parking lot.

[0031] The image is then filtered using the difference of Gaussian operator to enhance boundary features. The boundary response is extracted by the difference between the two Gaussian filtering results.

[0032] Edge detection and contour filtering are performed on the filtered image. If only two approximately parallel boundary lines remain, they are identified as the parking lot boundaries, and their position and direction information are recorded. In this embodiment, the two identified boundary lines are the left and right boundaries of the parking lot, with a direction of 15 degrees east of north.

[0033] S2. The drone flies along the marked parking lot boundary line, collects parking lot images and associates them with GPS trajectory, and extracts the location and outline information of parking spaces through analysis and processing; In step S2, the drone flies along the boundary of the parking lot and takes pictures of the parking lot using a camera. The obtained images are numbered sequentially and recorded on a GPS track. Then, a neural network model is used to analyze the parking space information, specifically including: S2.1 Extract and process the GPS trajectory to filter out the boundary areas; S2.11. Based on the calibrated parking lot boundary lines, extract the corresponding boundary lines from the image and calculate the adjacent GPS points on the GPS trajectory. and slope : ; in, and The first The and the first The planar coordinates of the GPS points; Based on the slope Determine the tangent direction of the GPS track at that point, and mark the image area on the GPS track whose tangent direction is consistent with the direction of the calibrated parking lot boundary line as a boundary. Traverse all GPS points to obtain all boundary areas. S2.12 If the obtained boundary line contains only one boundary line segment, then directly use the boundary line segment as the parking lot boundary area and output it to S2.2; if multiple boundary line segments are obtained, then each boundary line segment is used as a boundary candidate set, and each boundary line segment in the set is selected in turn as the processing object and output to S2.2 for subsequent processing.

[0034] S2.2 Based on the selected boundary regions, the image is divided into several segmentation blocks according to the segmentation ratio. The azimuth sequence of each segmentation block on the GPS trajectory is calculated. By clustering analysis and relation value calculation of the azimuth sequence, corner candidate regions and edge candidate regions in the segmentation block are identified. The optimal parking space is selected according to the sum of the relation values ​​of the corner candidate regions and the edge candidate regions. At the same time, a unique identification code is generated for each parking space and its GPS coordinates are recorded.

[0035] S2.21 Receive the boundary region or boundary candidate set output by S2.1, divide the image into several blocks according to the segmentation ratio, and calculate the azimuth angle of each segmented block to its adjacent points on the GPS track. And save it to get the sequence. Cluster these sequences sequentially to obtain a set of parking spaces. If there are overlapping parking spaces, the overlapping areas will be removed from the original set. S2.22. For each segment within a parking space, calculate its average azimuth angle. : ; in, This indicates the number of azimuth angles within the segmented block. Indicates the first In the segmented block, the first One azimuth angle; S2.23. Compare the azimuth sequence of each segment with the average azimuth sequence, and calculate the average distance between the two sequences. : ; The average distance is used as the relation value of the corresponding segmentation block. ,Right now ; S2.24. Sum the relationship values ​​between the line segments on the edge of each segmented region and the parking space segmentation blocks to obtain the sum of the relationship values ​​between each parking space segmentation block and its corresponding block. ; S2.25, Set the relation value Greater than the preset corner threshold The segmented blocks are identified as corner candidate areas for parking spaces, and the relation values ​​are... Less than or equal to The segmented blocks are determined as candidate edge regions for parking spaces; For each parking space in the polygonal parking space candidate set, sum the relationship values ​​between each edge and the corner candidate regions of the parking space, and take the parking space candidate region with the highest relationship value as the parking space output. If the sum of the relationship values ​​of all candidate regions is... All are below the preset screening threshold If so, then the parking lot will be abandoned.

[0036] In this embodiment, the drone flies along the marked boundary line of the parking lot, maintaining a flight altitude of 20 meters and a flight speed of 2 meters per second. The drone takes one image every 1 meter of flight, obtaining a total of 200 images. Each image is numbered according to the order in which it was taken and associated with the GPS coordinates at the corresponding time.

