Unmanned aerial vehicle path optimization method and system for high-speed service area distribution

Abstract

The invention relates to the technical field of path planning of unmanned aerial vehicles through non-electrical variable control, in particular to an unmanned aerial vehicle path optimization method and system for high-speed service area delivery, and the method comprises the steps: carrying out the path planning according to the unmanned aerial vehicle delivery order information and the initial position of an unmanned aerial vehicle, and determining a first delivery path, when the unmanned aerial vehicle arrives at a preset landing flight area along the first distribution path, identifying a motion track of a motion entity in the landing flight area; analyzing the static density distribution and the dynamic density distribution of the motion entity according to the motion track, and fusing the analysis results of the static density distribution and the dynamic density distribution to obtain the density distribution of the motion entity corresponding to each candidate landing point; selecting an optimal landing point from the plurality of candidate landing points according to the density distribution of the motion entities; and according to the optimal landing point, optimizing the first delivery path to obtain a second delivery path, so that the unmanned aerial vehicle lands along the second delivery path and completes delivery.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning for unmanned aerial vehicles (UAVs) through the control of non - electrical variables, and particularly relates to a method and system for optimizing the path of UAVs for distribution in high - speed service areas. Background Art

[0002] With the development of UAV technology, UAV distribution in highway service areas has gradually become popular with the "low - altitude economy". For UAV distribution in highway service areas, after a user parks their vehicle in the parking lot of the service area, they can place orders with restaurants and stores in the service area through a mobile terminal. After the restaurants and stores complete meal preparation and goods allocation, the UAV directly delivers items point - to - point. Along a preset flight path, it transports food and goods to the user's side, greatly facilitating the user's consumption in the service area.

[0003] Among them, based on UAV control technologies such as high - precision GPS (Global Positioning System) and communication networking, real - time data transmission can be ensured, and the high - efficiency distribution experience also helps to relieve congestion in service facilities in the service area. However, during the distribution process of UAVs in the service area, the landing location is determined according to the location where the user places an order. Generally, users are located in parking spaces in the service area, and this location usually does not have the conditions for the UAV to land directly. The traditional method is to set a fixed landing area and let users go to the fixed landing area to pick up goods. This method still requires users to walk a certain distance, resulting in low distribution efficiency.

[0004] Therefore, how to improve the efficiency of UAV distribution in highway service areas is an urgent problem to be solved currently. Summary of the Invention

[0005] In order to solve the technical problem of how to improve the efficiency of UAV distribution in highway service areas, the purpose of the present invention is to provide a method and system for optimizing the path of UAVs for distribution in high - speed service areas. The specific technical solutions adopted are as follows: An embodiment of the present application provides a method for optimizing the path of UAVs for distribution in high - speed service areas. The method includes: Obtain the UAV distribution order information and the initial position of the UAV within the target highway service area; Perform path planning based on the UAV distribution order information and the initial position of the UAV to determine the first distribution path. When the UAV arrives at a preset landing flight area along the first distribution path, identify the movement trajectories of moving entities within the landing flight area, where the landing flight area includes multiple candidate landing points; Analyze the static density distribution and dynamic density distribution of the moving entity according to the movement trajectory, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point; Select the optimal landing point from the multiple candidate landing points according to the moving entity density distribution; Optimize the first delivery path according to the optimal landing point to obtain a second delivery path, so that the drone lands along the second delivery path and completes the delivery.

[0006] In an embodiment of the present application, the path planning according to the drone delivery order information and the initial position of the drone to determine the first delivery path includes: Obtain the top-down map corresponding to the target highway service area; Rasterize the top-down map to obtain a raster map, and establish a three-dimensional coordinate system corresponding to the target highway service area, and correspond each grid in the raster map to the three-dimensional coordinate system; Determine the user position according to the drone delivery order information, and convert the user position and the initial position of the drone into coordinates in the raster map; Perform path planning according to the coordinates in the raster map to obtain the first delivery path.

[0007] In an embodiment of the present application, after determining the first delivery path, it further includes: Taking the coordinates corresponding to the user position in the raster map as the center, and determining the landing flight area according to a preset landing flight radius; When the distance between the coordinates of the drone and the center is equal to the landing flight radius when the drone is flying along the first delivery path, it is determined that the drone reaches the landing flight area.

[0008] In an embodiment of the present application, the identifying the movement trajectory of the moving entity in the landing flight area includes: Obtain the static image of the moving entity in the landing flight area at the current moment and the historical movement image of the moving entity in the landing flight area before the current moment through a pre-configured image acquisition device; Perform image recognition on the static image and the historical movement image to obtain the moving entity and the ground marker; Map the image pixels corresponding to the moving entity and the ground marker to the raster map to obtain the raster coordinates corresponding to the moving entity and the ground marker, and the raster coordinates are used to indicate the movement trajectory of the moving entity.

[0009] In an embodiment of the present application, the analysis of the static density distribution and the dynamic density distribution of the moving entity according to the motion trajectory includes: Determine the number of pixels of the moving entity in each grid at the current moment according to the grid coordinates, and determine the static density distribution according to the number of pixels; Determine a plurality of identical moving entities from the static image and the historical motion image according to the grid coordinates, and determine the dynamic density distribution according to the position changes of the plurality of identical moving entities in the grid map.

[0010] In an embodiment of the present application, the determination of the static density distribution according to the number of pixels includes: Compare the number of pixels of the moving entity in each grid with the average number of pixels at the current moment to obtain the moving entity density distribution of each grid; Combine the moving entity density distributions of each grid to obtain the static density distribution.

[0011] In an embodiment of the present application, the determination of a plurality of identical moving entities from the static image and the historical motion image according to the grid coordinates, and the determination of the dynamic density distribution according to the position changes of the plurality of identical moving entities in the grid map includes: Extract the feature descriptors corresponding to the moving entity according to the grid coordinates, and analyze the similarity between the moving entities in adjacent images according to the feature descriptors; Based on the similarity, determine a plurality of identical moving entities from the static image and the historical motion image; Obtain the grid motion vectors of the moving entity in each image according to the position changes of the plurality of identical moving entities in the grid map; Arrange the grid motion vectors of the moving entity in each image to obtain a trajectory sequence; Analyze the average change of the grid motion vectors in the trajectory sequence to obtain the predicted grid motion vector of the moving entity after the current moment, and perform curve fitting on the trajectory sequence to obtain the predicted grid motion direction of the moving entity after the current moment; Based on the predicted grid motion vector and the predicted grid motion direction, determine the predicted position of the moving entity; Compare the number of pixels of the moving entity at the predicted position with the average number of pixels after the current moment to obtain the moving entity density distribution at the predicted position, and obtain the dynamic density distribution according to the moving entity density distribution at the predicted position.

[0012] In one embodiment of the present application, the analysis result of fusing the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point includes: Obtain the speed of the moving entity at the current moment; Determine the density attention weight according to the speed; Based on the density attention weight, perform weighted summation on the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point.

[0013] In one embodiment of the present application, optimizing the first delivery path according to the optimal landing point to obtain a second delivery path, so that the unmanned aerial vehicle lands along the second delivery path and completes the delivery, includes: Obtain the position of the unmanned aerial vehicle at the current moment; Generate a second delivery path according to the position of the unmanned aerial vehicle at the current moment and the position of the optimal landing point; When the unmanned aerial vehicle lands along the second delivery path, update the optimal landing point until the delivery is completed.

[0014] An embodiment of the present application further provides an unmanned aerial vehicle path optimization system for high-speed service area delivery, and the system includes: An information acquisition module, configured to acquire unmanned aerial vehicle delivery order information and the initial position of the unmanned aerial vehicle within a target highway service area; A trajectory recognition module, configured to perform path planning according to the unmanned aerial vehicle delivery order information and the initial position of the unmanned aerial vehicle to determine a first delivery path, and recognize the movement trajectory of moving entities within the landing flight area when the unmanned aerial vehicle reaches a preset landing flight area along the first delivery path, where the landing flight area includes a plurality of candidate landing points; A density distribution analysis module, configured to analyze the static density distribution and the dynamic density distribution of the moving entities according to the movement trajectory, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point; An optimal landing point selection module, configured to select an optimal landing point from the plurality of candidate landing points according to the moving entity density distribution; A landing module, configured to optimize the first delivery path according to the optimal landing point to obtain a second delivery path, so that the unmanned aerial vehicle lands along the second delivery path and completes the delivery.

