A method for unmanned vehicle path planning based on relative positioning of regional images

Abstract

The invention discloses a method for planning the travel path of an unmanned vehicle based on relative positioning of regional images, which belongs to the technical field of unmanned vehicle control. The method comprises: obtaining comprehensive image data of a target area based on a drone configured by the unmanned vehicle, extracting the starting point position, the target point position and landform features of the unmanned vehicle from the comprehensive image data; establishing a DDQN network model, and generating a travel path of the unmanned vehicle from the starting point to the target point based on the starting point position, the target point position and the landform features; obtaining and processing local image data of the target area, obtaining the travel speed of the unmanned vehicle, obtaining the position coordinates of the unmanned vehicle in a Frenet coordinate system, setting navigation points on the travel path based on the position coordinates, optimizing the navigation points, and generating an optimized path of the unmanned vehicle based on the optimized navigation points. The accuracy of the path planning of the unmanned vehicle is improved by the invention.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned vehicle control, and in particular relates to an unmanned vehicle travel path planning method based on relative positioning of regional images. Background Art

[0002] The path planning technology of unmanned vehicles is one of the key technologies to achieve autonomous navigation of vehicles. With the development of intelligent transportation systems, the research on path planning technology of unmanned vehicles has become particularly important. With the development of autonomous driving technology, the positioning technology of unmanned vehicles in unknown environments has become the core of research. In the absence of GPS signals, technologies such as simultaneous localization and mapping (SLAM) technology and multi-sensor fusion solutions are used to achieve path planning and autonomous driving tasks for unmanned vehicles.

[0003] Traditional path planning methods, such as the Dijkstra algorithm, perform well in global path planning, but have shortcomings when dealing with the continuous maneuvers and obstacle avoidance of unmanned vehicles. These algorithms often ignore the dynamic characteristics of unmanned vehicles, resulting in the planned path being not smooth enough or too close to obstacles, and unable to avoid potential risks in time.

[0004] As an emerging technology, regional image relative positioning technology can provide accurate positioning information during the movement of unmanned vehicles. By analyzing the image information around the vehicle, this technology can accurately estimate the vehicle's position, which is crucial for the path planning of unmanned vehicles in complex environments. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a method for unmanned vehicle path planning based on relative positioning of regional images to solve the problems in the prior art.

[0006] In order to achieve the above-mentioned invention object, the present invention proposes a method for unmanned vehicle path planning based on relative positioning of regional images, comprising:

[0007] Step S1: The drone configured by the unmanned vehicle photographs the target area at a preset first height to obtain panoramic image data of the target area, and extracts the starting point position information of the unmanned vehicle, the target point position information and the geomorphic features of the target area from the panoramic image data;

[0008] Step S2: establishing a DDQN network model, wherein the DDQN network model generates a driving path of the unmanned vehicle from the starting point to the target point based on the starting point location information, the target point location information and the topographic features;

[0009] Step S3: the drone photographs a plurality of sub-areas in the target area at a preset second height to obtain local image data, and processes the local image data to obtain the driving speed of the unmanned vehicle;

[0010] Step S4: Establish a Frenet coordinate system, obtain the position coordinates of the unmanned vehicle in the Frenet coordinate system, set navigation points on the driving path based on the position coordinates and the driving speed, optimize the navigation points, and generate the final optimized path of the unmanned vehicle based on the optimized navigation points.

[0011] Furthermore, generating a driving path of the unmanned vehicle from a starting point to a target point comprises the following steps:

[0012] Establishing a simulation environment model of the target area based on the landform features, placing the unmanned vehicle in the simulation environment model for autonomous random exploration, obtaining exploration data of the unmanned vehicle, and storing the exploration data in an experience replay pool for storage;