[0037] First, boundary region filtering is performed. Based on the direction of the boundary line of S1, the slope of adjacent points on the GPS trajectory is calculated. The tangent direction is determined. Image regions on the GPS trajectory aligned with the boundary line direction are marked as boundaries. In this embodiment, two boundary line segments are marked and output as a candidate set of boundaries.

[0038] Then, parking space identification is performed. After receiving the candidate boundary set, each boundary segment is processed sequentially. Taking the first boundary segment as an example, the image is divided into 100 segments at a ratio of 1:10. The azimuth angles of adjacent points on the GPS track for each segment are calculated. , to obtain the sequence .

[0039] Clustering is performed on the azimuth sequence to obtain a set of possible parking spaces. In this embodiment, a total of 15 candidate parking spaces were identified, of which two overlapped areas. After removing these two overlapped areas, 13 candidates remained.

[0040] Calculate the average azimuth and related values Segments with a relation value greater than 15° are classified as corner candidate regions, while those with a relation value less than or equal to 15° are classified as edge candidate regions.

[0041] For each parking space in the polygonal parking space candidate set, the relationship values ​​between each edge and the corner candidate region are summed. The candidate with the highest sum of relationship values ​​is taken as the recognition result under that boundary segment, and its confidence level is recorded as 0.92. The second boundary segment is processed in the same way, resulting in a confidence level of 0.85. After comparison, the recognition result with a confidence level of 0.92 is selected as the final output. A unique identification code (such as P001 to P013) is generated for each parking space, and its GPS coordinates are recorded.

[0042] S3. The drone makes autonomous charging decisions based on the remaining battery power and the parking space occupancy status, including: establishing a parking space information database, confirming the actual status of parking spaces through inspection and photography, and selecting charging points based on overlapping areas. S3.1 Based on the parking space information obtained in S2, generate a unique identification code for each parking space, and record its GPS coordinates, charging pile location and occupancy status to establish a parking space status information database. S3.2, Preset charging decision threshold The drone monitors the remaining battery power in real time. ; S3.3, if The drone then takes pictures and samples according to the planned inspection path, uses a telephoto lens to obtain images of the parking space area, classifies and identifies the actual occupancy status of the parking spaces through image classification, and selects the best unoccupied parking spaces based on the S3.1 information database. After landing, it identifies the corresponding identification code, pairs the charging pile with the parking space, and starts charging. like The drone will then fly directly to the pre-planned emergency landing point. S3.4 After charging or making an emergency landing, the drone updates the occupancy status in the parking space status information database and uses the current landing point as the takeoff point for subsequent inspection tasks.

[0043] In this embodiment, a parking space status information database is first established. The drone inputs the information of 13 parking spaces output by S2 into the database. For each parking space, the identification code, GPS coordinates, charging pile location (as determined on-site, each parking space is equipped with a wireless charging pile), and initial occupancy status (all are unoccupied) are recorded.

[0044] Set a charging decision threshold The drone monitors the remaining battery power in real time. In this embodiment, the drone currently has 65% remaining battery power, which meets the requirements. .

[0045] The drone follows a planned inspection path to take photos and samples, using a telephoto lens (50mm focal length) to acquire images of the parking space area. The actual occupancy status of the parking spaces is determined through image classification and recognition. In this embodiment, parking spaces P003, P007, and P011 are identified as occupied, and their corresponding status is updated in the database.

[0046] The drone selects the best unoccupied parking space (the closest one with an available charging station) from the database, lands at parking space P005, identifies its identification code, pairs the charging station with the parking space, and begins charging.

[0047] After charging is complete, the drone updates the parking space status information database to show that P005 is occupied (occupied during charging) and records the current landing point as the takeoff point for subsequent inspection missions.

