WO2021093011A1 - 无人车驾驶决策方法、无人车驾驶决策装置及无人车 - Google Patents

无人车驾驶决策方法、无人车驾驶决策装置及无人车 Download PDF

Info

Publication number
WO2021093011A1
WO2021093011A1 PCT/CN2019/120559 CN2019120559W WO2021093011A1 WO 2021093011 A1 WO2021093011 A1 WO 2021093011A1 CN 2019120559 W CN2019120559 W CN 2019120559W WO 2021093011 A1 WO2021093011 A1 WO 2021093011A1
Authority
WO
WIPO (PCT)
Prior art keywords
decision
driving
unmanned vehicle
training
image
Prior art date
Application number
PCT/CN2019/120559
Other languages
English (en)
French (fr)
Inventor
李国法
李盛龙
杨一帆
纪泽锋
卢宗鹏
阳亮
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Publication of WO2021093011A1 publication Critical patent/WO2021093011A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
    • B60W30/18—Propelling the vehicle
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Definitions

  • This application belongs to the technical field of unmanned vehicles, and in particular relates to an unmanned vehicle driving decision-making method, an unmanned vehicle driving decision device, and an unmanned vehicle.
  • Unmanned vehicles are smart cars that perceive the road environment through the on-board sensor system, automatically generate driving decisions based on the road environment, and control the vehicle to reach a predetermined destination based on the driving decisions.
  • Unmanned vehicles integrate many technologies such as automatic control, architecture, artificial intelligence, and visual computing. It is a product of the highly developed computer science, pattern recognition and intelligent control technology, and has broad development prospects.
  • the existing decision-making methods for unmanned vehicle driving mainly establish a kinematics dynamic model of the self-car or other vehicles, and use this model as a decision-making model to predict driving actions.
  • a kinematics dynamic model of the self-car or other vehicles and use this model as a decision-making model to predict driving actions.
  • there are many time-varying uncertain factors so it is impossible to accurately establish the kinematics dynamic model of other cars or self-cars, which affects the driving decision-making ability of unmanned vehicles and cannot guarantee the driving process of unmanned vehicles. safety.
  • One of the objectives of the embodiments of the present application is to provide an unmanned vehicle driving decision-making method, an unmanned vehicle driving decision-making device, and an unmanned vehicle, which can solve the problem of the low decision-making ability of unmanned vehicle driving in the prior art, which leads to unmanned driving.
  • the problem of low safety during driving is to provide an unmanned vehicle driving decision-making method, an unmanned vehicle driving decision-making device, and an unmanned vehicle, which can solve the problem of the low decision-making ability of unmanned vehicle driving in the prior art, which leads to unmanned driving.
  • the problem of low safety during driving is to provide an unmanned vehicle driving decision-making method, an unmanned vehicle driving decision-making device, and an unmanned vehicle, which can solve the problem of the low decision-making ability of unmanned vehicle driving in the prior art, which leads to unmanned driving.
  • the problem of low safety during driving is to provide an unmanned vehicle driving decision-making method, an unmanned vehicle driving decision-making device, and an unmanned vehicle, which can solve the problem of the low decision-making ability
  • an embodiment of the present application provides an unmanned vehicle driving decision-making method, including:
  • the candidate driving action corresponding to the largest decision value in the decision result is determined as the target driving action of the unmanned vehicle at the next decision moment.
  • the method before the image to be decided is input into the trained decision model to obtain the decision result, the method further includes:
  • the training image is a photographed image of the road in front of the unmanned vehicle
  • the driving information corresponding to each training image is obtained separately;
  • Each set of training data is used to perform iterative training on the preset decision model to obtain a trained decision model, wherein each set of training data includes a training image and driving information corresponding to the training image.
  • the driving information corresponding to the training image includes a target driving action, an environmental return value corresponding to the target driving action, and an action image.
  • the method of obtaining driving information corresponding to each training image separately based on the preset decision model includes:
  • the first output result includes a plurality of candidate driving actions and a first output corresponding to each candidate driving action value
  • a captured image of the road ahead of the unmanned vehicle is acquired to obtain an action image, and an environmental reward value corresponding to the target driving action is calculated.