[0013] Extract sample data from the experience replay pool, and input the sample data into the DDQN network model, wherein the DDQN network model includes a behavior network and an evaluation network, wherein the behavior network analyzes the sample data to generate a plurality of action data under a driving state, and defines an objective function of the evaluation network, wherein the objective function is to minimize the distance between the current position of the unmanned vehicle and the position of a target point, wherein each action data corresponds to the current position of an unmanned vehicle, and the sample data is input into the objective function, and a function value of each action data under the current driving state is calculated, and a difference between the function value and a minimum value of the objective function is obtained, and an evaluation value corresponding to a plurality of the action data under the driving state is obtained based on the difference, and the evaluation value is inversely proportional to the difference, and the action data with the highest evaluation value is taken as the optimal action under the driving state, and the optimal action under each driving state from the starting point position to the target point position is continuously obtained, and a driving path of the unmanned vehicle is finally generated based on all the optimal actions.

[0014] Furthermore, obtaining the position coordinates of the unmanned vehicle in the Frenet coordinate system includes the following steps:

[0015] A path center line of the driving path is defined as a data reference line, a plurality of path points are set on the driving path, and position coordinates at the path points are obtained, wherein the longitudinal coordinates in the position coordinates are the driving distance between two adjacent path points along the driving path of the unmanned vehicle, and the transverse coordinates in the position coordinates are the vertical distance from the unmanned vehicle to the data reference line, and the position coordinates are coordinates in the Frenet coordinate system.

[0016] Furthermore, setting the navigation point includes the following steps:

[0017] The shooting frequency of the local image data is obtained, the quotient of the driving speed of the unmanned vehicle in the Frenet coordinate system and the shooting frequency is used as a control interval, and a navigation point is set on the driving path based on the control interval between two path points.

[0018] Furthermore, optimizing the navigation point includes the following steps:

[0019] Identify a turning area in the driving path, determine whether an area where each navigation point on the driving path is located belongs to the turning area, and if so, reduce the driving speed to a first threshold range and increase the number of navigation points to a second threshold range;

[0020] If the location of the navigation point does not belong to the turning area, the area where multiple consecutive navigation points that do not belong to the turning area are located is defined as a straight area, the driving speed of the unmanned vehicle in the straight area is increased to a third threshold range, and the number of navigation points is reduced to within the second threshold range.

[0021] Further, identifying the turning area on the driving path includes the following steps:

[0022] The driving path between two navigation points is obtained, and the road curvature of the driving path is calculated. If the curvature is greater than a fourth threshold, it indicates that the navigation point is in a turning area.

[0023] Furthermore, obtaining the driving speed of the unmanned vehicle includes the following steps:

[0024] The time interval for shooting each frame of the local image data and the position change of the unmanned vehicle in the continuous local image data are obtained, and the quotient of the position change and the time interval is used as the formal speed of the unmanned vehicle.

[0025] Furthermore, configuring the drone also includes fixing the field of view orientation of the drone.

[0026] Furthermore, the exploration data includes state data, action data and reward data of the unmanned vehicle.

[0027] Furthermore, the motion data includes direction data, speed data and operation data of the unmanned vehicle.

[0028] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0029] By installing a drone on an unmanned vehicle, a comprehensive view of the target area is obtained, the location information of the unmanned vehicle's starting point and target point, as well as the topographical features of the target area, and the relative position information of the starting point and target point is determined in the comprehensive image data, the navigation needs of the unmanned vehicle can be intuitively understood; a simulation environment model is set up, and the unmanned vehicle obtains the exploration data required to determine the path through autonomous random exploration in the simulation environment model, and the exploration data is used as samples to train the DDQN network model. The DDQN network model generates multiple possible actions through the behavior network, and evaluates the value of these actions through the evaluation network to obtain evaluation values, and selects the action with the highest evaluation value as the optimal action in the current state, thereby improving the adaptability and flexibility of path planning.

