Training method and device based on path planning and program product

Abstract

The invention discloses a training method and device based on path planning and a program product, and the method comprises the steps: obtaining a training data set, each training sample in the training data set comprising continuous frame samples and label data of a sample target in each continuous frame sample; constructing an initial intelligent path planning model; and performing iterative training on the intelligent path planning model based on training samples in the training data set and a loss function including physical information loss until an intelligent path planning model meeting training iteration conditions is obtained.

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving technology, and in particular to a training method based on path planning, a path planning method, an infrared imaging device, a computing device, and a computer program product. Background Art

[0002] The advanced artificial intelligence path integration and logical optimization technology (AI-PILOT) algorithms currently integrated into in-vehicle thermal imaging products use multi-sensor fusion to detect obstacles in real time. They also employ methods such as traditional neural networks (CNNs), long short-term memory neural networks (LSTMs), and reinforcement learning (RL) to predict local behavioral conflicts and adjust the planned path. However, the data collected through this method (including inputs and outputs) does not take into account the physical information and environmental differences of the scene, making it prone to inaccurate predictions and deviations in local behavior along the planned path. This makes it difficult to adapt to in-vehicle real-time decision-making systems in increasingly complex environments. Summary of the Invention

[0003] In order to solve the existing technical problems, the present invention provides a training method based on path planning, a path planning method, an infrared imaging device, a computing device and a computer program product, which can improve the real-time performance of local behavior of path planning in complex environments and the accuracy of conflict prediction.

[0004] In a first aspect, a path planning-based training method is provided, comprising: obtaining a training data set, wherein each training sample in the training data set includes a continuous frame sample and label data of a sample target in each continuous frame sample; constructing an initial intelligent path planning model; and iteratively training the intelligent path planning model based on the training samples in the training data set and a loss function including physical information loss until an intelligent path planning model that meets training iteration conditions is obtained.

[0005] In a second aspect, a path planning method is provided, which includes: acquiring collected actual scene data; forming an input of a pre-trained intelligent path planning model based on the actual scene data, and outputting global path planning data, wherein the intelligent path planning model is trained based on the path planning-based training method provided in the first aspect.

[0006] In a third aspect, an infrared imaging device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the path planning-based training method provided in the first aspect of this application, and / or the path planning method provided in the second aspect.

[0007] In a fourth aspect, a computing device is provided, comprising a storage device and a processing device, wherein the storage device stores a computer program, and when the computer program is executed by the processing device, the processing device executes the path planning-based training method provided in the first aspect.

[0008] In a fifth aspect, a computer program product is provided, comprising a computer program, which, when executed, implements the path planning-based training method provided in the first aspect of this application, and / or the path planning method provided in the second aspect.

[0009] During the training process, this application calculates the physical information loss based on physical information, which can enable the intelligent path planning model in training to better learn the characteristics of physical information in complex scene environments. In this way, after the pre-trained intelligent path planning model is applied to the actual environment, it can handle complex interactive behaviors in complex environments. The pre-trained intelligent path planning model can realize fast, accurate and real-time dynamic decision-making of local behaviors, improve the real-time performance of local behaviors in path planning in complex environments and the accuracy of conflict prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 FIG. 1 is a diagram illustrating an application environment of a training method based on path planning in one embodiment;

[0011] Figure 2 A diagram illustrating an application environment of a path planning method in one embodiment;

[0012] Figure 3 is a flow chart of a training method based on path planning in one embodiment;

[0013] Figure 4 is a schematic diagram of a dynamic sample target in one embodiment;

[0014] Figure 5 is a flow chart of a training method based on path planning in one embodiment;

[0015] Figure 6 Schematic diagram of an example output of a trained intelligent path planning model based on a physical information neural network in one embodiment;

[0016] Figure 7 is a flow chart of a training method based on path planning in another embodiment;

[0017] Figure 8 is a schematic diagram of a training method and apparatus based on path planning in one embodiment;

[0018] Figure 9 A schematic diagram of a computing device according to an embodiment. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the scope of protection of the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] In the following description, reference is made to “some embodiments” which describe a subset of all possible embodiments, but it should be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0022] See Figure 1 , is an application environment diagram of a training method based on path planning in one embodiment. The application environment diagram includes a computing device 30 and an infrared imaging device 10. The training method based on path planning is applied to the computing device 30, which serves as a training platform and can be a device with computing capabilities, including but not limited to: a cluster formed by one or more computers, one or more server groups, etc. The computing device 30 is used to train the intelligent path planning model according to the training data set, obtain the trained intelligent path planning model, and store the trained intelligent path planning model in the infrared imaging device 10. The computing device 30 can transmit data to the infrared imaging device 10 via wired or wireless means, so that the model parameters in the infrared imaging device 10 can be updated in a timely manner.

[0023] like Figure 2 As shown, Figure 2: This is an application environment diagram of a path planning method in one embodiment; the application environment diagram includes an infrared imaging device 10, which can be installed in a vehicle device, wherein the vehicle device includes but is not limited to: a vehicle, an electric vehicle, a hybrid electric vehicle, a motorcycle, a bicycle, a personal mobile device, an airplane, a drone, a ship or a robot, etc. The infrared imaging device 10 includes an image acquisition device 12, a processor 13 and a memory 14. The image acquisition device 12 is used to acquire actual scene data, such as infrared image data of an actual scene, and the processor 13 outputs global path planning data based on the actual scene data acquired by the image acquisition device 12 using a pre-trained intelligent path planning model. The memory 14 is used to store programs and various types of data (such as parameters of a pre-trained intelligent path planning model) corresponding to the training method based on path planning.

[0024] The image acquisition device 12 may be a combination of one or more sensors. The image acquisition device 12 may be a monocular vision sensor or a multi-camera vision sensor. For example, the image acquisition device 12 may be a combination of one or more of a thermal imaging sensor, a visible light image sensor, a millimeter wave sensor, a lidar sensor, an infrared thermal imaging sensor, a depth sensor, a back-illuminated CMOS sensor, an electron multiplying CCD sensor, a scientific-grade CMOS sensor, an InGaAs sensor, a microchannel plate sensor, and a quantum dot sensor.