[0015] The present invention has the following beneficial effects: First, obtain the drone delivery order information and the initial position of the drone within the target highway service area; then, perform path planning based on the drone delivery order information and the initial position of the drone to determine the first delivery path. When the drone reaches the preset landing flight area along the first delivery path, identify the movement trajectories of moving entities within the landing flight area, where the landing flight area includes multiple candidate landing points; then, analyze the static density distribution and dynamic density distribution of the moving entities according to the movement trajectories, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point; then, select the optimal landing point from the multiple candidate landing points according to the moving entity density distribution; finally, optimize the first delivery path according to the optimal landing point to obtain the second delivery path, so that the drone lands along the second delivery path and completes the delivery. In this application, planning the first delivery path based on the order coordinates and the initial position of the drone can achieve long-distance and fast passage, and avoid redundant hovering in non-sensitive areas; after entering the preset landing flight area, dynamically adjust the path in combination with the real-time identified movement trajectories of moving entities, avoiding obstacle avoidance or waiting caused by fixed routes, and reducing the ineffective flight time in local areas; by real-time analyzing multiple candidate landing points in the landing area, directly select the optimal landing point near the user's dismounting position, avoiding the time waste caused by the movement of the user; at the same time, by fusing the static and dynamic distributions of moving entities, anticipate the feasible landing points in advance, so that the drone can plan the optimal landing path during flight, shortening the hovering time after arrival; the real-time analysis of the moving entity density can provide basic data support for multi-drone collaborative delivery. By sharing the density distributions of each area, the dispatching system can dynamically allocate orders to low-risk areas, avoiding multiple drones concentrating in high-person-flow / vehicle-flow areas and improving the overall delivery efficiency of the service area. Through the fusion analysis of the static density distribution and the dynamic density distribution, the limitations of a single data dimension are solved. The static density reflects the immediate distribution of moving entities within the landing area at the current moment, capturing sudden static obstacles, and the dynamic density predicts the future distribution trend based on the historical trajectories of moving entities, anticipating the movement directions of movable obstacles. Through an adaptive weight mechanism, balance the proportions of the two, so that the system focuses on dynamic analysis in scenarios of high-speed moving objects and on static analysis in scenarios of low-speed or stationary objects, accurately identifying high-risk areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 Schematic diagram of the implementation environment of a method for optimizing the path of an unmanned aerial vehicle for high-speed service area distribution provided by an embodiment of the present invention; Figure 2 Flow chart of a method for optimizing the path of an unmanned aerial vehicle for high-speed service area distribution provided by an embodiment of the present invention; Figure 3 Schematic diagram of the trajectory sequence provided by an embodiment of the present invention; Figure 4 Schematic diagram of curve fitting and predicted positions provided by an embodiment of the present invention; Figure 5 Schematic diagram of the structure of a system for optimizing the path of an unmanned aerial vehicle for high-speed service area distribution provided by an embodiment of the present invention. Detailed implementation manners

[0018] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, details the specific implementation manners, structures, features and effects of a method and system for optimizing the path of an unmanned aerial vehicle for high-speed service area distribution proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0019] It should be noted that the terms "first", "second", etc. in the specification of this application and the above accompanying drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order different from those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0021] The following specifically describes the specific solutions of a method and system for optimizing the path of an unmanned aerial vehicle for high-speed service area distribution provided by the present invention with reference to the accompanying drawings.

[0022] Please refer to Figure 1 ,Figure 1 Schematic diagram of the implementation environment of an unmanned aerial vehicle path optimization method for high-speed service area distribution provided by an embodiment of the present invention. As Figure 1 shown, the implementation environment includes a control terminal 101 and an unmanned aerial vehicle 102. The control terminal 101 may be a terminal device configured with an unmanned aerial vehicle path optimization system for high-speed service area distribution, including but not limited to a laptop computer, a tablet computer, a personal digital assistant, a PAD (tablet computer), a desktop computer, etc. with local computing capabilities; the unmanned aerial vehicle path optimization system for high-speed service area distribution may be implemented in the form of a target client, and the target client may be a video client, an instant messaging client, a browser client, etc. that support the unmanned aerial vehicle path optimization for high-speed service area distribution; the control terminal 101 may communicate with the unmanned aerial vehicle 102 through a network, which may include but not limited to a wireless network, where the wireless network includes: Bluetooth, WIFI (Wireless Fidelity, a technology that allows electronic devices to connect to a wireless local area network), and other networks that implement wireless communication. The control terminal 101 may include but not limited to a human-computer interaction screen, a processor, and a memory. The human-computer interaction screen may be used to display unmanned aerial vehicle delivery order information, the initial position of the unmanned aerial vehicle, the first delivery path, the optimal landing point, and the second delivery path. The processor may be used to respond to human-computer interaction operations, execute corresponding operations, or generate corresponding instructions.

[0023] As an optional manner, the control terminal 101 may also be a server, which may be a single server, a server cluster composed of multiple servers, or a cloud server. The above is only an example, and no limitation is made thereto in this embodiment.

[0024] As an optional manner, the following steps of the unmanned aerial vehicle path optimization method for high-speed service area distribution may be executed on the control terminal 101: Obtain the unmanned aerial vehicle delivery order information and the initial position of the unmanned aerial vehicle within the target high-speed service area; Perform path planning according to the unmanned aerial vehicle delivery order information and the initial position of the unmanned aerial vehicle to determine the first delivery path. When the unmanned aerial vehicle reaches a preset landing flight area along the first delivery path, identify the movement trajectories of moving entities within the landing flight area, where the landing flight area includes multiple candidate landing points; Analyze the static density distribution and dynamic density distribution of the moving entities according to the movement trajectories, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point; Select an optimal landing point from the multiple candidate landing points according to the density distribution of the moving entities; Optimize the first delivery path according to the optimal landing point to obtain a second delivery path, so that the drone lands along the second delivery path and completes the delivery.

[0025] In the above method, planning the first delivery path based on the order coordinates and the initial position of the drone can achieve long-distance and fast passage, and avoid redundant hovering in non-sensitive areas. When entering the preset landing flight area, the path is dynamically adjusted by combining the real-time recognized trajectories of moving entities, avoiding obstacle avoidance or waiting caused by fixed routes, and reducing the ineffective flight time in local areas. By analyzing multiple candidate landing points in the landing area in real time, directly select the optimal landing point near the user's dismounting location, avoiding the time waste caused by the user's movement. At the same time, by fusing the static and dynamic distributions of moving entities, the feasible landing points can be predicted in advance, enabling the drone to plan the optimal landing path during flight, shortening the hovering time after arrival. The real-time analysis of the density of moving entities can provide basic data support for multi-drone collaborative delivery. By sharing the density distribution of each area, the dispatching system can dynamically allocate orders to low-risk areas, avoiding multiple drones concentrating in high-person flow / vehicle flow areas, and improving the overall delivery efficiency of the service area. Through the fusion analysis of static density distribution and dynamic density distribution, the limitations of a single data dimension are solved. The static density reflects the immediate distribution of moving entities in the landing area at the current moment, capturing sudden static obstacles. The dynamic density predicts the future distribution trend based on the historical trajectories of moving entities, predicting the movement direction of movable obstacles. The adaptive weight mechanism balances the proportion of the two, so that the system focuses on dynamic analysis in high-speed moving object scenarios and static analysis in low-speed or static object scenarios, accurately identifying high-risk areas.

[0026] As an optional example, the execution subject of the above drone path optimization method for high-speed service area delivery in this embodiment is not limited. The above drone path optimization method for high-speed service area delivery can be executed on the control terminal 101. For example, when the control terminal 101 is a desktop computer, some or all of the steps of the above drone path optimization method for high-speed service area delivery can be executed on the desktop computer.

[0027] The above part introduced the content of the exemplary implementation environment applying the technical solution of the present application. Next, the drone path optimization method for high-speed service area delivery of the present application will be continued to be introduced.

[0028] To solve the problem of how to improve the efficiency and safety of drones during distribution in highway service areas in the prior art, embodiments of the present application respectively propose a method for optimizing the path of drones for distribution in highway service areas and a system for optimizing the path of drones for distribution in highway service areas. The following will describe these embodiments in detail.

[0029] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of a method for optimizing the path of drones for distribution in highway service areas provided by an embodiment of the present invention. This method can be applied to Figure 1 the implementation environment shown. It should be understood that this method can also be applicable to other exemplary implementation environments and be specifically executed by devices in other implementation environments. This embodiment does not limit the implementation environment applicable to this method.

[0030] As Figure 2 shown, in an exemplary embodiment, the method for optimizing the path of drones for distribution in highway service areas at least includes steps S210 to S250, which are introduced in detail as follows: In step S210, obtain the drone delivery order information and the initial position of the drone within the target highway service area.

[0031] Among them, the drone delivery order information refers to the delivery requirement data submitted by users through the order system within the highway service area, and the core includes the geographical location (such as parking space coordinates) when the user places an order, which is used to determine the delivery target area. This information is obtained through the internal positioning technology of the service area (such as base station positioning), and the accuracy needs to meet the requirements for identifying the drone landing area.

[0032] Among them, the initial position of the drone refers to the starting point of the drone's delivery task, which is usually set as a fixed safe area within the service area (such as near service facilities). This location prohibits unauthorized personnel and vehicles from freely entering and exiting to ensure the safety during the takeoff stage.