  • the calculating the environmental return value corresponding to the target driving action includes:
  • the environmental reward value corresponding to the target driving action is calculated.
  • the calculation of the driving reward value according to the driving speed, the preset minimum speed limit, and the preset maximum speed limit includes:
  • R velocity is the driving return value
  • v t is the driving speed after the unmanned vehicle performs the target driving action
  • v min is the minimum speed limit
  • v max is the maximum speed limit
  • the calculating a deceleration reward value according to the target driving action includes:
  • the formula Calculate the deceleration reward value where R stop is the deceleration reward value, d is the remaining driving distance, K is the first preset value, and the remaining driving distance is after the unmanned vehicle performs the target driving action The distance between the position of and the preset target position;
  • the deceleration return value is set to a second preset value.
  • the i-th iterative training process in the step of iteratively training the preset decision model using each set of training data to obtain the trained decision model, the i-th iterative training process is performed include:
  • the reference model when i ⁇ N, the reference model is a preset decision model; when i>N, the reference model is a decision model after the i-1th iteration training; N is a positive integer.
  • an embodiment of the present application provides an unmanned vehicle driving decision-making device, which is characterized in that it includes:
  • a pending decision image acquisition unit configured to obtain a photographed image of the road ahead of the unmanned vehicle at the current decision moment to obtain the pending decision image
  • the decision-making unit is configured to input the image to be decided into a trained decision model to obtain a decision result, the decision result including multiple candidate driving actions and a decision value corresponding to each candidate driving action;
  • the result determining unit is configured to determine the candidate driving action corresponding to the largest decision value in the decision result as the target driving action of the unmanned vehicle at the next decision moment.
  • an embodiment of the present application provides an unmanned vehicle, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes The computer program implements the unmanned vehicle driving decision method according to any one of the above-mentioned first aspects.
  • an embodiment of the present application provides a computer-readable storage medium
  • an embodiment of the present application provides a computer-readable storage medium
  • the computer-readable storage medium stores a computer program, and is characterized in that the When the computer program is executed by the processor, the driverless vehicle driving decision method according to any one of the above-mentioned first aspects is realized.
  • the embodiments of the present application provide a computer program product, which when the computer program product runs on a terminal device, causes the terminal device to execute the unmanned vehicle driving decision method described in any one of the above-mentioned first aspects.
  • the embodiment of the application obtains the image to be decided by acquiring the photographed image of the road ahead of the unmanned vehicle at the current decision time; the image to be decided is input into the trained decision model to obtain the decision result, and the decision result includes multiple Candidate driving actions and the decision values corresponding to each candidate driving action; the trained decision model can more accurately reflect the mapping relationship between the road environment in front of the unmanned vehicle and the decision result, avoiding time-varying uncertain factors. Influence; determining the candidate driving action corresponding to the largest decision value in the decision result as the target driving action of the unmanned vehicle at the next decision moment.
  • Fig. 1 is a schematic diagram of an unmanned vehicle system provided by an embodiment of the present application
  • FIG. 2 is a schematic flowchart of an unmanned vehicle driving decision-making method provided by an embodiment of the present application
  • FIG. 3 is a schematic flowchart of a training method for an unmanned vehicle driving decision model provided by an embodiment of the present application
  • FIG. 4 is a schematic flowchart of a method for obtaining driving information provided by an embodiment of the present application.
  • FIG. 5 is a schematic diagram of an iterative training process provided by an embodiment of the present application.
  • FIG. 6 is a structural block diagram of an unmanned vehicle driving decision device provided by an embodiment of the present application.
  • Fig. 7 is a schematic structural diagram of an unmanned vehicle provided by an embodiment of the present application.
  • references described in the specification of this application to "one embodiment” or “some embodiments”, etc. mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in conjunction with the embodiment. Therefore, the sentences “in one embodiment”, “in some embodiments”, “in some other embodiments”, “in some other embodiments”, etc. appearing in different places in this specification are not necessarily All refer to the same embodiment, but mean “one or more but not all embodiments” unless it is specifically emphasized otherwise.
  • FIG. 1 is a schematic diagram of an unmanned vehicle system provided by an embodiment of this application.
  • a processor 10 a controller 11, and a camera 12 for taking pictures of the road in front of the unmanned vehicle may be provided on the unmanned vehicle.