[0025] The processor 13 may be one or more processors. When there are multiple processors 13, the multiple processors may be integrated on one chip or independently set on each chip. The vehicle equipment may also include other sensor modules, including but not limited to environmental perception sensors and motion posture sensors. Environmental perception sensors include but are not limited to one or more combinations of the following sensors: brightness sensors, temperature sensors, haze sensors and other environmental sensors. Motion posture sensors include but are not limited to one or more combinations of the following: inertial sensors (Inertial Measurement Unit, IMU), speed sensors, acceleration sensors, gyroscope sensors, geomagnetic sensors, rotation vector sensors, steering wheel angle sensors, level sensors, tilt sensors, vibration sensors, displacement sensors and gravity sensors, etc.

[0026] The vehicle equipment may further include a display terminal for displaying images.

[0027] See also Figure 3 , is a flow chart of a path planning-based training method provided in one embodiment of the present application. A path planning-based training method is applied to a computing device, and the path planning-based training method includes the following steps:

[0028] S11. Obtain a training data set, where each training sample in the training data set includes continuous frame samples and label data of sample targets in each continuous frame sample.

[0029] In this embodiment, continuous frame samples include but are not limited to one or more of the following: continuous frame point cloud data, continuous frame infrared images, and continuous frame visible light images. The label data includes but is not limited to one of the following: static driving area labels of static sample targets, activity area labels of dynamic sample targets, target type labels of sample targets, temperature labels of sample targets, and preset optimal path labels. The sample targets include static sample targets and dynamic sample targets. For example, static sample targets are the ground, trees, vehicles parked on the road, traffic lights, static obstacles, etc. For example, dynamic sample targets are walking pedestrians, moving vehicles, dynamic obstacles, etc. Figure 4 As shown, Figure 4 is a schematic diagram of a dynamic sample target in one embodiment, Figure 4 The red box in the middle is a walking pedestrian. Since the training samples are continuous frame samples, and the continuous frame samples include static sample targets and / or dynamic sample targets, the training dataset includes static / dynamic continuous space model data.

[0030] S12. Construct an initial intelligent path planning model.

[0031] In this embodiment, the intelligent path planning model adopts a three-part hierarchical planning architecture in the Artificial Intelligence Path Integration and Logical Optimization Technology (AI-PILOT) algorithm. The intelligent path planning model includes a global path planning network layer, a local behavior decision network layer, and a motion planning network layer. The global path planning network layer is used for the overall planning of the path from the starting point to the end point. The local behavior decision network layer is used to process dynamic change instructions in a short period of time, such as lane environment lane changes, following vehicles, etc., with real-time and accuracy requirements. The motion planning network layer is responsible for planning motion indicators related to the motion posture of the vehicle in the local environment, such as controlling the motion speed according to the scene, etc. When executing the global path planning network layer, if the local behavior conditions are met, the local behavior decision network layer is executed, and if the motion control conditions are met, the motion planning network layer is executed.

[0032] S13. Based on the training samples in the training data set and the loss function including the physical information loss, the intelligent path planning model is iteratively trained until an intelligent path planning model that meets the training iteration conditions is obtained.

[0033] Physical information includes temperature field information, flow field information, thermal radiation information, and more. Since infrared imaging relies on an optical system focusing infrared radiation from an object or its reflection onto a detector, the scene environment includes temperature field information. Since the temperature field is also a flow field and dynamically changes, changes in the environment cause changes in the temperature field, and the flow of the flow field affects the temperature field. As the principle of infrared imaging indicates, infrared imaging is based on thermal radiation, and therefore, physical information also includes thermal radiation information.

[0034] During the training process, learning the physical information in the complex scene environment in the training data set and calculating the physical information loss based on the physical information can enable the intelligent path planning model to better learn the characteristics of the physical information in the complex scene environment. In this way, after the pre-trained intelligent path planning model is applied to the actual environment, it can handle complex interactive behaviors in complex environments. The pre-trained intelligent path planning model can realize fast, accurate and real-time dynamic decision-making of local behaviors.

[0035] In the above embodiment, during the training process, the physical information loss based on physical information is calculated, so that the intelligent path planning model in training can better learn the characteristics of physical information in complex scene environments. In this way, after the pre-trained intelligent path planning model is applied to the actual environment, it can handle complex interactive behaviors in complex environments. The pre-trained intelligent path planning model can realize fast, accurate and real-time dynamic decision-making of local behaviors, improve the real-time performance of local behaviors in path planning in complex environments and the accuracy of conflict prediction.

[0036] In some embodiments, the iterative training of the intelligent path planning model based on the training samples in the training data set and the loss function including the physical information loss until the intelligent path planning model that meets the training iteration conditions is obtained includes:

[0037] Obtaining training input samples from the training data set, and forming input of the intelligent path planning model in the current iteration based on the training input samples;

[0038] Using the loss function and the training input samples, the current total loss value in the current iteration is calculated. Based on the current total loss value, training input samples are continued to be obtained from the training data set for iterative training until an intelligent path planning model that meets the training iteration conditions is obtained.

[0039] In this embodiment, during the training process, the input data for each training sample includes continuous frame samples and label data of the sample targets in each continuous frame sample, including various data such as static / dynamic continuous space model data, global path labels, dynamic obstacle labels, traffic light labels, etc., and a pre-trained intelligent path planning model is formed through iterative training. The training iteration conditions include but are not limited to: the number of iterations is greater than a preset number, the current total loss value is less than a preset loss value, etc. If it is determined based on the current loss value that the current conditions do not meet the training iteration conditions, training input samples are continuously obtained from the training data set for iterative training until an intelligent path planning model that meets the training iteration conditions is obtained.