[0030] Since most actual roads are not straight, the Frenet coordinate system is established. The Frenet coordinate system can better adapt to curved paths. Setting navigation points under the Frenet coordinate system can also obtain more accurate vehicle location information. The navigation points in turning areas and straight areas are optimized to improve driving safety and efficiency. The final optimized path is generated based on the optimized navigation points to ensure that in complex environments without GPS or Beidou signals, unmanned vehicles can efficiently and safely reach the target point from the starting point. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flowchart of the steps of a method for planning a travel path of an unmanned vehicle based on relative positioning of regional images according to the present invention;

[0032] Figure 2 It is the Frenet coordinate system diagram of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0034] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0035] like Figure 1 As shown, a method for unmanned vehicle travel path planning based on relative positioning of regional images includes:

[0036] S1: The drone configured for the unmanned vehicle photographs the target area at a preset first height to obtain panoramic image data of the target area, and extracts the starting point position information of the unmanned vehicle, the target point position information and the geomorphic features of the target area from the panoramic image data.

[0037] Specifically, in areas without GPS or Beidou signals, an endurance drone is equipped on the unmanned vehicle. The drone usually carries a short-focal-length high-definition camera to capture images of specific areas. The direction of the drone's field of view remains unchanged during the entire shooting process. Assuming the field of view is facing south, this means that no matter how the drone moves, its camera always points to the south, that is, the angle for shooting the unmanned vehicle remains unchanged.

[0038] The target area in the present invention is usually an unobstructed geographical area such as a desert, and the area range is assumed to be 50 kilometers × 50 kilometers. The altitude at which the drone is first launched must ensure that the entire target area can be clearly captured, as well as the location information of the starting point and target point of the unmanned vehicle. It is assumed that the first altitude of the drone is 60 meters.

[0039] S2: Establish a DDQN network model. The DDQN network model generates a driving path for the unmanned vehicle from the starting point to the target point based on the starting point location information, the target point location information and the terrain features.

[0040] Specifically, target detection is performed on the captured panoramic image data based on a real-time target detection algorithm to determine the relative position information between the starting point and the target point of the unmanned vehicle in the comprehensive image data. The unmanned vehicle uses sensors (such as lidar and cameras) to perceive the surrounding environment, including terrain features and obstacles. The dual deep Q network (DDQN network model) usually includes an input layer, a hidden layer and an output layer. The input layer receives state information, which includes the relative position information of the unmanned vehicle, sensor readings, etc. The hidden layer processes this information, and the output layer gives an evaluation value of the possible action information corresponding to each state information, and selects the action with the highest evaluation value as the optimal action in this state. By continuously selecting the optimal action, the unmanned vehicle moves from the starting point to the target point to generate a driving path. Due to the undulating terrain and steep slopes in target areas such as deserts,

[0041] Traditional algorithms may not be able to find an effective path in areas where GPS signals are missing, resulting in being stuck in a local optimal solution and being unable to reach the destination. However, the DDQN algorithm model can use past experience and current sensor data to continue to provide navigation instructions for unmanned vehicles. At the same time, the DDQN model trained through simulation enables unmanned vehicles to navigate autonomously in complex terrain in the real world.

[0042] S3: The UAV photographs multiple sub-areas in the target area at a preset second height to obtain local image data, and processes the local image data to obtain the driving speed of the unmanned vehicle.

[0043] Specifically, the drone is lowered from a first height to a second height. Assuming the second height is 10 meters, the drone can capture the overall posture of the unmanned vehicle at the second height. The captured local image data is processed based on image recognition technology and computer vision algorithms to extract the driving speed of the unmanned vehicle at multiple locations.

[0044] S4: Establish a Frenet coordinate system, obtain the position coordinates of the unmanned vehicle in the Frenet coordinate system, set navigation points on the driving path based on the position coordinates and the driving speed, optimize the navigation points, and generate the final optimized path of the unmanned vehicle based on the optimized navigation points.

[0045] Specifically, since most actual roads are not straight, a Frenet coordinate system is established. The Frenet coordinate system is a coordinate system commonly used in geometry and physics, especially in trajectory planning and robot control, which can more intuitively represent the position of the vehicle on a curved road. Setting navigation points in the Frenet coordinate system can also obtain more accurate vehicle position information, and then dynamically adjust the navigation points according to the real-time speed status of the unmanned vehicle and environmental changes, and optimize the navigation points. The optimization aspects include: reducing the driving speed at turns and increasing the number of navigation points. The number of navigation points should not be too many. Too many navigation points will increase the calculation burden and affect the speed and real-time performance of path planning. When it is determined that it is a straight line, the number of navigation points can be reduced to improve the driving efficiency of the unmanned vehicle.