[0033] In step S220, perform path planning based on the drone delivery order information and the initial position of the drone to determine the first delivery path. When the drone reaches the preset landing flight area along the first delivery path, identify the movement trajectories of moving entities within the landing flight area, where the landing flight area includes multiple candidate landing points.

[0034] Among them, the drone delivery in highway service areas usually involves delivering the goods and items required by users after obtaining user orders. During this process, relevant communication technologies can be used to obtain the location of the user when placing the order. However, in the service area, the addresses where users place orders are mostly within parking spaces. The parking areas on both sides and the road ahead are generally used as the target locations for the drone to land. There may be dynamic targets such as pedestrians and vehicles in the parking areas on both sides and the road ahead. Therefore, when the drone is about to enter the descent stage, it is necessary to identify the trajectories of pedestrians and vehicles on the ground and select the optimal landing location to complete the delivery.

[0035] Among them, path planning is to generate the flight route of the drone from the starting point to the target area through an algorithm based on order information and the initial position. This process needs to combine the map features of the service area, avoid no-fly zones and high-risk areas, and balance flight efficiency and safety.

[0036] Among them, the first delivery path refers to the global initial route generated by path planning for the long-distance flight of the drone from the starting point to the landing flight area. This path mainly uses the low-altitude stable flight mode (such as a height of 30 - 50 meters), focusing on quickly reaching the target area. When performing path planning based on order information and the initial position, the service area map can be obtained and rasterized, converting the user position and the initial position into map coordinates, and generating a global route (the first delivery path) from the starting point to the user position to ensure the efficiency and compliance of long-distance flight.

[0037] Among them, the landing flight area is a local area demarcated with the user's order placement location as the center (such as a circular area with a radius of dozens of meters), which is a key identification area before the drone enters the landing stage. This area contains multiple candidate landing points (such as the available open spaces around the parking space where the user is located), and it is necessary to analyze the ground environment in real time.

[0038] Among them, the moving entity refers to the dynamic targets within the landing flight area, including objects such as pedestrians and vehicles that may affect the safety of the drone's landing. It is necessary to detect and track them in real time through image recognition technology.

[0039] Among them, the motion trajectory is the sequence of position changes of the moving entity within continuous time. By collecting multiple frames of images through the camera carried by the drone and combining the feature matching algorithm to analyze the position migration of the same entity in adjacent frames, trajectory data is formed.

[0040] Among them, the candidate landing point is an alternative position within the landing flight area that meets the basic landing conditions (such as open spaces without obstacles, parking space gaps). It is necessary to evaluate the safety through density analysis and finally select the optimal landing point.

[0041] In step S230, based on the motion trajectory, analyze the static density distribution and dynamic density distribution of the moving entity, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point.

[0042] Among them, the static density distribution is to statistically analyze the immediate distribution of moving entities at each position within the landing area based on the image recognition results at the current moment, reflecting the spatial aggregation degree of people and vehicles in the current frame.

[0043] Among them, the dynamic density distribution is to predict the position distribution trend at a future moment through a trajectory fitting algorithm based on the historical trajectory data of the moving entity, reflecting the movement law and potential risks of movable obstacles.

[0044] Among them, the moving entity density distribution is comprehensive risk assessment data formed after fusing the static and dynamic density analysis results, used to quantify the safety level of each candidate landing point and provide a direct basis for the landing point selection.

[0045] Among them, when identifying the motion trajectory of moving entities within the landing area, the real-time image of the landing area can be collected through the UAV camera, and image recognition technology can be used to detect moving entities (such as pedestrians and vehicles), and the position changes of the entities can be tracked through multi-frame image matching (such as comparing feature points of adjacent frames) to form a continuous trajectory.

[0046] Among them, when fusing the static and dynamic density distributions, the static immediate data and dynamic prediction data can be combined through a weight mechanism (such as adjusting the weight according to the speed of the moving entity) to form a more comprehensive risk assessment result and avoid the limitations of a single data dimension.

[0047] In step S240, select the optimal landing point from the multiple candidate landing points according to the moving entity density distribution.

[0048] Among them, when selecting the optimal landing point, the position with the lowest density of people and vehicles and the highest safety can be screened from the candidate points according to the fused density distribution as the target point for the UAV to land.

[0049] In step S250, optimize the first delivery path according to the optimal landing point to obtain a second delivery path, so that the UAV lands along the second delivery path and completes the delivery.

[0050] Among them, the second delivery path is an optimized local fine path, used for the dynamic adjustment flight of the UAV within the landing area. The end point of the path is the optimally determined landing point in real time, with real-time adaptability and safety.

[0051] Among them, when optimizing the first delivery route to obtain the second delivery route, after entering the landing area, based on the optimally updated landing point in real time, the initial route can be locally adjusted to generate an accurate landing route adapted to the dynamic environment, ensuring that the drone completes the final stage of flight along a safe route.

[0052] Exemplarily, within a certain highway service area, a user places an order for food and beverage at position A in the parking space, and the drone departs from the fixed starting point S on the east side of the service area. The system first obtains the coordinates of the user's position A (latitude and longitude coordinates) and the starting point S of the drone. Then, the service area map is rasterized, and positions A and the drone starting point S are converted into raster coordinates; a global route from the drone starting point S to position A is generated, and the drone flies along this route at a height of 40 meters and a speed of 40 km / h to quickly approach the target area. When the drone is about 30 meters away from position A (i.e., enters the preset circular area), the camera is activated to collect ground images in real time to identify moving entities within the area (such as pedestrian B walking and vehicle C driving). Static density analysis: In the current frame, pedestrian B stays in grid X near position A, and vehicle C is passing through grid Y. Calculate the instantaneous density of each grid; Dynamic density analysis: Through the analysis of the historical 5-frame images, it is predicted that pedestrian B may move to grid Z, and vehicle C will drive out of the area along a fixed direction. Combining the static (instantaneous stay) and dynamic (vehicle about to leave) data, it is determined that the density of grid Y will be lower at a future moment and is selected as the optimal landing point. The original route is adjusted to generate a second delivery route from the current position to grid Y. The drone slowly descends and emits a prompt sound, and finally lands safely at grid Y, where the user can directly pick up the goods.

[0053] As can be seen from the above steps S210 to S250, in the solution proposed in this embodiment, based on the order coordinates and the initial position of the drone, the first delivery path is planned, which can achieve long-distance and fast passage, and avoid redundant hovering in non-sensitive areas. When entering the preset landing flight area, the path is dynamically adjusted in combination with the trajectory of the moving entity recognized in real time, avoiding obstacle avoidance or waiting caused by a fixed route, and reducing the ineffective flight time in the local area. By analyzing multiple candidate landing points in the landing area in real time, the optimal landing point is directly selected near the user's getting-off location, avoiding the time waste caused by the user's movement. At the same time, by fusing the static and dynamic distributions of the moving entities, the feasible landing points are predicted in advance, so that the drone can plan the optimal landing path during the flight, shortening the hovering time after arrival. The real-time analysis of the density of the moving entities can provide basic data support for multi-drone collaborative delivery. By sharing the density distribution of each area, the dispatching system can dynamically allocate orders to low-risk areas, avoiding multiple drones concentrating in high-person-flow / high-vehicle-flow areas, and improving the overall delivery efficiency of the service area. Through the fusion analysis of the static density distribution and the dynamic density distribution, the limitations of a single data dimension are solved. The static density reflects the immediate distribution of the moving entities in the landing area at the current moment, capturing sudden static obstacles. The dynamic density predicts the future distribution trend based on the historical trajectory of the moving entities, predicting the movement direction of the movable obstacles. The proportion of the two is balanced through an adaptive weight mechanism, so that the system focuses on dynamic analysis in the scene of high-speed moving objects and focuses on static analysis in the scene of low-speed or static objects, accurately identifying high-risk areas.

[0054] In an embodiment of the present application, the path planning according to the drone delivery order information and the initial position of the drone to determine the first delivery path includes: Obtain the top-down map corresponding to the target highway service area; Rasterize the top-down map to obtain a raster map, and establish a three-dimensional coordinate system corresponding to the target highway service area, and correspond each grid in the raster map to the three-dimensional coordinate system; According to the drone delivery order information, determine the user's position, and convert the user's position and the initial position of the drone into coordinates in the raster map; Perform path planning according to the coordinates in the raster map to obtain the first delivery path.

[0055] Among them, since the starting delivery point of the drone is generally a commercial goods supply point with a fixed position, the influence of pedestrians and vehicles on the starting point is not considered. When the drone takes off from the starting point, it enters the delivery process. When it is far from the landing target, it flies at a relatively high altitude until it reaches the range of the landing target area, and analyzes the captured ground image to obtain a real-time dynamic information image of the ground.

[0056] Among them, the top-down map corresponding to the target highway service area refers to the planar map of the service area presented from a horizontal perspective, covering geographical elements such as roads, parking spaces, and buildings within the service area, and is used to construct the spatial reference for the UAV path planning. This map needs to truly reflect the actual layout of the service area, including the lane directions, parking area divisions, etc., and provide basic data for subsequent rasterization processing.