[0040] In the above embodiment, during the training process, the physical information loss in each iteration is calculated, and the current total loss value is calculated. Based on the current total loss value, it is determined whether to continue training the intelligent path planning model. During the training process, the intelligent path planning model under training can better learn the characteristics of physical information in complex scene environments. In this way, after the pre-trained intelligent path planning model is applied to the actual environment, it can handle complex interactive behaviors in complex environments. The pre-trained intelligent path planning model can achieve fast, accurate and real-time dynamic decision-making of local behaviors, thereby improving the real-time performance of local behaviors in path planning in complex environments and the accuracy of conflict prediction.

[0041] In some embodiments, the physical information loss includes at least one of a thermal conduction residual loss based on a temperature field, a fluid flow residual loss based on a flow field, and a thermal radiation residual loss based on a thermal radiation model.

[0042] In this embodiment, the physical information loss can be one of the following: thermal conduction residual loss, fluid flow residual loss, and thermal radiation residual loss, or a loss calculated based on at least two of the aforementioned losses. To better accurately predict the local behavior of a vehicle on a road in real time, a deep learning intelligent optimization algorithm with a physical information neural network is used to train an intelligent path planning model and generate optimal decision parameters. When optimizing the local behavior decision of vehicle thermal imaging based on the physical information of thermal imaging and road environment information, it is necessary to integrate the laws of thermodynamics with multi-sensor fusion perception data to ensure that the planned path avoids high-temperature hazardous areas (areas with significant local behavior changes) while meeting energy allocation efficiency. Therefore, thermal conduction residual loss based on the temperature field, fluid flow residual loss based on the flow field, and thermal radiation residual loss based on the thermal radiation model can be calculated. This allows the intelligent path planning model to better learn the characteristics of physical information in complex scene environments during training, such as the information characteristics of the temperature field, the information characteristics of the flow field, the information characteristics of thermal radiation, and other physical information characteristics. By applying the above-mentioned physical information equation, physical information loss in the physical information neural network can be formed to influence the intelligent path planning model of local behavior decision-making in different extreme environments, so as to accurately predict conflicts in real time and generate response feedback.

[0043] In the above embodiment, by learning a variety of different physical information features in different complex scene environments in the training data set, the real-time performance of the vehicle's path planning local behavior in complex environments and the accuracy of conflict prediction can be improved. It also greatly enhances the adaptability and responsiveness of the real-time decision-making system in different complex environments, especially in extreme environments with complex physical information.

[0044] In some embodiments, calculating the current total loss value in the current iteration using the loss function and the training input sample includes:

[0045] Outputting current sample identification data corresponding to the training input sample through the intelligent path planning model in the current iteration;

[0046] Calculating a first heat change value of a current sample target in the current sample identification data based on the current sample identification data;

[0047] Calculating a second heat change value of the current sample target based on a temperature label corresponding to the current sample target in the training input sample;

[0048] The heat conduction residual loss of the current sample target is calculated based on the first heat change value and the second heat change value.

[0049] Optionally, the calculating, based on the current sample identification data, a first heat change value of the current sample target in the current sample identification data includes:

[0050] Obtaining the material density of the current sample target and the specific heat capacity parameter of the infrared imaging device;

[0051] Calculating the temperature change rate of the training input sample within a preset time;

[0052] Calculating the first heat change value based on the material density, the specific heat capacity parameter, and the temperature change rate;

[0053] Calculating the second heat change value of the current sample target based on the temperature label corresponding to the current sample target in the training input sample includes:

[0054] Obtaining the thermal conductivity of the infrared imaging device;

[0055] Calculating first gradient information of the training input sample;

[0056] Calculating the current heat of the current sample target at the current moment and the radiant energy indicating the heat change within a preset time period after the current moment;

[0057] The second heat change value is calculated based on the thermal conductivity, the first gradient information, the current heat amount, and the radiation energy.

[0058] In this embodiment, the current sample identification data refers to the data obtained from the training input sample identification in the current iteration. The current sample identification data includes, but is not limited to, the location area data of the current sample target, the target type of the current sample target, the temperature data of the current sample target, etc. Since the pixel values ​​of pixels in an infrared image are temperature values, after identifying the location area data of the current sample target, the temperature value of the center point or the average temperature value of the current sample target can be calculated based on the pixel values ​​within the location area data.

[0059] By establishing the heat conduction equation, the heat conduction residual loss is calculated. The heat conduction equation is:

[0060]

[0061] The left side of the equation is the first heat change value, and the right side of the equation is the second heat change value. The heat conduction residual loss of the current sample target can be the absolute value of the difference between the first heat change value and the second heat change value, or the square of the difference between the first heat change value and the second heat change value, etc. nIndicates the material density of the current sample target. When outputting the current sample identification data, the target type of the current sample target can be output. For each target type, the material density of each target type can be pre-configured. p,n is the specific heat capacity parameter. Training input sample T n Including continuous frame infrared images, therefore, T n Represents the temperature field of the training input sample, related to (x, y, z, t), Indicates T n The partial differential with respect to t is the temperature change rate of the training input sample within the preset time. K is the thermal conductivity, Represents the first-order gradient, where the first gradient information is Q s Indicates the current heat of the current sample target at the current moment, which can be calculated based on the temperature data of the location area where the current sample target is located in the current frame of the training input sample. l The radiant energy representing the heat change within a preset time period after the current moment can be calculated based on the temperature data of the location area of ​​the current sample target in consecutive frames of the training input sample. Various parameter data, such as specific heat capacity parameters and thermal conductivity, can be pre-stored in memory 14.

[0062] In this embodiment, the heat conduction residual loss of each current sample target is calculated according to the above formula to obtain the heat conduction residual loss of each current sample target, and then the heat conduction residual loss of the training input sample is calculated based on the heat conduction residual loss of each current sample target. For example, the heat conduction residual loss of the training input sample is equal to the cumulative sum of the heat conduction residual losses of each current sample target.

[0063] In the above embodiment, based on the current sample identification data, the first heat change value of the current sample target in the current sample identification data is calculated, based on the temperature label corresponding to the current sample target in the training input sample, the second heat change value of the current sample target is calculated, and based on the first heat change value and the second heat change value, the thermal conduction residual loss of the current sample target is calculated. The thermal conduction residual loss in the training process can be calculated more accurately, and the intelligent path planning model in training can more accurately learn the temperature field change data in the complex environment in the training data set, so that the adaptability and responsiveness of the real-time decision-making system in different complex environments can be greatly improved in the future.