[0046] As a preferred technical solution of the present invention, generating a driving path of an unmanned vehicle from a starting point to a target point includes the following steps:

[0047] A simulation environment model of the target area is established based on the topographic features, and the unmanned vehicle is placed in the simulation environment model for autonomous random exploration to obtain the exploration data of the unmanned vehicle, which is then stored in the experience replay pool for storage.

[0048] Sample data is extracted from the experience replay pool and input into the DDQN network model. The DDQN network model includes a behavior network and an evaluation network. The behavior network analyzes the sample data to generate multiple action data under the driving state, and defines the objective function of the evaluation network. The objective function is to minimize the distance between the current position of the unmanned vehicle and the position of the target point. Each action data corresponds to the current position of an unmanned vehicle. The sample data is input into the objective function, and the function value of each action data under the current driving state is calculated, and the difference between the function value and the minimum value of the objective function is obtained. Based on the difference, the evaluation value corresponding to the multiple action data under the driving state is obtained, and the evaluation value is inversely proportional to the difference. The action data with the highest evaluation value is taken as the optimal action under the driving state, and the optimal action under each driving state from the starting point to the target point is continuously obtained. Based on all the optimal actions, the driving path of the unmanned vehicle is finally generated.

[0049] Specifically, use simulation software to build a simulation environment model of the target area. This model should include key elements such as unmanned vehicles, obstacles, roads, etc. This simulation environment model allows us to test and optimize the path planning algorithm under safe and risk-free conditions, and can quickly iterate and adjust parameters.

[0050] The unmanned vehicle is placed in the simulation environment model and allowed to conduct autonomous random exploration to obtain exploration data. The exploration data includes state data, action data and reward data. The state data includes the position of the unmanned vehicle in the current state and the position in the next state. The action data includes the direction data, speed data and operation data of the unmanned vehicle. A reward function is set, and the speed change and angle change taken by the unmanned vehicle in the two positions are input into the reward function to obtain reward data. The reward data reflects the reward index of the two action changes. The higher the reward index, the more the action change and path planning match. Autonomous random exploration helps the unmanned vehicle learn various possible paths and environmental reactions, and the data generated during the exploration process is stored in the experience replay pool. Samples are randomly extracted from the experience replay pool. These samples will be used to train the DDQN network model. The DDQN network model generates multiple possible actions through the behavior network. In the control system of the unmanned vehicle, the objective function of the evaluation network is defined to quantify the pros and cons of each possible action in order to select the optimal action to achieve a specific goal. The objective function is defined as minimizing the distance between the current position of the unmanned vehicle and the target point. Each action data corresponds to a current position of the unmanned vehicle. In the path planning of unmanned vehicles, each possible action (such as going straight, turning left, turning right, etc.) will move the vehicle to a new position, and the difference between the objective function value of each action and the minimum value of the objective function (the minimum distance under ideal conditions) is calculated. This difference reflects the deviation of each action from the optimal path, and based on the difference, an evaluation value for each action is generated. The evaluation value is inversely proportional to the difference, which means that the smaller the difference, the higher the evaluation value, indicating that the action is closer to the ideal state. The evaluation network evaluates the value of these actions to obtain the evaluation value, and selects the action with the highest evaluation value as the optimal action in the current state. This method ensures that the unmanned vehicle can choose the best path in each state, thereby improving the performance and safety of path planning.

[0051] Obtaining the position coordinates of the unmanned vehicle in the Frenet coordinate system includes the following steps:

[0052] The center line of the driving path is defined as the data reference line, and multiple path points are set on the driving path to obtain the position coordinates of the path points. The longitudinal coordinates in the position coordinates are the driving distance between two adjacent path points along the driving path of the unmanned vehicle, and the transverse coordinates in the position coordinates are the vertical distance from the unmanned vehicle to the data reference line. The position coordinates are coordinates in the Frenet coordinate system.