[0048] S4. After the drone completes charging or makes an emergency landing, it enters the target area for inspection and obtains several sub-areas through cruise photography, image stitching, vehicle recognition and area division, and establishes a status record table for each sub-area. S4.1, Preset minimum power requirement for inspection After the drone completes S3, it reads the current remaining battery power. ; S4.2, if Then take off from the current landing point and set the flight inspection path according to the inspection target area; S4.3 The drone cruises and takes pictures according to the planned path, and uses a pixel-based image stitching algorithm to stitch together each captured area map to generate a complete panoramic image of the road surface. S4.31. For each area map to be stitched, sort the data of points in the GPS track that contain that area map in ascending order, and calculate the slope of each data point. Mark the coordinates of each point as The distance between two adjacent points is calculated. : ; in, and These represent the numbers on the GPS track. The point and the first The coordinates of a point in the preset projection direction; S4.32. For each point, calculate its distance to the previous point and the average distance to all its neighboring points. proportion : ; ; in, This represents the total number of points on the GPS track. This represents the average distance between all adjacent points. Indicates the first The ratio of segment distance to average distance; like ,in If the preset segmentation weight threshold is not met, the current point will be used as the dividing point; otherwise, the current point will be used as an undivided point. S4.33. Filter the obtained boundary points and take the point with the farthest distance as the first boundary point. If the distance from the boundary point to the farthest point is less than the distance between the nearest points between the boundary points, delete the boundary points between adjacent boundary points. Then calculate the distance between each boundary point and the farthest point, and take the point with the farthest distance as the new boundary point. Repeat this process until all boundary points meet the requirements. S4.34. Based on the selected boundary points, the GPS track is divided into several continuous intervals. Each interval is composed of two adjacent boundary points and all the unbounded points between them. Each boundary point is connected to its nearest previous point, and two adjacent boundary points are connected to form several rectangular areas. Each rectangular area corresponds to a splicing sub-area. S4.35. Use a feature-based block method to find the non-deformable regions within each stitching sub-region, and stitch the images of each stitching sub-region at their corresponding original positions to generate a panoramic image.

[0049] S4.4 Use a neural network model trained from the dataset to identify vehicles in panoramic images and extract obstacle points; S4.5 Extract the identified obstacle points, generate a grid using Voronoi Diagrams, label the area occupied by the vehicle, and divide the overall identification area into several sub-regions using the edge information-based region division method, and establish an independent region identifier and status record table for each sub-region. like If the battery level is low, it will continue to charge at the current charging point until the battery reaches its maximum capacity. Then, execute steps S4.2 through S4.5.

[0050] S4.51. For obstacle areas, use Voronoi Diagrams to obtain the dividing lines of the corresponding areas, use edge detection to extract the boundary lines of the obstacle areas, and intersect the dividing lines in the areas with the boundary lines to obtain the coordinates of the intersection points. S4.52. For each intersection point, starting from that point, traverse the boundary lines adjacent to it. Let the Euclidean distance between the two points be... ,like ,in If the preset boundary merging distance threshold is not met, then delete one of the boundary lines until all intersection points have been traversed, and use the remaining boundary lines as edge candidate regions; S4.53, for each edge candidate region, treat it as a polygon in the Voronoi Diagram, and let the i-th edge candidate region be... The side length is The Euclidean distance from the midpoint of one side to the midpoint of the adjacent side is . Then the proportion corresponding to this side length for: ; in, Indicates the first The length of the strip, Let represent the Euclidean distance between the midpoint of an edge and the midpoint of its adjacent edge; normalize the sum of all weights to obtain the normalized weight. : ; in, This represents the sum of the weights corresponding to all edges; right Sort the edges in ascending order. For the edges corresponding to the lengths of the first K edges, find the midpoint and connect the midpoint to the center point of the candidate edge region. Pair the closed region obtained by the connection with the parking space, where K is the preset number of candidate edges. S4.54. Based on the pairing results, the closed areas that are successfully paired with parking spaces are marked as vehicle-occupied areas, and the closed areas that are not paired with any parking spaces are marked as other obstacle areas. Based on the spatial relationship of each closed area, the overall identification area is divided into several sub-areas. Each sub-area contains at least one parking space and its surrounding environment information. An independent area identification and status record table is established for each sub-area.