  • the processor 10 is provided with an unmanned vehicle driving decision device provided in an embodiment of the present application.
  • the unmanned vehicle driving decision-making device is used to implement the unmanned vehicle driving decision-making method provided in the embodiment of the present application.
  • the unmanned vehicle driving decision-making device may first control the camera 12 through the controller 11 to obtain the photographed image of the road in front of the unmanned vehicle at the current decision-making moment to obtain the image to be decided, and then use the image provided by the embodiment of this application
  • the unmanned vehicle driving decision-making method and the acquired image to be decided are used for decision-making, and the target driving action of the unmanned vehicle at the next decision-making moment is obtained, and then the target driving action can be fed back to the unmanned vehicle controller 11, which is controlled by the controller 11
  • the unmanned vehicle performs the target driving action.
  • the decision-making process Before the decision-making process, it also includes the training process of the decision-making model.
  • the unmanned vehicle driving decision-making device can control the camera device 12 through the controller 11 to obtain multiple training images, and use the training method of the unmanned vehicle driving decision-making model provided by the embodiments of the application and the obtained multiple training images to compare preset values.
  • the decision model is iteratively trained to obtain the trained decision model.
  • FIG. 2 shows a schematic flow chart of an unmanned vehicle driving decision method provided by an embodiment of the present application.
  • the method may include the following steps:
  • S202 Input the image to be decided into a trained decision model to obtain a decision result, where the decision result includes multiple candidate driving actions and a decision value corresponding to each candidate driving action.
  • S203 Determine the candidate driving action corresponding to the largest decision value in the decision result as the target driving action of the unmanned vehicle at the next decision moment.
  • the interval of the decision time can be manually set. For example, it can be set to make a decision every 3s, then the next decision time of the current decision time is the time 3s after the current decision time.
  • step S202 before the image to be decided is input into the trained decision model to obtain the decision result, the method further includes a training process of the decision model.
  • the training method of the unmanned vehicle driving decision model may include the following steps:
  • training images in different time periods and in different environments should be selected as far as possible.
  • S302 Obtain driving information corresponding to each training image based on a preset decision model.
  • the preset decision network may be a convolutional neural network.
  • the preset decision-making network may be composed of 6 layers of convolutional layers, 3 layers of pooling layers, and 2 layers of fully connected layers.
  • each set of training data uses iterative training on the preset decision model to obtain a trained decision model, where each set of training data includes a training image and driving information corresponding to the training image.
  • each iteration of training a set of training data is used for training.
  • the training data selected for each iterative training is different.
  • the driving information corresponding to the training image includes a target driving action, an environmental return value corresponding to the target driving action, and an action image.
  • FIG. 4 is a schematic flowchart of a method for obtaining driving information provided by an embodiment of this application.
  • the method of obtaining driving information corresponding to a training image based on a preset decision model in step S302 may include the following steps:
  • the driving candidate driving actions may include slow driving, faster acceleration, slow deceleration, faster deceleration, emergency braking, and the like.
  • S402 Determine the candidate driving action corresponding to the largest first output value in the first output result as the target driving action of the unmanned vehicle.
  • calculating the environmental return value corresponding to the target driving action in S403 may include:
  • S41 Obtain the driving speed of the unmanned vehicle after executing the target driving action, and calculate a driving reward value according to the driving speed, a preset minimum speed limit, and a preset maximum speed limit.
  • R velocity is the driving return value
  • v t is the driving speed after the unmanned vehicle performs the target driving action
  • v min is the minimum speed limit
  • v max is the maximum speed limit
  • S42 Obtain a driving state after the unmanned vehicle performs the target driving action, and determine a preset collision report value corresponding to the driving state, and the driving state includes collision and non-collision.
  • the preset collision report value when the driving state is collision, the preset collision report value is -10; when the driving state is non-collision, the preset collision report value is 0.
  • the preset collision return value can be defined based on experimental experience, and there is no specific limitation here.
  • S43 may include:
  • the formula Calculate the deceleration reward value where R stop is the deceleration reward value, d is the remaining driving distance, K is the first preset value, and the remaining driving distance is after the unmanned vehicle performs the target driving action The distance between the position and the preset target position.