[0053] Specifically, the shooting frequency of the UAV to obtain local image data is determined, that is, the number of times the image is taken per unit time. According to the driving path of the unmanned vehicle, a series of path points are pre-set. These path points define the ideal driving trajectory of the unmanned vehicle, such as Figure 2As shown, a Frenet coordinate system is established, and the center line d=0 of the driving path is defined as the data reference line. The ordinate of the position coordinate is the driving distance between two adjacent path points along the driving path. For example, d=1 indicates that the driving distance between two adjacent path points is 1m. The abscissa is the vertical distance from the unmanned vehicle to the data reference line. For example, s=1 indicates that the vertical distance from the unmanned vehicle to the data reference line is 1m.

[0054] Setting a navigation point involves the following steps:

[0055] The shooting frequency of the local image data is obtained, the quotient of the driving speed of the unmanned vehicle in the Frenet coordinate system and the shooting frequency is used as the control interval, and the navigation point is set on the driving path based on the control interval between two path points.

[0056] Specifically, setting navigation points in the Frenet coordinate system can also provide more accurate vehicle position information than the Cartesian coordinate system, and is more in line with actual road conditions. The control intervals between navigation points are set based on the real-time driving speed, which means that path planning can respond quickly to speed changes, thereby improving the real-time performance of path planning.

[0057] Optimizing navigation points includes the following steps:

[0058] Identify the turning area in the driving path, and determine whether the area where each navigation point on the driving path is located belongs to the turning area. If yes, reduce the driving speed to a first threshold range and increase the number of navigation points to a second threshold range.

[0059] If the location of the navigation point does not belong to the turning area, the area where multiple consecutive navigation points that do not belong to the turning area are located is defined as a straight area, the driving speed of the unmanned vehicle in the straight area is increased to the third threshold range, and the number of navigation points is reduced to within the second threshold range.

[0060] Specifically, identifying turning areas and adjusting speed and navigation points, reducing speed in turning areas, reducing centrifugal force on unmanned vehicles when turning, reducing the risk of rollover or loss of control, and increasing the number of navigation points in turning areas can help to more accurately control the vehicle's driving path, avoid deviation from the predetermined trajectory, and reduce the possibility of collision. However, there should not be too many navigation points, otherwise the driving efficiency of the unmanned vehicle will be reduced. Increasing speed in straight areas can improve driving efficiency, reduce unnecessary deceleration, and thus shorten driving time.

[0061] The driving speed is reduced to a first threshold range, and the calculation of the maximum driving speed within the first threshold range includes the following steps: obtaining the maximum driving speed in the turning area based on a first formula, the first formula being: ,in is the maximum curvature value of the turning area, is the stiffness coefficient of the unmanned vehicle steering system, is the distance between the front and rear wheels, is the maximum angle that the front wheels of the unmanned vehicle can turn, is the mass of the unmanned vehicle, is the acceleration of the unmanned vehicle in the y-axis direction, is the distance from the center of the unmanned vehicle to the front wheel, is the distance from the center of the unmanned vehicle to the rear wheel, For the maximum driving speed, Formula 1 takes into account the physical limitations of the vehicle when turning, such as the stiffness of the steering system, the vehicle's wheelbase, mass, steering angle, etc. By ensuring that the vehicle does not exceed these physical limitations, the maximum safe speed when turning is determined, which helps prevent skidding or rollover accidents when turning at high speeds, thereby improving the safety and reliability of the vehicle.