[0051] In this embodiment, a preset The drone currently has 85% battery remaining, which meets the requirements. Take off from the current landing point (P005) and enter the inspection target area.

[0052] The drone cruised and took photos along a planned path, flying at an altitude of 30 meters, a speed of 3 meters per second, and a shooting interval of 2 seconds, collecting a total of 150 images of the area. The collected images were then stitched together to generate a panoramic image.

[0053] During image stitching, data of points in the area map contained in the GPS track are sorted in ascending order, and the distance between adjacent points is calculated. and average distance Calculate the proportion Set the segmented weight threshold. ,like If a point is marked as a boundary point, it is considered a boundary point; otherwise, it is considered an unbounded point. In this embodiment, a total of 8 boundary points are marked.

[0054] After filtering the boundary points and removing redundant boundary points, 5 valid boundary points remain.

[0055] Based on the selected boundary points, the GPS track is divided into 6 consecutive intervals. Each interval consists of adjacent boundary points and the unbounded points between them. Using the two endpoints of each interval as the two diagonal vertices of a rectangular region, the image regions corresponding to all unbounded points within the interval are incorporated into the rectangular region, forming 6 rectangular regions. Each rectangular region corresponds to a stitched sub-region.

[0056] The non-deformable regions within each stitching sub-region are found using a feature-based segmentation method, and then stitched together at the corresponding original positions to generate a panoramic image.

[0057] A YOLOv5 neural network model trained from a dataset of 2000 parking lot images was used to identify vehicles and extract obstacle points from panoramic images. In this embodiment, a total of 28 obstacle points were identified, corresponding to 24 cars and 4 other obstacles.

[0058] The obstacle area is divided into sub-regions. A mesh is generated using Voronoi Diagrams, and the area occupied by the vehicle is labeled. A boundary merging distance threshold is set. Meters are used to merge adjacent boundary lines to obtain candidate edge regions. The weight of each edge is then calculated. After normalization, we get ,Pick For the top 3 edges, find their midpoints and connect them to the center point of the region. Then, pair the closed regions obtained by connecting the lines with the parking spaces.

[0059] Pairing results: 24 closed areas were successfully paired and marked as vehicle-occupied areas; 4 closed areas were not paired and marked as other obstacle areas. Based on the spatial relationship of the closed areas, the overall identification area was divided into 6 sub-areas, each containing 2-3 parking spaces and their surrounding environment information. An independent area identifier (e.g., Z01 to Z06) and status record table were established for each sub-area.

[0060] S5. After the inspection mission is completed, the drone lands to charge and updates the status record table based on the inspection results of each sub-area, generating an inspection report and uploading it to the cloud.

[0061] S5.1 For each parking space, record its corresponding identification code, location, whether it is occupied, and charging pile information, and plan the parking location according to the historical records in the next planning. S5.2 For the inspection area, based on the sub-areas divided in S4 and their corresponding status record tables, the information inspected by the drone is compared with the status record tables of each sub-area to obtain the inspection results of each sub-area, generate an inspection report containing the parking space occupancy status and obstacle distribution of each sub-area, and upload it to the cloud. S5.3 At the end of the inspection, the drone will land in the designated landing area to charge. If the drone does not reach the landing area within the designated time, it will be determined that the drone is damaged or has insufficient power to reach the landing point. When the drone is low on power, it will land in a nearby charging area and send an alarm message to the staff.

[0062] In this embodiment, after the inspection task is completed, the drone lands in the designated landing area to charge. Based on the established parking space status information database, the identification code, location, occupancy status, and charging pile information of each parking space are recorded. In this embodiment, the occupancy status of parking spaces P003, P007, P011, and P005 has been updated, while the remaining parking spaces remain vacant.