[0064] In some embodiments, calculating the current total loss value in the current iteration using the loss function and the training input sample includes:

[0065] Acquire sensor data corresponding to the training input sample, and calculate flow velocity field data corresponding to the training input sample based on the sensor data corresponding to the training input sample;

[0066] Calculating the temperature change rate of the training input sample within a preset time and first gradient information of the training input sample;

[0067] Calculating a first flow field value corresponding to the training input sample based on the flow velocity field data, the temperature change rate, and the first gradient information;

[0068] Outputting current sample identification data corresponding to the training input sample through the intelligent path planning model in the current iteration, and obtaining the thermal diffusion coefficient of the current sample target in the current sample identification data;

[0069] Calculating a second flow field value corresponding to the training input sample based on the thermal diffusion coefficient and the first gradient information;

[0070] The fluid flow residual loss corresponding to the training input sample is calculated based on the first flow field value and the second flow field value.

[0071] In this embodiment, the temperature field is also a flow field and changes dynamically. That is, changes in the environment will cause changes in the temperature field. The flow of the flow field affects the temperature field. It is necessary to consider the dynamic changes in the temperature field caused by fluid convection. The Navier-Stokes equation and the heat conduction equation are combined to construct a multi-physics field model of the physical information neural network. The formula of the multi-physics field model is as follows:

[0072]

[0073] The left side of the above equation is the first flow field value, and the right side of the equation is the second flow field value. represents the first gradient information. The fluid flow residual loss can be the absolute value of the difference between the first flow field value and the second flow field value, or the square of the difference between the first flow field value and the second flow field value, etc. Where μ represents the flow velocity field data. The training input samples are acquired based on sensor data. Therefore, each training input sample corresponds to a set of sensor data. Indicates the temperature change rate. α represents the thermal diffusion coefficient. Each target type corresponds to a different thermal diffusion coefficient, which can be collected and configured in advance. is the first gradient information.

[0074] In the above embodiment, based on the flow velocity field data, thermal diffusion coefficient, temperature change rate and first gradient information, the first flow field value corresponding to the training input sample is calculated, and based on the thermal diffusion coefficient and the first gradient information, the second flow field value corresponding to the training input sample is calculated; based on the first flow field value and the second flow field value, the fluid flow residual loss corresponding to the training input sample is calculated, which can more accurately calculate the fluid flow residual loss during the training process, and enable the intelligent path planning model in training to more accurately learn the flow field change data in the complex environment in the training data set, so as to greatly improve the adaptability and responsiveness of the real-time decision-making system in different complex environments in the future.

[0075] In some embodiments, calculating the current total loss value in the current iteration using the loss function and the training input sample includes:

[0076] Acquire the radiation power of the infrared imaging device, the emissivity parameter of the infrared imaging device, the radiation area of ​​the infrared imaging device, and the ambient temperature data corresponding to the training input sample;

[0077] Calculating sample temperature data of the training input sample;

[0078] Calculating a radiation power estimation value corresponding to the training input sample based on the emissivity parameter, the radiation area, the ambient temperature data, and the sample temperature data;

[0079] Based on the radiation power and the radiation power estimation value, a thermal radiation residual loss corresponding to the training input sample is calculated.

[0080] In this embodiment, high-temperature objects can affect the thermal imaging product. Thermal imagers themselves generate heat, and excessive heat can affect data acquisition. To address the infrared radiation effects of high-temperature objects on the sensor, a thermal radiation model is established based on the Stefan-Boltzmann law to correct the temperature measurement value of the thermal imaging sensor. The radiation model is expressed as follows:

[0081] P=εσA(T n 4 ―T env 4 )

[0082] The left side of the equation, P, represents the radiated power, and the right side represents the estimated radiated power. The thermal radiation residual loss can be the absolute value of the difference between the radiated power and the estimated radiated power, or the square of the difference between the radiated power and the estimated radiated power, etc. A represents the radiated area. ε represents the emissivity parameter. T n Represents the training input sample. T envrepresents the ambient temperature data, which indicates the ambient temperature of the environment where the training input samples are collected. σ represents the Stefan-Boltzmann constant.

[0083] In the above embodiment, based on the emissivity parameter, radiation area, ambient temperature data and sample temperature data, the radiation power estimation value corresponding to the training input sample is calculated; based on the radiation power and the radiation power estimation value, the thermal radiation residual loss corresponding to the training input sample is calculated, which can more accurately calculate the thermal radiation residual loss in the training process, and enable the intelligent path planning model in training to more accurately learn the thermal radiation change data in the complex environment in the training data set, so as to greatly improve the adaptability and responsiveness of the real-time decision-making system in different complex environments in the future.

[0084] In some embodiments, the current total loss value further includes at least one of a data loss indicating a temperature loss and a path optimization constraint loss.

[0085] In this embodiment, the current total loss value can be calculated based on the physical information loss and the aforementioned data loss and / or path optimization constraint loss. The data loss corresponding to the current sample target is the loss calculated between the predicted temperature data of the current sample target and the corresponding temperature label. For training input samples, the data loss corresponding to the training input sample is the cumulative sum of the data losses corresponding to each current sample target. The path optimization constraint loss indicates the constraint loss during the path planning process.

[0086] In this embodiment, by applying the above-mentioned physical information equations and constraints, physical information loss in the physical information neural network can be formed to influence the intelligent path planning model of local behavior decision-making in different extreme environments, so as to accurately predict conflicts in real time and generate response feedback.

[0087] More specifically, the loss function of the physical information neural network model based on the AI-PILOT hierarchical planning architecture is divided into three parts: data loss, physical information loss, and path optimization constraint loss. The physical information loss R is the sum of the thermal conduction residual loss R1 based on the temperature field, the fluid flow residual loss R2 based on the flow field, and the thermal radiation residual loss R3 based on the thermal radiation model. The path optimization constraints include the temperature field obstacle avoidance constraints in path planning and the temperature constraints of the path itself in extreme environments. Therefore, the current total loss value is It can be:

[0088]

[0089] Where T p and T s are the data losses of the predicted temperature data of the current sample target and the corresponding temperature label, R is the physical information loss, and λJ is the path optimization constraint.