  • the deceleration return value is set to a second preset value.
  • the braking action can also be attributed to the deceleration action, that is, when the target driving action is deceleration or braking, the formula must be passed Calculate the deceleration return value; and other target driving actions correspond to the second preset value.
  • S44 Calculate an environmental reward value corresponding to the target driving action according to the driving reward value, the preset collision reward value, and the deceleration reward value.
  • the driving reward value, the preset collision reward value and the deceleration reward value can be directly added to obtain the environmental reward value.
  • the driving reward value, the preset collision reward value and the deceleration reward value can also be weighted and summed to obtain the environmental reward value.
  • the specific calculation method is not limited.
  • FIG. 5 is a schematic diagram of the iterative training process provided in this embodiment of the application.
  • the process of performing the i-th iterative training includes:
  • S501 Input the training images in the i-th group of training data into the decision model after the i-1th iterative training to obtain a second output result.
  • the second output result includes a plurality of candidate driving actions and a second output value corresponding to each candidate driving action.
  • S502 Input the action image in the i-th set of training data into a reference model to obtain a third output result.
  • the third output result includes a plurality of candidate driving actions and a third output value corresponding to each candidate driving action.
  • the candidate driving action in the third output result is the same as the candidate driving action in the second output result.
  • S503 Use the second output result and the third output result to train the decision model after the i-1th iteration training.
  • the reference model when i ⁇ N, the reference model is a preset decision model; when i>N, the reference model is a decision model after the i-1th iteration training; N is a positive integer.
  • a reference model is introduced.
  • the driving N is 3.
  • the reference models are all preset decision models, that is, the initially defined decision models.
  • the reference model is the decision model after the first iteration training;
  • the reference model is the decision model after the second iteration training; analogy.
  • the reference model is a decision model that is different from the decision model in the current iterative training process by N iterations of the training process. Using such a training method can effectively ensure the stability of the iterative training process.
  • step S503 using the second output result and the third output result to train the decision model after the i-1th iteration training may include the following steps:
  • A. Obtain the second output value corresponding to the target driving action in the i-th set of training data in the second output result to obtain the first calculated value, obtain the largest third output value in the third output result to obtain the second calculated value, and The error value of the decision model after the i-1th iteration training is calculated using the first calculated value, the second calculated value, and the environmental return value in the i-th set of training data.
  • the error value can be calculated by the following formula:
  • L( ⁇ ) is the error value
  • Q(s, a; ⁇ ) is the first calculated value
  • s is the training image
  • a is the target driving action
  • ⁇ is the weight of the decision model after the i-1th iteration training. Value
  • TargetQ r+ ⁇ max a′ Q(s′,a′; ⁇ ′);
  • r is the environmental return value
  • ⁇ is the discount factor
  • max a′ Q(s′, a′; ⁇ ′) is the second calculated value
  • s′ represents the action image
  • a′ represents the candidate driving corresponding to the second calculated value Action
  • ⁇ ′ represents the weight of the preset decision model.
  • the decision model trained in the i-1th iteration is used as the trained decision model.
  • the gradient descent method can be used to update the weights.
  • the embodiment of the application obtains the image to be decided by acquiring the photographed image of the road ahead of the unmanned vehicle at the current decision time; the image to be decided is input into the trained decision model to obtain the decision result, and the decision result includes multiple Candidate driving actions and the decision values corresponding to each candidate driving action; the trained decision model can more accurately reflect the mapping relationship between the road environment in front of the unmanned vehicle and the decision result, avoiding time-varying uncertain factors. Influence; determining the candidate driving action corresponding to the largest decision value in the decision result as the target driving action of the unmanned vehicle at the next decision moment.
  • FIG. 6 shows a structural block diagram of the unmanned vehicle driving decision-making apparatus provided by an embodiment of the present application. For ease of description, only the same as the embodiment of the present application is shown. The relevant part.
  • the device includes:
  • the decision-to-decision image acquisition unit 61 is configured to obtain the photographed image of the road ahead of the unmanned vehicle at the current decision moment to obtain the decision-to-decision image.
  • the decision unit 62 is configured to input the image to be decided into the trained decision model to obtain a decision result.