[0062] Increasing the driving speed of the unmanned vehicle in the straight area to a third threshold range includes the following steps: obtaining the roughness and average slope of the road surface in the straight area, adjusting the driving speed in the straight area based on the second formula and the third formula to obtain a second driving speed and the third travel speed , the second formula is: ,in, is the original driving speed, The unevenness of the road surface, is the preset first value, and the third formula is: , is the average slope of the road surface, is a preset second value, and the third threshold range is between the second driving speed and the third driving speed. The second formula is used to reduce the original driving speed when the road surface roughness is greater than the first value, and to accelerate the original speed when the road surface roughness is less than the first value, that is, when the road surface is sufficiently flat. The third formula is the same as the second formula in principle, and when the average slope of the road surface is greater than the second value, the original driving speed is reduced, and when the average slope of the road surface is less than the second value, the original speed is accelerated.

[0063] Identifying a turning area on a driving path includes the following steps:

[0064] A driving path between two navigation points is obtained, and a road curvature of the driving path is calculated. If the curvature is greater than a fourth threshold, it indicates that the navigation point is in a turning area.

[0065] Specifically, the position coordinates of the navigation point in the Frenet coordinate system are converted into coordinate points in the Cartesian coordinate system, a polynomial (usually a 5th-order polynomial) is used to fit the navigation point, the polynomial coefficients are adjusted by an optimization algorithm (such as minimum Jerk optimization), the optimized polynomial coefficients are used to generate a smooth trajectory curve, a fourth threshold is set, and the calculated curvature is compared with the preset fourth threshold. If the curvature is greater than this threshold, it is considered that the current navigation point is in the turning area.

[0066] Obtaining the speed of the unmanned vehicle includes the following steps:

[0067] The time interval for capturing each frame of local image data and the position change of the unmanned vehicle in the continuous local image data are obtained, and the quotient of the position change and the time interval is used as the formal speed of the unmanned vehicle.

[0068] By installing a drone on an unmanned vehicle, a comprehensive view of the target area is obtained, the location information of the unmanned vehicle's starting point and target point, as well as the topographical features of the target area, and the relative position information of the starting point and target point is determined in the comprehensive image data, the navigation needs of the unmanned vehicle can be intuitively understood; a simulation environment model is set up, and the unmanned vehicle obtains the exploration data required to determine the path through autonomous random exploration in the simulation environment model, and the exploration data is used as samples to train the DDQN network model. The DDQN network model generates multiple possible actions through the behavior network, and evaluates the value of these actions through the evaluation network to obtain evaluation values, and selects the action with the highest evaluation value as the optimal action in the current state, thereby improving the adaptability and flexibility of path planning.

[0069] Since most actual roads are not straight, the Frenet coordinate system is established. The Frenet coordinate system can better adapt to curved paths. Setting navigation points under the Frenet coordinate system can also obtain more accurate vehicle location information. The navigation points in turning areas and straight areas are optimized to improve driving safety and efficiency. The final optimized path is generated based on the optimized navigation points to ensure that in complex environments without GPS or Beidou signals, unmanned vehicles can efficiently and safely reach the target point from the starting point.

[0070] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0071] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0072] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0073] The above embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