[0090] Optionally, the path optimization constraint loss includes an obstacle avoidance constraint loss based on a temperature field, and the calculating of the current total loss value in the current iteration using the loss function and the training input sample includes:

[0091] Outputting current sample identification data corresponding to the training input sample by the intelligent path planning model in the current iteration, obtaining a current sample obstacle target from the current sample identification data, and obtaining a current sample temperature of the current sample obstacle target at the current position at the current moment from the current sample identification data;

[0092] Get the sample temperature threshold;

[0093] The obstacle avoidance constraint loss is calculated according to the current sample temperature and the sample temperature threshold.

[0094] In this embodiment, the obstacle avoidance constraints are expressed as follows:

[0095] T n (p(t),t)≤T max

[0096] The left side of the equation represents the current sample temperature, and the right side of the equation represents the sample temperature threshold. The obstacle avoidance constraint loss can be the absolute value of the difference between the current sample temperature and the sample temperature threshold, or the square of the difference between the current sample temperature and the sample temperature threshold. Where p(t) represents the path position at the current time t, T n (p(t), t) represents the temperature at the path position p(t) at the current time t, that is, the current sample temperature. max Indicates the temperature threshold.

[0097] In the above embodiment, the obstacle avoidance constraint loss is calculated based on the current sample temperature and the sample temperature threshold, which can more accurately calculate the obstacle avoidance constraint loss during the training process, and enable the intelligent path planning model in training to more accurately learn the obstacle avoidance constraints in path planning in complex environments in the training data set, so as to greatly improve the adaptability and responsiveness of the real-time decision-making system in different complex environments.

[0098] In some embodiments, the path optimization constraint loss includes a temperature constraint loss under extreme conditions, and calculating the current total loss value in the current iteration using the loss function and the training input sample includes:

[0099] Outputting current sample identification data corresponding to the training input sample by the intelligent path planning model in the current iteration, and obtaining the quality of the current sample target in the current sample identification data;

[0100] Obtain the specific heat capacity parameters of infrared imaging equipment;

[0101] Calculating the temperature change value of the training input sample within a preset time;

[0102] Calculating a first temperature constraint value corresponding to the current sample target according to the mass of the current sample target, the specific heat capacity parameter, and the temperature change value;

[0103] Obtaining a preset first extreme ambient temperature value and a preset second extreme ambient temperature value, and calculating a second temperature constraint value based on the first extreme ambient temperature value and the second extreme ambient temperature value;

[0104] The self-temperature constraint loss corresponding to the current sample target is calculated based on the first temperature constraint value and the second temperature constraint value corresponding to the current sample target.

[0105] In this embodiment, in order to avoid and monitor the thermodynamic changes of the vehicle-mounted thermal imaging product itself during operation, the temperature constraints of the vehicle-mounted thermal imaging product itself in extreme environments need to be considered. The expression of the temperature constraints of the vehicle-mounted thermal imaging product in extreme environments is as follows:

[0106]

[0107] The left side of the equation represents the first temperature constraint value, and the right side of the equation represents the second temperature constraint value. The self-temperature constraint loss corresponding to the current sample target represents the absolute value of the difference between the first temperature constraint value and the second temperature constraint value, or the square of the absolute value of the difference between the first temperature constraint value and the second temperature constraint value. The self-temperature constraint loss corresponding to the training input sample is the cumulative sum of the self-temperature constraint losses corresponding to each current sample target. n The quality of the current sample target can be pre-collected and configured for various target types. p,n represents the specific heat capacity parameter, Indicates the temperature change value, which is T n The first derivative with respect to t. Q a represents the first extreme ambient temperature value, γ represents the weight coefficient corresponding to the first extreme ambient temperature value, Q c represents the second extreme ambient temperature value, and τ represents the weight coefficient corresponding to the second extreme ambient temperature value, wherein the first extreme ambient temperature value and the second extreme ambient temperature value can be pre-configured different temperatures, and the weight coefficient is also pre-configured.

[0108] In the above embodiment, based on the first temperature constraint value and the second temperature constraint value corresponding to the current sample target, the self-temperature constraint loss corresponding to the current sample target is calculated, which can more accurately calculate the self-temperature constraint loss corresponding to the current sample target, and enable the intelligent path planning model in training to more accurately learn the constraints under extreme environments in path planning in complex environments in the training data set, so as to influence the intelligent path planning model of local behavior decisions in different extreme environments, so as to accurately predict conflicts in real time and generate response feedback.

[0109] In some embodiments, obtaining a training data set includes:

[0110] Obtain sample sensor data and training samples;

[0111] Modeling and generating static simulation environment data based on the sample sensor data and the training samples;

[0112] Continuous trajectory data of the dynamic sample target is obtained from the training sample, and dynamic simulation environment data is modeled and generated based on the continuous trajectory data and the static simulation environment data. The dynamic simulation environment data includes dynamic simulation environment data of path data of the dynamic sample target that changes with time parameters.

[0113] In this embodiment, ground sample data can be extracted from the training samples, and then the position coordinates (x, y, z) of the terrain are determined in three-dimensional Euclidean space based on the ground sample data. The sample sensor data obtained by multi-sensor fusion (such as: laser radar (LiDAR) scanning, millimeter wave radar (mmWave Radar) scanning, camera, etc.) are directly modeled to obtain static simulation environment data to obtain a static simulation environment. Then, physical constraints such as the maximum speed, acceleration, minimum turning radius of the vehicle and other information are imposed in the static simulation environment. For dynamic sample targets (such as dynamic obstacles), the path is parameterized by time, and the dynamic coordinates (x(t), y(t), z(t)) are set to establish a continuous space model containing dynamic obstacles. The modeling method of dynamic obstacles needs to add the time factor, so the time variation is introduced into the position coordinates to form dynamic coordinates (x(t), y(t), z(t)), and each position is a variation with respect to time.