[0057] Among them, rasterization is the process of dividing the top-down map into regular grid cells (rasters). Each raster corresponds to a fixed area (such as 1 square meter) in the actual space of the service area. By discretizing the continuous geographical space, it is transformed into structured data that can be processed by a computer, facilitating subsequent spatial analysis and path calculation.

[0058] Among them, the raster map is the service area map after rasterization processing, composed of multiple regularly arranged rasters. Each raster can be marked as a feasible area (such as roads, open spaces) or a no-fly area (such as the top of a building), and carries information such as geographical coordinates and obstacle distributions, becoming the digital carrier for UAV path planning.

[0059] Among them, the three-dimensional coordinate system is a spatial coordinate system established based on the ground of the service area, usually including the x-axis (horizontal direction), y-axis (vertical direction), and z-axis (height direction). The x-axis and y-axis correspond to the ground plane, and the z-axis represents the UAV flight height, which is used to associate the two-dimensional raster map with the actual three-dimensional spatial position to achieve the three-dimensional planning of the UAV flight trajectory.

[0060] Among them, when corresponding each raster in the raster map to the three-dimensional coordinate system, a unique three-dimensional coordinate (x, y, z) can be assigned to each raster, where the z-axis value can be preset according to the terrain of the service area or flight height rules (such as the default z = 0 representing the ground). Through this corresponding relationship, the position of the UAV, the user coordinates, and the obstacle positions can all be uniformly mapped to the same coordinate system, facilitating the coordinate calculation and spatial analysis of the path planning algorithm.

[0061] Among them, when determining the user's position according to the UAV delivery order information, the geographical location (such as GPS coordinates, base station positioning coordinates within the service area) when the user places an order can be obtained through the order system, and combined with the service area map to match to a specific location (such as near a certain parking space, a certain building), to clarify the target point of the path planning.

[0062] Among them, when converting the user's position and the UAV's initial position into coordinates in the raster map, through the coordinate conversion algorithm, the geographical coordinates (such as longitude and latitude) of the user's position and the UAV's initial position can be mapped to the row and column numbers in the raster map (such as the raster in the i-th row and j-th column), so that both are in the same digital spatial reference, facilitating the path planning algorithm to search for and calculate the path based on the grid cells.

[0063] Among them, when performing path planning according to the coordinates in the grid map to obtain the first delivery path, based on the distribution of the feasible area and no-fly area of the grid map, a path planning algorithm can be used to search for the optimal path between the grid coordinates of the initial position of the UAV and the user position. The path needs to avoid no-fly grids (such as buildings and crowded areas), and preferably select a grid sequence with a short distance and high safety to form a global route from the starting point to the landing flight area.

[0064] Exemplarily, a certain highway service area has a rectangular layout, including an east-west main lane, a north-south parking strip, and a central service building. The user places an order at parking space P in the north area, and the initial position of the UAV is the fixed takeoff and landing point S beside the service building on the east side. Obtain the plane drawing of the service area through surveying and mapping or electronic maps, and mark elements such as lanes, parking spaces, and service buildings. Divide the map into a 50×50 grid matrix, with each grid corresponding to an actual area of 1 square meter; establish a three-dimensional coordinate system, with the northwest corner of the service area as the origin (x = 0, y = 0, z = 0), the x-axis extending eastward, the y-axis extending southward, and the z-axis representing the flight height (by default, z = 30 meters is the low-altitude flight layer). The user position P is located at (x = 20, y = 35, z = 0) through base station positioning, corresponding to the grid at the 20th column and 35th row in the grid map; the initial position S of the UAV is (x = 10, y = 10, z = 0), corresponding to the grid at the 10th column and 10th row. The algorithm identifies the service building area in the grid map as a no-fly grid (such as grids with x = 15 - 25, y = 20 - 30); plan the path from S(10, 10) to P(20, 35): go east along the x-axis to x = 18, and then go south along the y-axis to y = 35, avoiding the no-fly area of the central service building, to form the first delivery path.

[0065] In this embodiment, rasterization transforms complex geographical spaces into discrete grid cells, enabling a computer to rapidly analyze feasible paths through matrix operations, avoiding complex geometric calculations of traditional continuous space algorithms, significantly enhancing the path planning speed, and being applicable to the real-time scheduling requirements of unmanned aerial vehicles (UAVs). By means of a three-dimensional coordinate system, the user's position, the UAV's position, and obstacles are uniformly mapped onto a raster map, eliminating coordinate deviations of different positioning systems (such as GPS and base station positioning), achieving an improvement in path planning accuracy from the meter level to the grid level (such as an accuracy of 1 square meter), and ensuring that the UAV flies precisely along a preset route. The raster map can pre-mark no-fly zones (such as buildings and high-voltage power lines), and the path planning algorithm automatically avoids such grids, avoiding static risks that may be overlooked in traditional manual route planning. Meanwhile, rasterized data provides a basis for the real-time identification and avoidance of subsequent dynamic obstacles (such as moving vehicles), forming a complete safety system combining static planning with dynamic adjustment. The rasterization model can flexibly adapt to service areas of different shapes and scales (such as rectangles and irregular polygons), balancing calculation accuracy and efficiency by adjusting the grid granularity (such as 0.5 m × 0.5 m or 2 m × 2 m). For service areas with complex terrains (such as sections with uphill and downhill slopes), different height levels can be distinguished through the z-axis value of the three-dimensional coordinate system to achieve three-dimensional path planning (such as low-altitude obstacle avoidance and high-altitude fast passage). The structured data of the raster map and the three-dimensional coordinate system can be seamlessly docked with the UAV navigation system, the real-time communication module, and the multi-UAV collaborative scheduling platform, providing a unified data standard for subsequent function expansion (such as multi-UAV path coordination and dynamic traffic data access), and enhancing the scalability and compatibility of the system.

[0066] In one embodiment of the present application, after determining the first delivery path, the following steps are further included: Taking the coordinates corresponding to the user's position in the raster map as the center, and determining the landing flight area according to a preset landing flight radius; When the distance between the coordinates of the UAV and the center is equal to the landing flight radius during the flight of the UAV along the first delivery path, it is determined that the UAV arrives at the landing flight area.

[0067] Among them, the landing flight radius refers to the radius parameter of a preset circular area centered on the coordinates of the user's position in the raster map. This radius is used to delimit the key recognition range before the UAV enters the landing stage, that is, the landing flight area. When the distance between the UAV and the center of the user's position is equal to this radius, it triggers the UAV to start the real-time image acquisition and moving entity recognition process, providing a data basis for subsequent dynamic path optimization. This radius needs to be set according to factors such as the flight speed of the UAV and the image recognition delay to ensure that the UAV has sufficient time to complete the ground environment analysis and landing point selection.

[0068] Exemplarily, the user places an order at the parking space coordinates (x0, y0) in the highway service area, and a preset landing flight radius is R (for example, the actual distance is 30 meters, corresponding to 30 grid units in the grid map). Taking (x0, y0) as the center, the circular area with a radius of R is the landing flight area, which covers multiple candidate landing points (such as adjacent parking spaces, lane gaps, etc.) around the user's position. When the drone flies along the first delivery path to the coordinates (x1, y1), calculate its distance from the center (x0, y0). When d = R, it is determined that the drone has reached the landing flight area, and the camera is immediately activated to collect the ground image of this area, and identify the real-time positions and trajectories of moving entities such as pedestrians and vehicles.

[0069] In this embodiment, by presetting the landing flight radius, it is ensured that the drone starts to identify the ground conditions at a certain safe distance from the user's position (instead of directly arriving above), avoiding emergency obstacle avoidance or landing failure caused by sudden obstacles at close range. For example, if the flight speed of the drone is 10 m / s, a radius of 30 meters can provide 3 seconds of advance analysis time, which is sufficient to complete multi-frame image recognition and trajectory prediction, improving the decision-making calmness. The landing flight area limits the analysis scope to the local space around the user's position, avoiding global real-time calculation of the entire service area and significantly reducing the data processing volume.

[0070] In an embodiment of the present application, the identifying the movement trajectories of moving entities in the landing flight area includes: Obtaining a static image of the moving entity in the landing flight area at the current moment, and a historical movement image of the moving entity in the landing flight area before the current moment, through a pre-configured image acquisition device; Performing image recognition on the static image and the historical movement image to obtain moving entities and ground markers; Mapping the image pixels corresponding to the moving entities and the ground markers to the grid map to obtain the grid coordinates corresponding to the moving entities and the ground markers, and the grid coordinates are used to indicate the movement trajectories of the moving entities.

[0071] Among them, the pre-configured image acquisition device refers to devices such as cameras mounted on the drone, which are used to collect ground images of the landing flight area in real time. This device needs to meet the requirements of image clarity and acquisition frequency, so as to accurately identify moving entities and ground details and provide a data basis for trajectory analysis.