[0063] Based on the six sub-regions and their corresponding status logs, the information detected by the drone is compared with the status logs of each sub-region. For example, sub-region Z01 contains three parking spaces: P001, P002, and P003. The inspection results show that P003 is occupied, consistent with the status log; P001 and P002 are vacant, and their status logs remain unchanged.

[0064] The system generates an inspection report, which includes: parking space occupancy status (number of vacant / occupied spaces) in each sub-area, obstacle distribution (vehicle location and other obstacle location), and an overall area status assessment. The inspection report is uploaded to the cloud in JSON format for real-time viewing and historical querying by management personnel.

[0065] If a drone fails to reach the designated landing area within a preset time (e.g., 30 minutes) while charging, it is deemed damaged or has insufficient battery power to reach the landing point. In this case, the drone will automatically land in a nearby charging area (e.g., an emergency charging station) and send an alarm message (including GPS coordinates, remaining battery power, and fault code) to the staff.

[0066] This invention provides a method for taking pictures of parking inspections using drones. In S1, image color difference analysis is performed to extract the color difference values ​​of red, green and blue and perform threshold filtering. Combined with the Gaussian difference operator to enhance boundary features, it can automatically identify and mark the parking lot boundary line without the need for manual preset or reliance on pre-stored maps. This significantly improves the adaptability and autonomy of drones in unknown parking lot environments and realizes the autonomous identification and marking of parking lot boundaries.

[0067] In step S2, the acquired images are associated with and recorded with GPS trajectories. By using azimuth sequence analysis, clustering operations, and relation value calculation, the corner candidate regions and edge candidate regions of parking spaces can be accurately identified, and the location and contour information of the parking spaces can be output. At the same time, a unique identification code is generated for each parking space and GPS coordinates are recorded, providing an accurate location benchmark for subsequent charging decisions and inspection tasks, thus achieving high-precision parking space information extraction and GPS trajectory association.

[0068] By establishing a parking space information database containing GPS coordinates, charging station locations, and occupancy status through S3, and introducing a charging decision threshold, the system intelligently selects charging points based on the comparison between the real-time remaining battery power and the threshold, combined with the parking space occupancy status and area overlap judgment: when the battery is sufficient, it prioritizes landing in a parking space matching the database for pairing and charging; when the battery is insufficient, it flies directly to an emergency landing point. This mechanism effectively avoids drone disconnection due to battery depletion, improves the safety and reliability of mission execution, and establishes an intelligent autonomous charging decision-making mechanism.

[0069] In step S4, patrol photography, image stitching, vehicle recognition, and area division are performed within the target inspection area. An independent status record table is established for each sub-area, allowing the inspection results to be refined down to specific sub-areas and parking spaces. Step S5 further compares and updates the inspection results with the status record table, forming a data closed loop of "charging decision → inspection execution → status update." This achieves dynamic maintenance and continuous optimization of parking space status information, as well as deep integration of inspection tasks and parking space management.

[0070] Voronoi Diagrams generates a grid, merges boundaries, calculates edge weights, and pairs closed areas with parking spaces, dividing the overall identification area into several sub-regions. Each sub-region contains at least one parking space and its surrounding environment information, and establishes independent area identifiers and status record tables. Based on the status record tables of each sub-region, S5 generates inspection reports containing information on parking space occupancy and obstacle distribution in the sub-regions, making the inspection results more refined and structured, facilitating subsequent intelligent scheduling and decision analysis, and establishing a refined status record mechanism based on sub-regions.

[0071] The various steps in this invention form a clear progressive relationship, achieving logical coordination and data closure among multiple steps: S1 outputs boundary lines for S2 navigation; S2 outputs parking space information for S3 to establish an information database; S3 outputs charging point locations for S4 as takeoff points; S4 outputs sub-area status record tables for S5 to generate inspection reports; and the status update results of S5 provide historical data support for the next task. The entire technical solution constitutes a complete closed-loop system of "boundary recognition → parking space positioning → charging decision → inspection division → status update".