[0114] In the above embodiment, based on the collected actual scene data, real static simulation environment data is established, and then the path data of the dynamic sample target is added to the static simulation environment data. Time changes can be introduced to obtain dynamic simulation environment data. In this way, the training data set can include static and / or dynamic simulation environment data, which can provide more feature information for subsequent training.

[0115] like Figure 5 As shown, Figure 5 Flowchart of a path planning-based training method in one embodiment; the method is applied to infrared imaging equipment or vehicle equipment, and the method includes:

[0116] S51: Acquire collected actual scene data.

[0117] S52. Forming an input of a pre-trained intelligent path planning model based on the actual scenario data, and outputting global path planning data, wherein the intelligent path planning model is trained based on a training method for path planning.

[0118] In this embodiment, the actual scene data is scene data collected based on one or more sensor data. Figure 6 As shown, Figure 6 This is a schematic diagram showing an example of the output of a training intelligent path planning model based on a physical information neural network in one embodiment. Figure 6 It can be seen that the real-time performance of local behaviors in path planning and the accuracy of conflict prediction in complex environments can be improved in real time.

[0119] like Figure 7 As shown, Figure 7 FIG. 1 is a flow chart of a training method based on path planning in another embodiment, the flow chart including the following steps:

[0120] S71 . Obtain a training data set, where each training sample in the training data set includes continuous frame samples and label data of a sample target in each continuous frame sample.

[0121] S72. Construct an initial intelligent path planning model.

[0122] S73. Based on the training samples in the training data set and the loss function including physical information loss, the following loss values ​​are calculated: data loss, physical information loss, and path constraint loss.

[0123] S74. Calculate the current total loss value based on the above three losses. Based on the current total loss value, continue to obtain training input samples from the training data set for iterative training until an intelligent path planning model that meets the training iteration conditions is obtained.

[0124] S75: Acquire the collected actual scene data.

[0125] S76. Form an input of a pre-trained intelligent path planning model based on the actual scenario data, and output global path planning data, wherein the intelligent path planning model is trained based on a training method for path planning.

[0126] The steps in this embodiment have been described in one or more of the above embodiments and will not be repeated here.

[0127] Through the combination of one or more of the above embodiments, this application has at least the following features:

[0128] 1. Introducing physical information neural networks into conventional AI-PILOT intelligent path planning algorithms, and designing real-time decision-making systems for vehicle-mounted product series to improve the real-time performance of local path planning behaviors and the accuracy of conflict prediction in complex environments.

[0129] 2. The path planning method not only enables the thermal imaging products of outdoor vehicle systems to meet the original performance indicators, but also greatly improves its real-time decision-making system, and its adaptability and responsiveness in different complex environments, especially in extreme environments with complex physical information.

[0130] 3. The path planning method can also be applied to different models of outdoor vehicle-mounted series products and other related models. The built-in software algorithm does not require any external equipment or devices.

[0131] On the other hand, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the path planning-based training method described in any embodiment of the present application.

[0132] Among them, in the computer program product, an optional implementation form of the program module architecture of the computer program that implements each step of the path planning-based training method can be a path planning-based training method device. Figure 8 An embodiment of the present application provides a training method and device based on path planning, including: an acquisition module 71, used to obtain a training data set, each training sample in the training data set includes continuous frame samples and label data of the sample target in each continuous frame sample; a training module 72, used to construct an initial intelligent path planning model; the training module 72 is also used to iteratively train the intelligent path planning model based on the training samples in the training data set and a loss function including physical information loss, until an intelligent path planning model that meets the training iteration conditions is obtained.

[0133] Optionally, the training module 72 is further configured to obtain training input samples from the training data set, and form inputs of the intelligent path planning model in the current iteration based on the training input samples;

[0134] Using the loss function and the training input samples, the current total loss value in the current iteration is calculated. Based on the current total loss value, training input samples are continued to be obtained from the training data set for iterative training until an intelligent path planning model that meets the training iteration conditions is obtained.

[0135] Optionally, the physical information loss includes at least one of a thermal conduction residual loss based on a temperature field, a fluid flow residual loss based on a flow field, and a thermal radiation residual loss based on a thermal radiation model.

[0136] Optionally, the training module 72 is further configured to:

[0137] Outputting current sample identification data corresponding to the training input sample through the intelligent path planning model in the current iteration;

[0138] Calculating a first heat change value of a current sample target in the current sample identification data based on the current sample identification data;

[0139] Calculating a second heat change value of the current sample target based on a temperature label corresponding to the current sample target in the training input sample;

[0140] The heat conduction residual loss of the current sample target is calculated based on the first heat change value and the second heat change value.

[0141] Optionally, the training module 72 is further configured to:

[0142] Obtaining the material density of the current sample target and the specific heat capacity parameter of the infrared imaging device;

[0143] Calculating the temperature change rate of the training input sample within a preset time;

[0144] Calculating the first heat change value based on the material density, the specific heat capacity parameter, and the temperature change rate;

[0145] Calculating the second heat change value of the current sample target based on the temperature label corresponding to the current sample target in the training input sample includes:

[0146] Obtaining the thermal conductivity of the infrared imaging device;

[0147] Calculating first gradient information of the training input sample;

[0148] Calculating the current heat of the current sample target at the current moment and the radiant energy indicating the heat change within a preset time period after the current moment;

[0149] The second heat change value is calculated based on the thermal conductivity, the first gradient information, the current heat amount, and the radiation energy.

[0150] Optionally, the training module 72 is further configured to:

[0151] Acquire sensor data corresponding to the training input sample, and calculate flow velocity field data corresponding to the training input sample based on the sensor data corresponding to the training input sample;

[0152] Calculating the temperature change rate of the training input sample within a preset time and first gradient information of the training input sample;

[0153] Calculating a first flow field value corresponding to the training input sample based on the flow velocity field data, the temperature change rate, and the first gradient information;

[0154] Outputting current sample identification data corresponding to the training input sample through the intelligent path planning model in the current iteration, and obtaining the thermal diffusion coefficient of the current sample target in the current sample identification data;

[0155] Calculating a second flow field value corresponding to the training input sample based on the thermal diffusion coefficient and the first gradient information;

[0156] The fluid flow residual loss corresponding to the training input sample is calculated based on the first flow field value and the second flow field value.