[0072] Among them, the static image refers to the ground image of the landing flight area collected by the drone at the current moment, which reflects the instant scene at this moment and is used to analyze the current positions and distributions of moving entities.

[0073] Among them, the historical motion images refer to multiple frames of ground images continuously collected by the drone before the current moment, which are used to extract the position changes of moving entities in the time series to form historical data for trajectory analysis.

[0074] Among them, image recognition is to use computer vision algorithms (such as object detection models) to process static images and historical motion images, identify moving entities (such as pedestrians, vehicles) and ground markers (such as parking space markings, road boundaries) in the images, and mark their positions in the images.

[0075] Among them, ground markers refer to fixed geographical features within the landing flight area, such as clear parking space markings, road centerlines, building outlines, etc. These markers are used to assist in calibrating the coordinate mapping between image pixels and the grid map to ensure the accuracy of position conversion.

[0076] Among them, image pixels are the smallest units that make up an image, and each pixel corresponds to a point in the image. By analyzing the pixel distribution, the contours and positions of moving entities and the shapes of ground markers can be identified.

[0077] Among them, grid coordinates are the unique numbers (such as row and column numbers) of each grid in the grid map of the landing flight area, corresponding to specific positions in the actual space of the service area (such as a 1-square-meter area). Grid coordinates are used to convert the pixel positions in the image into actual geographical coordinates to achieve the spatial quantization of the moving entity trajectory.

[0078] Among them, when performing image recognition on static images and historical motion images to obtain moving entities and ground markers, the dynamic targets (moving entities) and static references (ground markers) in the images can be detected through image recognition algorithms. For example, the contours of pedestrians and vehicles can be segmented from the image, and the positions of parking space markings can be marked to provide annotation data for subsequent coordinate mapping.

[0079] Among them, when mapping the image pixels corresponding to the moving entities and ground markers to the grid map to obtain grid coordinates, based on the known actual positions of the ground markers (such as the grid coordinates of parking space markings), a proportional mapping relationship between image pixels and grid coordinates can be established (such as 100 pixels corresponding to 1 meter). Furthermore, the pixel coordinates of the moving entity in the image can be converted into the actual position in the grid map to form a grid coordinate sequence of the moving entity, that is, the motion trajectory.

[0080] Exemplarily, an existing YOLO model is used to identify moving entities and markers in high-definition images captured by a drone. The moving entities mainly include complete personnel and vehicles within the image, and the markers are mainly complete parking space markings and numbers on the ground without occlusion; this model needs to be retrained and parameter-tuned. The training set is images of personnel, vehicles, and markers (markers refer to complete empty parking lines and corresponding numbers on the ground) manually annotated. Using the parking markings and numbers on the ground in the images respectively, according to the corresponding relationship between the actual ground and the image markings, the horizontal and vertical scale ratios of the image, as well as the position corresponding relationship, are calculated respectively, and each image pixel is mapped to the actual grid coordinates, and then the identified personnel, vehicles, etc. in the image are transformed into the corresponding three-dimensional coordinates. Among them, the three-dimensional coordinates of people and vehicles are represented by the central coordinate mapping relationship in the image. For example, if the grid coordinate corresponding to the central coordinate of a person identified in the image is , then it is considered that the grid coordinate of this person in the real world is ; Exemplarily, after the drone enters the landing flight area (radius 30 meters), it captures images at a frequency of 5 frames per second: at the current moment, a static image is captured, showing that there is pedestrian A and vehicle B near position A, and the ground marker is a clear parking space marking; in the first 4 frames of images captured at historical moments, the pixel coordinates of pedestrian A and the pixel coordinates of vehicle B. The pedestrian A, vehicle B, and parking space markings in the static image are identified through the object detection model and marked as moving entities and ground markers. The historical images are identified frame by frame to track the position changes of pedestrian A and vehicle B in consecutive images. Given that the actual grid coordinate corresponding to the parking space marking is (20, 35), and its pixel coordinate in the image, the ratio of pixel to grid is calculated as 100 pixels / meter (i.e., 1 pixel = 0.01 meter). The current pixel coordinate of pedestrian A is converted to the grid coordinate (22, 36).

[0081] Similarly, the pixel coordinate sequence of pedestrian A in the historical images is converted to the grid coordinate sequence [(21, 34), (21, 35), (22, 35), (22, 36)] to form its movement trajectory.

[0082] In this embodiment, through high-frequency image acquisition (such as multiple frames per second) and target recognition technology, real-time detection of moving entities such as pedestrians and vehicles in the landing area is achieved, avoiding misdetection or missed detection problems of traditional sensors (such as radar) in complex environments, and ensuring the timely response of the UAV to dynamic obstacles. Combining the immediate position of the static image and the trajectory sequence of the historical image, a complete spatio-temporal path of the moving entity can be constructed (such as the speed and direction from grid A' to grid B'), providing key data for subsequent dynamic density distribution prediction and improving the ability to predict movement trends (such as identifying the intention of pedestrians to cross the lane in advance). Using ground markers (such as fixed parking space markings) as a calibration reference, precise alignment between the image space and the actual space is achieved through pixel-grid mapping, solving the problem of meter-level errors relying solely on GPS positioning, and improving the position perception accuracy of the UAV for moving entities to the grid level (such as 1 square meter), providing a reliable basis for safe landing point selection. The identified and mapped grid coordinate data can be directly used to calculate the static density (the number of entities in the current grid) and dynamic density (the future distribution predicted by the historical trajectory) of the moving entity, forming a complete risk assessment system to support closed-loop control, and ultimately realizing autonomous and safe navigation of the UAV in complex environments.

[0083] In one embodiment of the present application, the analyzing the static density distribution and the dynamic density distribution of the moving entity according to the motion trajectory includes: Determining the number of pixels of the moving entity in each grid at the current moment according to the grid coordinates, and determining the static density distribution according to the number of pixels; Determining a plurality of identical moving entities from the static image and the historical motion image according to the grid coordinates, and determining the dynamic density distribution according to the position changes of the plurality of identical moving entities in the grid map.

[0084] Among them, determining the number of pixels of the moving entity in each grid at the current moment according to the grid coordinates means that in the current frame image, based on the grid coordinates of the moving entity, the number of pixel points belonging to the moving entity (such as pedestrians, vehicles) in the image area corresponding to each grid is counted. For example, if a certain grid corresponds to a 100×100 pixel area in the image, and 30 pixels are occupied by pedestrians or vehicles, then the number of pixels of the moving entity in this grid is 30.

[0085] Among them, determining the static density distribution according to the number of pixels is to quantify the aggregation degree of the moving entity in each grid at the current moment through the proportional relationship between the number of pixels and the total number of pixels in the grid. For example, for the above-mentioned grid with 30 pixels, if the total number of pixels is 10,000 (corresponding to an actual area of 1 square meter), then the static density is 30 / 10,000 = 0.3%. The higher the density, the denser the people or vehicles in the grid, and the higher the landing risk.

[0086] Among them, according to the grid coordinates, multiple identical moving entities are determined from the static image and the historical motion image. By using the grid coordinate continuity of the moving entities in multiple frames of images and through a feature matching algorithm, moving targets belonging to the same physical entity in different images are identified. For example, in three consecutive frames of images, the feature of the moving entity corresponding to a certain grid coordinate sequence is highly similar, and it can be determined as the same pedestrian or vehicle.

[0087] Among them, the position changes of multiple identical moving entities in the grid map are to track the grid coordinate migration trajectory of the same moving entity within continuous time to form a position change sequence (such as moving from grid (10, 10) to (12, 10) and then to (13, 11)), which is used to analyze its moving direction, speed, and trend, providing data support for dynamic density distribution prediction.

[0088] Exemplarily, the recognition range of the image is the target area where the drone flies forward and lands. Due to the high recognition frequency of the camera, the far viewing angle of the drone from the target ground area, and the low moving speed of people, vehicles, etc. in the service area, the recognized people and vehicles will not only appear in one frame, but slowly enter the picture until they disappear. Therefore, according to the trajectory changes of the same person or vehicle moving entity recognized in a period of history (a period of time before the current time), trajectory curve fitting can be performed to obtain the future predicted density distribution (weight) of people and vehicles, which is used as dynamic data. At the same time, considering the mutability of the moving trajectories of people and vehicles, that is, the characteristic that the direction may suddenly change, the density distribution existing in each grid of the current frame is obtained as static data; because generally, when moving objects move at a high speed, it is difficult to suddenly change direction, so the speed magnitude of the moving entity is used as the combined weight of dynamic data and static data, and thus the weighted ground density distribution is obtained.

[0089] In an embodiment of the present application, the determining the static density distribution according to the number of pixels includes: Comparing the number of pixels of the moving entity in each grid with the average number of pixels at the current moment to obtain the moving entity density distribution of each grid; Combining the moving entity density distributions of each grid to obtain the static density distribution.