[0072] This invention provides a method for drone-based parking inspection and photography, applicable to intelligent management systems for various parking lots. It enables drones to autonomously identify boundaries, accurately locate parking spaces, make intelligent charging decisions, and perform refined inspection area division and status recording management. This method significantly reduces labor costs in parking lot management, improves parking space utilization and management efficiency, and has good industrial practical value and promising prospects for widespread application.

[0073] Based on the disclosure and teachings of the foregoing specification, those skilled in the art can make changes and modifications to the above embodiments. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and some modifications and changes to the present invention should also fall within the protection scope of the claims of the present invention. Furthermore, although some specific terms are used in this specification, these terms are only for convenience of explanation and do not constitute any limitation on the present invention.

Claims

1. A method for taking photos during unmanned aerial vehicle (UAV) parking inspections, characterized in that, Includes the following steps: S1. The drone collects images of the parking lot scene through a camera, and identifies and marks the parking lot boundary line based on image color difference analysis; S2. The drone flies along the marked parking lot boundary line, collects parking lot images and associates them with GPS trajectory, and extracts the location and outline information of parking spaces through analysis and processing; S3. The drone makes autonomous charging decisions based on the remaining battery power and the parking space occupancy status, including: establishing a parking space information database, confirming the actual status of parking spaces through inspection and photography, and selecting charging points based on overlapping areas. S4. After the drone completes charging or makes an emergency landing, it enters the target area for inspection and obtains several sub-areas through cruise photography, image stitching, vehicle recognition and area division, and establishes a status record table for each sub-area. S5. After the inspection mission is completed, the drone lands to charge and updates the status record table based on the inspection results of each sub-area, generating an inspection report and uploading it to the cloud.

2. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 1, characterized in that, Specifically, S1 includes: The drone takes pictures of the parking lot scene using a camera and extracts the difference between the maximum and minimum values ​​of the red, green, and blue chromaticity values ​​in the image. : ； in, These represent the chromaticity values ​​of each pixel in the red, green, and blue channels, respectively; for Threshold filtering is performed, assuming the total number of pixels in the image is... The area threshold is If the following conditions are met: ； in, Represents pixel coordinates. The preset color difference threshold, For indicator functions; If the current image is determined to contain the parking lot boundary, the image is filtered using the difference of Gaussian operator to enhance the boundary features; otherwise, the image is reacquired. Edge recognition is performed on the filtered image, and the detected contours are filtered based on the position of the parking lot boundary lines. If only two boundary lines remain after filtering, the area is determined to be a parking lot area, and the parking lot boundary lines are marked.

3. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 1, characterized in that, In step S2, the drone flies along the boundary of the parking lot and takes pictures of the parking lot using a camera. The obtained images are numbered sequentially and recorded on a GPS track. Then, a neural network model is used to analyze the parking space information, specifically including: S2.1 Extract and process the GPS trajectory to filter out the boundary areas; S2.2 Based on the selected boundary regions, the image is divided into several segmentation blocks according to the segmentation ratio. The azimuth sequence of each segmentation block on the GPS trajectory is calculated. By clustering analysis and relation value calculation of the azimuth sequence, corner candidate regions and edge candidate regions in the segmentation block are identified. The optimal parking space is selected according to the sum of the relation values ​​of the corner candidate regions and the edge candidate regions. At the same time, a unique identification code is generated for each parking space and its GPS coordinates are recorded.

4. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 3, characterized in that, In step S2.1, the specific steps for filtering out the boundary region are as follows: S2.

11. Based on the calibrated parking lot boundary lines, extract the corresponding boundary lines from the image and calculate the adjacent GPS points on the GPS trajectory. and slope : ； in, and The first The and the first The planar coordinates of the GPS points; Based on the slope Determine the tangent direction of the GPS track at that point, and mark the image area on the GPS track whose tangent direction is consistent with the direction of the calibrated parking lot boundary line as a boundary. Traverse all GPS points to obtain all boundary areas. S2.12 If the obtained boundary line contains only one boundary line segment, then directly use the boundary line segment as the parking lot boundary area and output it to S2.2; if multiple boundary line segments are obtained, then each boundary line segment is used as a boundary candidate set, and each boundary line segment in the set is selected in turn as the processing object and output to S2.2 for subsequent processing.

5. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 4, characterized in that, In step S2.2, the specific steps for identifying parking spaces based on boundary areas are as follows: S2.21 Receive the boundary region or boundary candidate set output by S2.1, divide the image into several blocks according to the segmentation ratio, and calculate the azimuth angle of each segmented block to its adjacent points on the GPS track. And save it to get the sequence. Cluster these sequences sequentially to obtain a set of parking spaces. ; If there are overlapping parking spaces, the overlapping area will be removed from the original set; S2.

22. For each segment within a parking space, calculate its average azimuth angle. : ； in, This indicates the number of azimuth angles within the segmented block. Indicates the first In the segmented block, the first One azimuth angle; S2.

23. Compare the azimuth sequence of each segment with the average azimuth sequence, and calculate the average distance between the two sequences. : ； The average distance is used as the relation value of the corresponding segmentation block. ,Right now ; S2.

24. Sum the relationship values ​​between the line segments on the edge of each segmented region and the parking space segmentation blocks to obtain the sum of the relationship values ​​between each parking space segmentation block and its corresponding block. ; S2.25, Set the relation value Greater than the preset corner threshold The segmented blocks are identified as corner candidate areas for parking spaces, and the relation values ​​are... Less than or equal to The segmented blocks are determined as candidate edge regions for parking spaces; For each parking space in the polygonal parking space candidate set, sum the relationship values ​​between each edge and the corner candidate regions of the parking space, and take the parking space candidate region with the highest relationship value as the parking space output. If the sum of the relationship values ​​of all candidate regions is... All are below the preset screening threshold If so, then the parking lot will be abandoned.

6. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 1, characterized in that, Specifically, S3 includes: S3.1 Based on the parking space information obtained in S2, generate a unique identification code for each parking space, and record its GPS coordinates, charging pile location and occupancy status to establish a parking space status information database. S3.2, Preset charging decision threshold The drone monitors the remaining battery power in real time. ; S3.3, if The drone then takes pictures and samples according to the planned inspection path, uses a telephoto lens to obtain images of the parking space area, classifies and identifies the actual occupancy status of the parking spaces through image classification, and selects the best unoccupied parking spaces based on the S3.1 information database. After landing, it identifies the corresponding identification code, pairs the charging pile with the parking space, and starts charging. like The drone will then fly directly to the pre-planned emergency landing point. S3.4 After charging or making an emergency landing, the drone updates the occupancy status in the parking space status information database and uses the current landing point as the takeoff point for subsequent inspection tasks.

7. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 6, characterized in that, Specifically, S4 includes: S4.1, Preset minimum power requirement for inspection After the drone completes S3, it reads the current remaining battery power. ; S4.2, if Then take off from the current landing point and set the flight inspection path according to the inspection target area; S4.3 The drone cruises and takes pictures according to the planned path, and uses a pixel-based image stitching algorithm to stitch together each captured area map to generate a complete panoramic image of the road surface. S4.4 Use a neural network model trained from the dataset to identify vehicles in panoramic images and extract obstacle points; S4.5 Extract the identified obstacle points, generate a grid using Voronoi Diagrams, label the area occupied by the vehicle, and divide the overall identification area into several sub-regions using the edge information-based region division method, and establish an independent region identifier and status record table for each sub-region. like If the battery level is low, it will continue to charge at the current charging point until the battery reaches its maximum capacity. Then, execute steps S4.2 through S4.

5.

8. The method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 7, characterized in that, Specifically, S4.3 includes: S4.

31. For each area map to be stitched, sort the data of points in the GPS track that contain that area map in ascending order, and calculate the slope of each data point. Mark the coordinates of each point as The distance between two adjacent points is calculated. : ； in, and These represent the numbers on the GPS track. The point and the first The coordinates of a point in the preset projection direction; S4.