[0074] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for unmanned vehicle path planning based on relative positioning of regional images, characterized in that: The method comprises the following steps: Step S1: The drone configured by the unmanned vehicle photographs the target area at a preset first height to obtain panoramic image data of the target area, and extracts the starting point position information of the unmanned vehicle, the target point position information and the geomorphic features of the target area from the panoramic image data; Step S2: establishing a DDQN network model, wherein the DDQN network model generates a driving path of the unmanned vehicle from the starting point to the target point based on the starting point location information, the target point location information and the topographic features; Step S3: the drone photographs a plurality of sub-areas in the target area at a preset second height to obtain local image data, and processes the local image data to obtain the driving speed of the unmanned vehicle; Step S4: establishing a Frenet coordinate system, obtaining the position coordinates of the unmanned vehicle in the Frenet coordinate system, setting navigation points on the driving path based on the position coordinates and the driving speed, optimizing the navigation points, and generating a final optimized path for the unmanned vehicle based on the optimized navigation points; The optimization process of the navigation point includes the following steps: Identify the turning area in the driving path, determine whether the area where each navigation point on the driving path is located belongs to the turning area, and if so, reduce the driving speed to a first threshold range and increase the number of navigation points to a second threshold range; If the location of the navigation point does not belong to the turning area, the area where multiple consecutive navigation points that do not belong to the turning area are located is defined as a straight area, the driving speed of the unmanned vehicle in the straight area is increased to a third threshold range, and the number of navigation points is reduced to within the second threshold range; The driving speed is reduced to a first threshold range, and the calculation of the maximum driving speed within the first threshold range includes the following steps: obtaining the maximum driving speed in the turning area based on a first formula, the first formula being: ,in is the maximum curvature value of the turning area, is the stiffness coefficient of the unmanned vehicle steering system, is the distance between the front and rear wheels, is the maximum angle that the front wheels of the unmanned vehicle can turn, is the mass of the unmanned vehicle, is the acceleration of the unmanned vehicle in the y-axis direction, is the distance from the center of the unmanned vehicle to the front wheel, is the distance from the center of the unmanned vehicle to the rear wheel, is the maximum travel speed; Increasing the driving speed of the unmanned vehicle in the straight area to a third threshold range includes the following steps: obtaining the roughness and average slope of the road surface in the straight area, adjusting the driving speed in the straight area based on the second formula and the third formula to obtain a second driving speed and the third travel speed , the second formula is: ,in, is the original driving speed, The unevenness of the road surface, is the preset first value, and the third formula is: , is the average slope of the road surface, is a preset second value, and the third threshold range is between the second driving speed and the third driving speed.

2. The method according to claim 1, characterized in that Generating the driving path of the unmanned vehicle from the starting point to the target point includes the following steps: Establishing a simulation environment model of the target area based on the landform features, placing the unmanned vehicle in the simulation environment model for autonomous random exploration, obtaining exploration data of the unmanned vehicle, and storing the exploration data in an experience replay pool for storage; Sample data is extracted from the experience replay pool and input into the DDQN network model, wherein the DDQN network model includes a behavior network and an evaluation network. The behavior network analyzes the sample data to generate a plurality of action data under a driving state. The values ​​of these actions are evaluated through the evaluation network to obtain evaluation values. The action data with the highest evaluation value is used as the optimal action under the driving state. The optimal action under each driving state from the starting point to the target point is continuously obtained, and the driving path of the unmanned vehicle is finally generated based on all the optimal actions.

3. The method according to claim 1, characterized in that Obtaining the position coordinates of the unmanned vehicle in the Frenet coordinate system includes the following steps: A path center line of the driving path is defined as a data reference line, a plurality of path points are set on the driving path, and position coordinates at the path points are obtained, wherein the longitudinal coordinates in the position coordinates are the driving distance between two adjacent path points along the driving path of the unmanned vehicle, and the transverse coordinates in the position coordinates are the vertical distance from the unmanned vehicle to the data reference line, and the position coordinates are coordinates in the Frenet coordinate system.

4. The method according to claim 3, characterized in that Setting a navigation point involves the following steps: The shooting frequency of the local image data is obtained, the quotient of the driving speed of the unmanned vehicle in the Frenet coordinate system and the shooting frequency is used as a control interval, and a navigation point is set on the driving path based on the control interval between two path points.

5. The method according to claim 1, characterized in that Identifying a turning area on the driving path includes the following steps: The driving path between two navigation points is obtained, and the road curvature of the driving path is calculated. If the curvature is greater than a fourth threshold, it indicates that the navigation point is in a turning area.

6. The method according to claim 1, characterized in that Obtaining the speed of the unmanned vehicle includes the following steps: The time interval for shooting each frame of the local image data and the position change of the unmanned vehicle in the continuous local image data are obtained, and the quotient of the position change and the time interval is used as the driving speed of the unmanned vehicle.

7. The method according to claim 1, characterized in that The configuration of the drone also includes fixing the field of view orientation of the drone.

8. The method according to claim 2, characterized in that: The exploration data includes state data, action data and reward data of the unmanned vehicle.

9. The method according to claim 2, characterized in that: The motion data includes direction data, speed data and operation data of the unmanned vehicle.

Images