[0090] Exemplarily, during the flight of the drone, the obtained image is the area where the drone flies forward. Assuming that the image obtained at the current moment is the latest data, for each pixel recognized from the image, using its corresponding grid position, if the more the pixels of people and vehicles corresponding to the grid, the higher the density of people and vehicles, thereby calculating the moving entity density distribution of the grid.

[0091] The representation method of the moving entity density distribution of the grid can be: Among them, represents the density distribution of moving entities in the i-th grid; represents the number of pixels of moving entities in the i-th grid; represents the average number of pixels at the current moment.

[0092] Among them, the immediate risk is quantified by the proportion of moving entity pixels in the grid. The larger the value, the denser the people or vehicles in the grid.

[0093] Then, the grid ranges corresponding to the image can be calculated one by one to obtain the density distribution of moving entities in different grids. The density distributions of moving entities in all grids are arranged to obtain the static density distribution, denoted as .

[0094] In this embodiment, by comparing the number of pixels in each grid with the global average value at the current moment, the difference in the pixel number benchmark caused by different lighting conditions, image acquisition angles, or UAV flight heights is eliminated. For example, if the overall pixel brightness of the image is high when the lighting is strong, the average number of pixels will automatically adapt to this change, ensuring that the density data at different times has a unified measurement standard and avoiding misjudgment. Compared with directly using the absolute value of the pixel number, the comparison with the average value can more sensitively identify the density abnormal area. The average number of pixels is equivalent to performing low-pass filtering on the whole image data, which can effectively suppress the interference of accidentally appearing isolated pixels (such as noise points formed by the shaking of leaves and light reflection). For example, if there are 2 misdetected pixels in a grid due to reflection, and its number is close to the average value (assuming the average value is 3), the system will determine it as a normal density, avoiding unnecessary obstacle avoidance actions triggered by noise and improving the robustness of static analysis.

[0095] In an embodiment of the present application, the determining a plurality of identical moving entities from the static image and the historical motion image according to the grid coordinates, and determining the dynamic density distribution according to the position changes of the plurality of identical moving entities in the grid map includes: Extracting a feature descriptor corresponding to the moving entity according to the grid coordinates, and analyzing the similarity between the moving entities in adjacent images according to the feature descriptor; Based on the similarity, determining a plurality of identical moving entities from the static image and the historical motion image; Obtaining the grid motion vector of the moving entity in each image according to the position changes of the plurality of identical moving entities in the grid map; Arranging the grid motion vectors of the moving entity in each image to obtain a trajectory sequence; Analyze the average change of the grid motion vectors in the trajectory sequence to obtain the predicted grid motion vector of the moving entity after the current moment, and perform curve fitting on the trajectory sequence to obtain the predicted grid motion direction of the moving entity after the current moment; Based on the predicted grid motion vector and the predicted grid motion direction, determine the predicted position of the moving entity; Compare the number of pixels of the moving entity at the predicted position with the average number of pixels after the current moment to obtain the moving entity density distribution at the predicted position, and obtain the dynamic density distribution according to the moving entity density distribution at the predicted position.

[0096] Among them, the feature descriptor corresponding to the moving entity refers to the high-dimensional feature vector extracted from the image of the moving entity (such as pedestrians, vehicles), which is used to uniquely identify the visual features (such as shape, texture, contour, etc.) of the entity. For example, by using an algorithm to extract the key point features of the entity and generate a feature descriptor, the same entity in different images can be matched through the similarity of the feature vectors.

[0097] Among them, the similarity between moving entities in adjacent images is based on the comparison result of the feature descriptors, which measures the probability that the moving entities in two adjacent frames of images are the same physical entity. Usually, the cosine similarity is used to calculate the included angle of the feature vectors. The smaller the included angle, the higher the similarity. When the similarity is greater than a preset threshold (such as 0.8), it is determined as the same entity.

[0098] Among them, determining multiple identical moving entities based on similarity is to classify the moving entities with high similarity as the same target through feature similarity matching in static images and historical motion images, so as to achieve cross-frame tracking. For example, if the similarity of the feature descriptors of a certain vehicle in three consecutive frames of images reaches 0.9, it is determined as the same vehicle to avoid misjudging different entities as the same target.

[0099] Among them, the position change of multiple identical moving entities in the grid map is the record of the grid coordinate migration of the same moving entity in consecutive images, such as moving from grid (10, 10) to (12, 10) and then to (13, 11), which reflects its motion trajectory and direction in space.

[0100] Among them, the grid motion vector is the grid coordinate difference of the same moving entity in two adjacent frames of images, which is used to quantify the amplitude and direction of the motion and reflect its displacement per unit time.

[0101] Among them, arranging the grid motion vectors to obtain a trajectory sequence is to arrange the motion vectors of the same entity in consecutive multiple frames in chronological order to form time-series motion trajectory data, which is used to analyze the regularity of the motion trend.

[0102] Among them, the average change of the grille motion vectors in the trajectory sequence is to calculate the average change rate of the magnitude and direction of the historical motion vectors, which is used to predict the motion amplitude at the next moment.

[0103] Among them, the predicted grille motion vector is the motion vector at the next moment deduced based on the historical average change amount, and its direction is consistent with the historical trend.

[0104] Among them, performing curve fitting on the trajectory sequence is to fit the trajectory curve of the motion vectors through mathematical methods (such as linear regression, polynomial fitting), smooth the noise points in the historical data, and extract the change trend of the motion direction (such as straight-line motion, turning trend, etc.).

[0105] Among them, the predicted grille motion direction is the tangent direction of the trajectory obtained by curve fitting, which reflects the future motion tendency of the moving entity. For example, if the fitting curve shows that the entity is moving northeast at an angle of 15°, the predicted direction is 15° east of northeast.

[0106] Among them, determining the predicted position based on the predicted vector and direction is to combine the magnitude and direction of the predicted motion vector and calculate the position of the moving entity at the next moment in the grid map. For example, if the current position is (20, 30) and the predicted vector is (3, 1), then the predicted position is (23, 31).

[0107] Among them, the density distribution of the moving entities at the predicted position is to count the ratio of the number of pixels of the predicted moving entities in this area to the average number of pixels at the future moment according to the grid coordinates of the predicted position, and quantify the future risk level of this position.

[0108] Exemplarily, in order to estimate the motion trajectories of each moving entity in the target area, the recognition results of historical continuous n-frame images are obtained, and the moving entities recognized in adjacent frames are matched; there is computational redundancy when the moving entities in one image are respectively matched with all the moving entities in the adjacent image. Since the motion of people and vehicles is relatively slow and the motion range in adjacent images is small, the matching range of the next frame image is reduced to a preset multiple range around the position in the previous frame. That is, if it is recognized that there is a person within 10*20 pixels at a certain position in the i-th frame, then in the (i + 1)-th frame, the recognition range is: with the entity at the same position as in the i-th frame as the core, and the range radius is 100*200 pixels. Thus, the main process of matching is: obtaining the recognized entities in each image, using an algorithm to extract feature points, generating feature descriptors, calculating the similarity between the feature descriptors of any two entities recognized in the two images, and considering the entities with the closest similarity as the same entity. Among them, the feature descriptor is a high-dimensional vector, and the similarity between the two feature descriptors is calculated using cosine similarity. Calculate the motion vectors of the same moving entity in two consecutive frames, which is mainly the difference between the grid positions of the latter frame and the previous frame.

[0109] Exemplarily, the grid motion vector of moving entity j at the i-th frame can be expressed as: where, represents the grid motion vector of moving entity j at the i-th frame; and respectively represent the grid coordinates of moving entity j at the i-th frame and the (i - 1)-th frame.

[0110] Among them, the motion vector is calculated by the difference in grid coordinates between two adjacent frames. The magnitude of the vector reflects the speed (e.g., 1 grid / frame = 5 m / s, assuming a frame rate of 5 frames / second), and the direction reflects the motion trend.

[0111] Exemplarily, referring to Figure 3 , Figure 3 which is a schematic diagram of the trajectory sequence provided by an embodiment of the present invention. In Figure 3 , the grid motion vector at the i-th frame, the grid motion vector at the (i - 1)-th frame, the grid motion vector at the (i - 2)-th frame, and the grid motion vector at the (i - 3)-th frame are shown. After obtaining the grid motion vector sequence of consecutive n frames for each moving entity, it can be arranged in chronological order to form a trajectory sequence as shown in Figure 3 .

[0112] Since the change of the trajectory is a slow and continuous process, it can be considered that the change in the magnitude of the historical motion vector has a slow linear relationship; and the final motion vector changes continuously in one direction, and can be obtained by curve fitting. Then, calculate the average change in the magnitude of the historical consecutive n-frame motion vectors as the change in the final predicted motion vector to obtain the predicted grid motion vector of the moving entity after the current moment; and fit the historical n-frame trajectory curve to obtain the predicted grid motion direction of the moving entity after the current moment.