32. For each point, calculate its distance to the previous point and the average distance to all its neighboring points. proportion : ； ； in, This represents the total number of points on the GPS track. This represents the average distance between all adjacent points. Indicates the first The ratio of segment distance to average distance; like ,in If the preset segmentation weight threshold is not met, the current point will be used as the dividing point; otherwise, the current point will be used as an undivided point. S4.

33. Filter the obtained boundary points and take the point with the farthest distance as the first boundary point. If the distance from the boundary point to the farthest point is less than the distance between the nearest points between the boundary points, delete the boundary points between adjacent boundary points. Then calculate the distance between each boundary point and the farthest point, and take the point with the farthest distance as the new boundary point. Repeat this process until all boundary points meet the requirements. S4.

34. Based on the selected boundary points, the GPS track is divided into several continuous intervals. Each interval is composed of two adjacent boundary points and all the unbounded points between them. Each boundary point is connected to its nearest previous point, and two adjacent boundary points are connected to form several rectangular areas. Each rectangular area corresponds to a splicing sub-area. S4.

35. Use a feature-based block method to find the non-deformable regions within each stitching sub-region, and stitch the images of each stitching sub-region at their corresponding original positions to generate a panoramic image.

9. A method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 7, characterized in that, Specifically, S4.5 includes: S4.

51. For obstacle areas, use Voronoi Diagrams to obtain the dividing lines of the corresponding areas, use edge detection to extract the boundary lines of the obstacle areas, and intersect the dividing lines in the areas with the boundary lines to obtain the coordinates of the intersection points. S4.

52. For each intersection point, starting from that point, traverse the boundary lines adjacent to it. Let the Euclidean distance between the two points be... ,like ,in If the preset boundary merging distance threshold is not met, then delete one of the boundary lines until all intersection points have been traversed, and use the remaining boundary lines as edge candidate regions; S4.53, for each edge candidate region, treat it as a polygon in the Voronoi Diagram, and let the i-th edge candidate region be... The side length is The Euclidean distance from the midpoint of one side to the midpoint of the adjacent side is . Then the proportion corresponding to this side length for: ； in, Indicates the first The length of the strip, Let represent the Euclidean distance between the midpoint of an edge and the midpoint of its adjacent edge; normalize the sum of all weights to obtain the normalized weight. : ； in, This represents the sum of the weights corresponding to all edges; right Sort the edges in ascending order. For the edges corresponding to the lengths of the first K edges, find the midpoint and connect the midpoint to the center point of the candidate edge region. Pair the closed region obtained by the connection with the parking space, where K is the preset number of candidate edges. S4.

54. Based on the pairing results, the closed areas that are successfully paired with parking spaces are marked as vehicle-occupied areas, and the closed areas that are not paired with any parking spaces are marked as other obstacle areas. Based on the spatial relationship of each closed area, the overall identification area is divided into several sub-areas. Each sub-area contains at least one parking space and its surrounding environment information. An independent area identification and status record table is established for each sub-area.

10. A method for taking photos of unmanned aerial vehicle (UAV) parking inspections according to claim 1, characterized in that, Specifically, S5 includes: S5.1 For each parking space, record its corresponding identification code, location, whether it is occupied, and charging pile information, and plan the parking location according to the historical records in the next planning. S5.2 For the inspection area, based on the sub-areas divided in S4 and their corresponding status record tables, the information inspected by the drone is compared with the status record tables of each sub-area to obtain the inspection results of each sub-area, generate an inspection report containing the parking space occupancy status and obstacle distribution of each sub-area, and upload it to the cloud. S5.3 At the end of the inspection, the drone will land in the designated landing area to charge. If the drone does not reach the landing area within the designated time, it will be determined that the drone is damaged or has insufficient power to reach the landing point. When the drone is low on power, it will land in a nearby charging area and send an alarm message to the staff.

Images