[0157] Optionally, the training module 72 is further configured to:

[0158] Acquire the radiation power of the infrared imaging device, the emissivity parameter of the infrared imaging device, the radiation area of ​​the infrared imaging device, and the ambient temperature data corresponding to the training input sample;

[0159] Calculating sample temperature data of the training input sample;

[0160] Calculating a radiation power estimation value corresponding to the training input sample based on the emissivity parameter, the radiation area, the ambient temperature data, and the sample temperature data;

[0161] Based on the radiation power and the radiation power estimation value, a thermal radiation residual loss corresponding to the training input sample is calculated.

[0162] Optionally, the current total loss value also includes at least one of data loss indicating temperature loss and path optimization constraint loss.

[0163] Optionally, the path optimization constraint loss includes an obstacle avoidance constraint loss based on a temperature field, and the training module 72 is further configured to:

[0164] Outputting current sample identification data corresponding to the training input sample by the intelligent path planning model in the current iteration, obtaining a current sample obstacle target from the current sample identification data, and obtaining a current sample temperature of the current sample obstacle target at the current position at the current moment from the current sample identification data;

[0165] Get the sample temperature threshold;

[0166] The obstacle avoidance constraint loss is calculated according to the current sample temperature and the sample temperature threshold.

[0167] Optionally, the path optimization constraint loss includes a temperature constraint loss under extreme conditions. The training module 72 is further configured to:

[0168] Outputting current sample identification data corresponding to the training input sample by the intelligent path planning model in the current iteration, and obtaining the quality of the current sample target in the current sample identification data;

[0169] Obtain the specific heat capacity parameters of infrared imaging equipment;

[0170] Calculating the temperature change value of the training input sample within a preset time;

[0171] Calculating a first temperature constraint value corresponding to the current sample target according to the mass of the current sample target, the specific heat capacity parameter, and the temperature change value;

[0172] Obtaining a preset first extreme ambient temperature value and a preset second extreme ambient temperature value, and calculating a second temperature constraint value based on the first extreme ambient temperature value and the second extreme ambient temperature value;

[0173] The self-temperature constraint loss corresponding to the current sample target is calculated based on the first temperature constraint value and the second temperature constraint value corresponding to the current sample target.

[0174] Optionally, the training module 72 is further configured to:

[0175] Obtain sample sensor data and training samples;

[0176] Based on the sample sensor data, modeling is performed to generate static simulation environment data;

[0177] Continuous trajectory data of the dynamic sample target is obtained from the training sample, and dynamic simulation environment data is modeled and generated based on the continuous trajectory data and the static simulation environment data. The dynamic simulation environment data includes dynamic simulation environment data of path data of the dynamic sample target that changes with time parameters.

[0178] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of the method. In order to achieve the above functions, the training method device based on path planning includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0179] The embodiment of the present application can, according to the above method, exemplarily divide the path planning-based training method device into functional modules. For example, the path planning-based training method device can include various functional modules corresponding to the various functional divisions, or two or more functions can be integrated into one processing module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation.

[0180] See also Figure 9 On the other hand, an embodiment of the present application further provides a computing device 30, including a processing device 303 and a storage device 304, wherein the storage device 304 stores a computer program. When the computer program is executed by the processor, the processing device 303 executes the steps of a path planning method provided in any of the above embodiments of the present application.

[0181] The processing device 303 serves as the control center, connecting the various components of the infrared imaging device using various interfaces and circuits. By running or executing software programs and / or modules stored in the storage device 304 and accessing data stored in the storage device 304, the processing device 303 performs various functions and processes data. Optionally, the processing device 303 may include one or more processing cores. Preferably, the processing device 303 may integrate an application processor and a modem processor, with the application processor primarily processing the operating system, user interfaces, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into the processing device 303.

[0182] The storage device 304 can be used to store software programs and modules. The processing device 303 executes various functional applications and data processing by running the software programs and modules stored in the storage device 304. The storage device 304 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created based on the use of an infrared imaging device. In addition, the storage device 304 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the storage device 304 may also include a memory processor to provide the processing device 303 with access to the storage device 304.

[0183] On the other hand, an embodiment of the present application further provides a computer-readable non-volatile storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of a path planning-based training method provided in any of the above embodiments of the present application, and / or the steps of the path planning method provided in any embodiment.

[0184] On the other hand, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements a training method based on path planning as described in any embodiment of the present application, and / or the steps of the path planning method provided in any embodiment.

[0185] On the other hand, an embodiment of the present application provides a vehicle device, including a controller and a storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the controller, the controller executes the path planning method provided in the second aspect of the present application.

[0186] Those skilled in the art will appreciate that all or part of the processes in the methods provided in the above embodiments can be implemented by instructing the relevant hardware through a computer program. The 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 methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0187] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. The scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A training method based on path planning, characterized in that: include: Acquire a training data set, wherein each training sample in the training data set includes continuous frame samples and label data of sample targets in each continuous frame sample; Build an initial intelligent path planning model; Based on the training samples in the training data set and the loss function including the physical information loss, the intelligent path planning model is iteratively trained until an intelligent path planning model that meets the training iteration conditions is obtained.

2. The training method based on path planning according to claim 1, characterized in that: The iterative training of the intelligent path planning model based on the training samples in the training data set and the loss function including the physical information loss until the intelligent path planning model that meets the training iteration conditions is obtained includes: Obtaining training input samples from the training data set, and forming input of the intelligent path planning model in the current iteration based on the training input samples; Using the loss function and the training input samples, the current total loss value in the current iteration is calculated. Based on the current total loss value, training input samples are continued to be obtained from the training data set for iterative training until an intelligent path planning model that meets the training iteration conditions is obtained.