[0113] Among them, the representation of the average change of the grid motion vectors in the trajectory sequence can be: where, represents the average change of the grid motion vectors of moving entity j in the trajectory sequence (including n frames); n represents the number of frames of the historical images corresponding to the trajectory sequence; represents the motion vector of the j-th moving entity in the k-th frame image.

[0114] Then, the representation of the predicted grid motion vector of the moving entity after the current moment can be: where, Represents the predicted grid motion vector of the moving entity after the current moment (i.e., the (i + 1)-th frame).

[0115] Exemplarily, refer to Figure 4 , Figure 4 which is a schematic diagram of curve fitting and predicted position provided by an embodiment of the present invention. Figure 4 In it, the curve is obtained by curve fitting the trajectory sequence. As Figure 4 shown, by curve fitting the trajectory sequence, the predicted grid motion direction of the moving entity after the previous moment is obtained; using the magnitude is intercepted on the curve, and the end point of the vector direction falls on the curve to obtain the predicted position of the moving entity.

[0116] Then, the predicted positions of each moving entity in the next frame are calculated respectively, the predicted positions of the moving entities in the next frame are converted into pixels in the latest image, and the dynamic density distribution is calculated, denoted as .

[0117] In this embodiment, through the feature descriptor and similarity analysis, the problem of identity confusion of moving entities in consecutive images (such as distinguishing similar vehicles) is effectively solved, ensuring the continuity and reliability of the trajectory sequence, and avoiding trajectory breakage or incorrect prediction caused by incorrect matching. By the average change of historical motion vectors and curve fitting, the speed and direction laws of moving entities (such as uniform motion, acceleration, turning) are captured, and their future positions are predicted in advance. For example, it can be predicted that a vehicle 100 meters away will enter the landing area in 5 seconds, leaving time for obstacle avoidance; the mathematical fitting of the trajectory sequence can smooth accidental jitters (such as a pedestrian staying briefly), refining the true motion trend and improving prediction stability. The dynamic density distribution extends the risk analysis from the current moment to the future moment, combining the immediate risk of static density to form a two-dimensional evaluation system covering the present to the future. For example: static density identifies sudden stationary obstacles (such as a temporary stop); dynamic density warns of potential moving risks (such as an approaching vehicle), achieving double insurance. Based on the dynamic density prediction, the UAV can adjust its path in advance (such as flying around the predicted risk area), avoiding sudden stops or flight path oscillations caused by traditional immediate obstacle avoidance, improving flight smoothness; at the same time, reducing the risk of hindsight caused by relying on a single-frame image, such as avoiding landing at the landing point where a vehicle is about to arrive, reducing the collision probability.

[0118] In an embodiment of the present application, the analysis results of fusing the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point include: Obtain the speed of the moving entity at the current moment; Determine the density attention weight according to the speed; Based on the density attention weight, perform a weighted sum of the static density distribution and the dynamic density distribution to obtain the motion entity density distribution corresponding to each of the candidate landing points.

[0119] Among them, because when a moving object has a high speed, due to inertia, it is difficult to change the trend of motion. Therefore, only when the final speed of the vehicles and pedestrians on the ground is relatively slow, it is possible to make a sudden turn, resulting in the distribution trajectory not following the historical change distribution. Therefore, when the speed of the moving object on the ground is smaller, the density distribution predicted according to the trend may not be accurate enough, and it may be necessary to pay more attention to the current static density distribution data; on the contrary, when the speed of the moving object on the ground is larger, it is necessary to pay more attention to the predicted dynamic density distribution data.

[0120] Among them, obtaining the speed of the motion entity at the current moment means calculating its instantaneous motion speed through the position change of the motion entity in the grid map. The specific method is as follows: Using the historical continuous multi-frame grid motion vectors (such as the coordinate difference between two adjacent frames), combined with the time interval of image acquisition (such as 5 frames per second, with an interval of 0.2 seconds), calculate the displacement per unit time to obtain the magnitude and direction of the speed. For example, if a vehicle moves 2 grids (each grid is 1 meter) between two frames and the time interval is 0.2 seconds, then the speed is 10 m / s.

[0121] Among them, the density attention weight is a parameter used to balance the importance of the static density distribution and the dynamic density distribution, and its value range is usually [0, 1]. The weight value is dynamically adjusted according to the speed of the motion entity, reflecting the degree of emphasis of the system on the current immediate risk and the future predicted risk.

[0122] Among them, determining the density attention weight according to the speed is to establish a mapping relationship between the speed and the weight: when the speed of the motion entity is relatively high (such as a vehicle driving at high speed), its motion trajectory has strong inertia and predictability. Therefore, the credibility of the dynamic density distribution is higher, and the static density attention weight is reduced (such as the weight η = 0.3, and the dynamic weight 1 - η = 0.7); when the speed of the motion entity is relatively low or stationary (such as a pedestrian standing still, a vehicle temporarily parking), its trajectory is prone to mutation (such as a sudden turn). At this time, the static density distribution can better reflect the immediate risk, and the static weight is increased (such as η = 0.8, and the dynamic weight is 0.2).

[0123] Among them, the weighted sum is to perform a linear combination of the static density distribution and the dynamic density distribution through the density attention weight. The lower the value, the higher the safety of the candidate landing point.

[0124] Exemplarily, calculate the average of the speeds of the objects such as people and vehicles recognized on the ground in the current i-th frame and the historical (i - 1)-th frame, and perform normalization to obtain the density attention weight, which can be expressed as: Among them, represents normalization, represents the density attention weight; represents the velocity of the moving entity at the i-th frame (i.e., the current moment).

[0125] Based on the density attention weight, a weighted sum of the static density distribution and the dynamic density distribution is performed to obtain the moving entity density distribution corresponding to each candidate landing point. The representation of the moving entity density distribution can be: Among them, represents the moving entity density distribution; represents the static density distribution; represents the dynamic density distribution.

[0126] In this embodiment, through the velocity adaptive weight mechanism, the system can intelligently switch the focus of risk assessment according to the behavior characteristics of moving entities: for high-speed moving entities (such as moving vehicles), rely on the dynamic density to predict their trajectories to avoid collision risks caused by the lag of static data; for low-speed moving entities (such as pedestrians), focus on the static density to capture the immediate position to avoid misjudgment caused by dynamic prediction errors (such as sudden turning). The weight determination process is based on simple velocity threshold judgment (such as setting high-speed / low-speed critical values) or linear mapping (such as the greater the velocity, the smaller η), without complex algorithms and with high computational efficiency. Compared with global optimization or deep learning models, this method can run in real time in the UAV embedded system to meet the millisecond-level decision requirements in the landing stage.

[0127] In an embodiment of the present application, the optimizing the first delivery path according to the optimal landing point to obtain a second delivery path, so that the UAV lands along the second delivery path and completes the delivery, includes: Obtaining the position of the UAV at the current moment; Generating a second delivery path according to the position of the UAV at the current moment and the position of the optimal landing point; When the UAV lands along the second delivery path, updating the optimal landing point until the delivery is completed.

[0128] Exemplarily, when the drone enters the landing area range, the density distribution of moving entities is obtained using the above steps to obtain the target landing point. There is a landing point after each frame of the image ends; when a frame of the image ends, the landing point is dynamically updated. If the grid position difference of the landing point is small, the drone continues to fly towards the target landing point and emits a prompt sound during the flight to remind the nearby personnel to avoid. Generally, if the position difference of the landing point is less than 5 grid lengths, it is normal, otherwise the landing process is stopped, and the drone hovers in the air for 3 s and continuously analyzes whether the position of the landing grid point is stable. If it is stable, the landing process is continued, otherwise the landing is stopped and it continues to hover in the air for 3 s. Until finally the drone lands on the ground and waits for the user to pick up the goods.

[0129] In this embodiment, the drone can cope with sudden risks during the landing process (such as pedestrians suddenly entering the original landing point and vehicles temporarily parking). For example, when a dynamic obstacle appears at the originally planned landing point, the system immediately re-selects a safe landing point and generates a new path to avoid collision accidents caused by fixed routes. Compared with the traditional fixed-path scheme, the safety risk is reduced. In the traditional scheme, if the drone encounters an obstacle, it needs to hover and wait or return to the starting point to re-plan. However, in this scheme, through real-time local path optimization, the flight path is directly adjusted within the landing area without global re-planning. For example, in the above scenario, it only takes 1 - 2 seconds for the drone to complete the path adjustment from discovering the obstacle, saving 5 - 10 seconds of invalid hovering time compared with the traditional scheme, and improving the single-trip delivery efficiency.

[0130] Figure 5 FIG. is a schematic structural diagram of a drone path optimization system for high-speed service area delivery provided by an embodiment of the present invention. This system can be applied to Figure 1 the shown implementation environment. This system can also be applicable to other exemplary implementation environments and be specifically configured in other devices. This embodiment does not limit the implementation environment applicable to this system.

[0131] As Figure 5 shown, the exemplary drone path optimization system for high-speed service area delivery includes: An information acquisition module 501, configured to acquire the drone delivery order information and the initial position of the drone within the target high-speed service area; A trajectory recognition module 502, configured to perform path planning according to the drone delivery order information and the initial position of the drone to determine the first delivery path, and when the drone arrives at a preset landing flight area along the first delivery path, recognize the movement trajectories of moving entities within the landing flight area, where the landing flight area includes a plurality of candidate landing points; The density distribution analysis module 503 is configured to analyze the static density distribution and the dynamic density distribution of the moving entity according to the movement trajectory, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each of the candidate landing points; The optimal landing point selection module 504 is configured to select an optimal landing point from the multiple candidate landing points according to the moving entity density distribution; The landing module 505 is configured to optimize the first delivery path according to the optimal landing point to obtain a second delivery path, so that the drone lands along the second delivery path and completes the delivery.

[0132] In this exemplary UAV path optimization system for high-speed service area delivery, planning the first delivery path based on the order coordinates and the initial position of the UAV can achieve long-distance and fast passage, and avoid redundant hovering in non-sensitive areas. When entering the preset landing flight area, the path is dynamically adjusted in combination with the real-time identified movement entity trajectory, avoiding obstacle avoidance or waiting caused by fixed routes and reducing the ineffective flight time in local areas. By real-time analyzing multiple candidate landing points in the landing area, the optimal landing point is directly selected near the user's drop-off location, avoiding time waste caused by user movement. At the same time, by fusing the static and dynamic distributions of moving entities, feasible landing points are predicted in advance, enabling the UAV to plan the optimal landing path during flight and shortening the hovering time after arrival. The real-time analysis of the moving entity density can provide basic data support for multi-UAV collaborative delivery. By sharing the density distribution of each area, the dispatching system can dynamically allocate orders to low-risk areas, avoiding multiple aircraft concentrating in high-person-flow / vehicle-flow areas and improving the overall delivery efficiency of the service area. Through the fusion analysis of the static density distribution and the dynamic density distribution, the limitations of a single data dimension are solved. The static density reflects the immediate distribution of moving entities in the landing area at the current moment, capturing sudden static obstacles. The dynamic density predicts the future distribution trend based on the historical trajectory of moving entities, predicting the movement direction of movable obstacles. The system emphasizes dynamic analysis in the high-speed moving object scenario and static analysis in the low-speed or static object scenario by balancing the proportions of the two through an adaptive weight mechanism, accurately identifying high-risk areas.

[0133] It should be noted that the UAV path optimization system for high-speed service area distribution provided in the above embodiments and the UAV path optimization method for high-speed service area distribution provided in the above embodiments belong to the same concept. The specific manners in which each module and unit perform operations have been described in detail in the method embodiments and will not be elaborated herein. In actual application, the UAV path optimization system for high-speed service area distribution provided in the above embodiments can, according to needs, allocate the above functions to different functional modules, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above. This is not limited herein either.

[0134] It should be noted that the above sequence of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0135] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area, characterized in that, The method includes: Obtaining the drone delivery order information and the initial position of the drone within the target highway service area; Performing path planning based on the drone delivery order information and the initial position of the drone to determine a first delivery path. When the drone reaches a preset landing flight area along the first delivery path, identifying the movement trajectories of moving entities within the landing flight area, where the landing flight area includes multiple candidate landing points; Analyzing the static density distribution and the dynamic density distribution of the moving entities according to the movement trajectories, and fusing the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each of the candidate landing points; Selecting an optimal landing point from the multiple candidate landing points according to the moving entity density distribution; Optimizing the first delivery path according to the optimal landing point to obtain a second delivery path, so that the drone lands along the second delivery path and completes the delivery.

2. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 1, wherein, The performing path planning based on the drone delivery order information and the initial position of the drone to determine a first delivery path includes: Obtaining the top-down map corresponding to the target highway service area; Rasterizing the top-down map to obtain a raster map, and establishing a three-dimensional coordinate system corresponding to the target highway service area, and corresponding each grid in the raster map to the three-dimensional coordinate system; Determining the user position according to the drone delivery order information, and converting the user position and the initial position of the drone into coordinates in the raster map; Performing path planning according to the coordinates in the raster map to obtain the first delivery path.

3. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 2, wherein After determining the first delivery path, it further includes: Taking the coordinates corresponding to the user position in the raster map as the center, and determining the landing flight area according to a preset landing flight radius; When the distance between the coordinates of the drone and the center is equal to the landing flight radius during the flight of the drone along the first delivery path, it is determined that the drone reaches the landing flight area.

4. The method for optimizing the path of a drone for high-speed service area distribution according to claim 3, wherein, The identifying the movement trajectories of moving entities within the landing flight area includes: Obtaining a static image of the moving entities within the landing flight area at the current moment and a historical movement image of the moving entities within the landing flight area before the current moment through a pre-configured image acquisition device; Performing image recognition on the static image and the historical movement image to obtain moving entities and ground markers; Mapping the image pixels corresponding to the moving entities and the ground markers to the raster map to obtain the raster coordinates corresponding to the moving entities and the ground markers, and the raster coordinates are used to indicate the movement trajectories of the moving entities.

5. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 4, wherein The analyzing the static density distribution and the dynamic density distribution of the moving entities according to the movement trajectories includes: Determining the number of pixels of the moving entities in each grid at the current moment according to the raster coordinates, and determining the static density distribution according to the number of pixels; Determine a plurality of identical moving entities from the static image and the historical motion image according to the grid coordinates, and determine the dynamic density distribution according to the position changes of the plurality of identical moving entities in the grid map.

6. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 5, wherein, The determining the static density distribution according to the number of pixels includes: Compare the number of pixels of the moving entities in each grid with the average number of pixels at the current moment to obtain the density distribution of the moving entities in each grid; Combine the density distributions of the moving entities in each grid to obtain the static density distribution.

7. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 5, wherein The determining a plurality of identical moving entities from the static image and the historical motion image according to the grid coordinates, and determining the dynamic density distribution according to the position changes of the plurality of identical moving entities in the grid map includes: Extract the feature descriptors corresponding to the moving entities according to the grid coordinates, and analyze the similarity between the moving entities in adjacent images according to the feature descriptors; Based on the similarity, determine a plurality of identical moving entities from the static image and the historical motion image; According to the position changes of the plurality of identical moving entities in the grid map, obtain the grid motion vectors of the moving entities in each image; Arrange the grid motion vectors of the moving entities in each image to obtain a trajectory sequence; Analyze the average change of the grid motion vectors in the trajectory sequence to obtain the predicted grid motion vector of the moving entity after the current moment, and perform curve fitting on the trajectory sequence to obtain the predicted grid motion direction of the moving entity after the current moment; Based on the predicted grid motion vector and the predicted grid motion direction, determine the predicted position of the moving entity; Compare the number of pixels of the moving entity at the predicted position with the average number of pixels after the current moment to obtain the density distribution of the moving entity at the predicted position, and obtain the dynamic density distribution according to the density distribution of the moving entity at the predicted position.

8. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 5, wherein, The fusing the analysis results of the static density distribution and the dynamic density distribution to obtain the density distribution of the moving entity corresponding to each candidate landing point includes: Obtain the speed of the moving entity at the current moment; Determine the density attention weight according to the speed; Based on the density attention weight, perform weighted summation on the static density distribution and the dynamic density distribution to obtain the density distribution of the moving entity corresponding to each candidate landing point.

9. The method for optimizing the path of an unmanned aerial vehicle for distribution in a high-speed service area according to claim 1, wherein, The optimizing the first delivery path according to the optimal landing point to obtain a second delivery path, so that the drone lands along the second delivery path and completes the delivery includes: Obtain the position of the drone at the current moment; Generate a second delivery path according to the position of the drone at the current moment and the position of the optimal landing point; When the drone lands along the second delivery path, update the optimal landing point until the delivery is completed.

10. An unmanned aerial vehicle path optimization system for high-speed service area distribution, characterized in that, The system includes: An information acquisition module for acquiring the drone delivery order information and the initial position of the drone in the target highway service area; A trajectory recognition module, which is used to perform path planning according to the UAV delivery order information and the initial position of the UAV, determine the first delivery path, and recognize the motion trajectory of the moving entity in the landing flight area when the UAV arrives at the preset landing flight area along the first delivery path, wherein the landing flight area includes a plurality of candidate landing points; A density distribution analysis module, which is used to analyze the static density distribution and dynamic density distribution of the moving entity according to the motion trajectory, and fuse the analysis results of the static density distribution and the dynamic density distribution to obtain the moving entity density distribution corresponding to each candidate landing point; An optimal landing point selection module, which is used to select the optimal landing point from the plurality of candidate landing points according to the moving entity density distribution; A landing module, which is used to optimize the first delivery path according to the optimal landing point to obtain a second delivery path, so that the UAV lands along the second delivery path and completes the delivery.

Images