3. The path planning-based training method according to claim 2, wherein: The physical information loss includes at least one of a heat conduction residual loss based on a temperature field, a fluid flow residual loss based on a flow field, and a heat radiation residual loss based on a heat radiation model.

4. The path planning-based training method according to claim 3, wherein: Calculating the current total loss value in the current iteration using the loss function and the training input sample includes: Outputting current sample identification data corresponding to the training input sample through the intelligent path planning model in the current iteration; Calculating a first heat change value of a current sample target in the current sample identification data based on the current sample identification data; Calculating a second heat change value of the current sample target based on a temperature label corresponding to the current sample target in the training input sample; The heat conduction residual loss of the current sample target is calculated based on the first heat change value and the second heat change value.

5. The path planning-based training method according to claim 4, wherein: The calculating, based on the current sample identification data, a first heat change value of the current sample target in the current sample identification data includes: Obtaining the material density of the current sample target and the specific heat capacity parameter of the infrared imaging device; Calculating the temperature change rate of the training input sample within a preset time; Calculating the first heat change value based on the material density, the specific heat capacity parameter, and the temperature change rate; Calculating the second heat change value of the current sample target based on the temperature label corresponding to the current sample target in the training input sample includes: Obtaining the thermal conductivity of the infrared imaging device; Calculating first gradient information of the training input sample; Calculating the current heat of the current sample target at the current moment and the radiant energy indicating the heat change within a preset time period after the current moment; The second heat change value is calculated based on the thermal conductivity, the first gradient information, the current heat amount, and the radiation energy.

6. The path planning-based training method according to claim 3, wherein: Calculating the current total loss value in the current iteration using the loss function and the training input sample includes: Acquire sensor data corresponding to the training input sample, and calculate flow velocity field data corresponding to the training input sample based on the sensor data corresponding to the training input sample; Calculating the temperature change rate of the training input sample within a preset time and first gradient information of the training input sample; Calculating a first flow field value corresponding to the training input sample based on the flow velocity field data, the temperature change rate, and the first gradient information; Outputting current sample identification data corresponding to the training input sample through the intelligent path planning model in the current iteration, and obtaining the thermal diffusion coefficient of the current sample target in the current sample identification data; Calculating a second flow field value corresponding to the training input sample based on the thermal diffusion coefficient and the first gradient information; The fluid flow residual loss corresponding to the training input sample is calculated based on the first flow field value and the second flow field value.

7. The training method based on path planning according to claim 3, wherein: Calculating the current total loss value in the current iteration using the loss function and the training input sample includes: Acquire the radiation power of the infrared imaging device, the emissivity parameter of the infrared imaging device, the radiation area of ​​the infrared imaging device, and the ambient temperature data corresponding to the training input sample; Calculating sample temperature data of the training input sample; Calculating a radiation power estimation value corresponding to the training input sample based on the emissivity parameter, the radiation area, the ambient temperature data, and the sample temperature data; Based on the radiation power and the radiation power estimation value, a thermal radiation residual loss corresponding to the training input sample is calculated.

8. The training method based on path planning according to claim 1, wherein: The current total loss value further includes at least one of a data loss indicating a temperature loss and a path optimization constraint loss.

9. The path planning-based training method according to claim 8, wherein: The path optimization constraint loss includes an obstacle avoidance constraint loss based on a temperature field, and the calculation of the current total loss value in the current iteration using the loss function and the training input sample includes: Outputting current sample identification data corresponding to the training input sample by the intelligent path planning model in the current iteration, obtaining a current sample obstacle target from the current sample identification data, and obtaining a current sample temperature of the current sample obstacle target at the current position at the current moment from the current sample identification data; Get the sample temperature threshold; The obstacle avoidance constraint loss is calculated according to the current sample temperature and the sample temperature threshold.

10. The training method based on path planning according to claim 8, characterized in that: The path optimization constraint loss includes the temperature constraint loss under extreme conditions. The calculation of the current total loss value in the current iteration using the loss function and the training input sample includes: Outputting current sample identification data corresponding to the training input sample by the intelligent path planning model in the current iteration, and obtaining the quality of the current sample target in the current sample identification data; Obtain the specific heat capacity parameters of infrared imaging equipment; Calculating the temperature change value of the training input sample within a preset time; Calculating a first temperature constraint value corresponding to the current sample target according to the mass of the current sample target, the specific heat capacity parameter, and the temperature change value; Obtaining a preset first extreme ambient temperature value and a preset second extreme ambient temperature value, and calculating a second temperature constraint value based on the first extreme ambient temperature value and the second extreme ambient temperature value; The self-temperature constraint loss corresponding to the current sample target is calculated based on the first temperature constraint value and the second temperature constraint value corresponding to the current sample target.

11. The training method based on path planning according to claim 1, wherein: The obtaining of the training data set comprises: Obtain sample sensor data and training samples; Based on the sample sensor data, modeling is performed to generate static simulation environment data; Continuous trajectory data of the dynamic sample target is obtained from the training sample, and dynamic simulation environment data is modeled and generated based on the continuous trajectory data and the static simulation environment data. The dynamic simulation environment data includes dynamic simulation environment data of path data of the dynamic sample target that changes with time parameters.

12. A path planning method, characterized in that: The method comprises: Obtain the actual scene data collected; Based on the actual scene data, an input of a pre-trained intelligent path planning model is formed, and global path planning data is output, wherein the intelligent path planning model is trained based on the path planning-based training method described in any one of claims 1 to 11.

13. An infrared imaging device, characterized in that: It includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the training method based on path planning as described in any one of claims 1 to 11, and / or the path planning method as described in claim 11.

14. A computing device, characterized in that The system comprises a storage device and a processing device, wherein the storage device stores a computer program, and when the computer program is executed by the processing device, the processing device executes the training method based on path planning as claimed in any one of claims 1 to 11.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the training method based on path planning as claimed in any one of claims 1 to 11 and / or the path planning method as claimed in claim 12 is implemented.

Images