  • the decision result includes a plurality of candidate driving actions and a decision value corresponding to each candidate driving action.
  • the result determination unit 63 is configured to determine the candidate driving action corresponding to the largest decision value in the decision result as the target driving action of the unmanned vehicle at the next decision moment.
  • the device 6 further includes:
  • the training image acquisition unit 64 is configured to acquire multiple training images before inputting the image to be decided into the trained decision model to obtain the decision result, wherein the training image is a photograph of the road in front of the unmanned vehicle image.
  • the driving information acquiring unit 65 is configured to separately acquire driving information corresponding to each training image based on a preset decision model.
  • the decision model training unit 66 is configured to use each set of training data to perform iterative training on the preset decision model to obtain a trained decision model, where each set of training data includes a training image and a training image corresponding to the training image. Driving information.
  • the driving information corresponding to the training image includes a target driving action, an environmental return value corresponding to the target driving action, and an action image.
  • the driving information acquiring unit 65 includes:
  • the first output result module is used to input the training image into the preset decision model to obtain a first output result for each training image.
  • the first output result includes a plurality of candidate driving actions and each The first output value corresponding to the candidate driving action.
  • the target action module is configured to determine the candidate driving action corresponding to the largest first output value in the first output result as the target driving action of the unmanned vehicle.
  • the calculation module is configured to obtain a photographed image of the road ahead of the unmanned vehicle to obtain an action image after the unmanned vehicle performs the target driving action, and calculate the environmental reward value corresponding to the target driving action.
  • the calculation module includes:
  • the speed acquisition sub-module is used to acquire the driving speed of the unmanned vehicle after executing the target driving action, and calculate the driving reward value according to the driving speed, the preset minimum speed and the preset maximum speed.
  • the state acquisition sub-module is used to acquire the driving state of the unmanned vehicle after the target driving action is executed, and determine a preset collision report value corresponding to the driving state, and the driving state includes collision and non-collision.
  • the first calculation sub-module is used to calculate the deceleration return value according to the target driving action.
  • the second calculation sub-module is configured to calculate the environmental reward value corresponding to the target driving action according to the driving reward value, the preset collision reward value and the deceleration reward value.
  • the speed acquisition sub-module is also used to pass formula Calculate the driving reward value.
  • R velocity is the driving return value
  • v t is the driving speed after the unmanned vehicle performs the target driving action
  • v min is the minimum speed limit
  • v max is the maximum speed limit
  • the first calculation sub-module is further configured to pass the formula if the target driving action is deceleration Calculate the deceleration reward value, where R stop is the deceleration reward value, d is the remaining driving distance, K is the first preset value, and the remaining driving distance is after the unmanned vehicle performs the target driving action The distance between the position of and the preset target position; if the target driving action is not deceleration, the deceleration return value is set to a second preset value.
  • the decision model training unit 66 includes:
  • the second output result module is used to input the training images in the i-th group of training data into the decision model after the i-1th iterative training to obtain the second output result.
  • the third output result module is used to input the action image in the i-th group of training data into the reference model to obtain the third output result.
  • the iterative training module is configured to use the second output result and the third output result to train the decision model after the i-1th iterative training.
  • the reference model when i ⁇ N, the reference model is a preset decision model; when i>N, the reference model is a decision model after the i-1th iteration training; N is a positive integer.
  • FIG. 6 shows a structural block diagram of the unmanned vehicle driving decision-making apparatus provided by an embodiment of the present application. For ease of description, only the same as the embodiment of the present application is shown. The relevant part.
  • the device shown in FIG. 6 can be a software unit, a hardware unit, or a combination of software and hardware built into an existing terminal device, can also be integrated into the terminal device as an independent pendant, or can be used as an independent The terminal device exists.
  • FIG. 7 is a schematic structural diagram of an unmanned vehicle provided by an embodiment of the application.
  • the unmanned vehicle 7 of this embodiment includes: at least one processor 70 (only one is shown in FIG. 7), a processor, a memory 71, and a memory 71 that is stored in the memory 71 and can be stored in the at least one processor.
  • a computer program 72 running on the processor 70 when the processor 70 executes the computer program 72, the steps in any of the embodiments of the driverless vehicle driving decision-making method described above are implemented.
  • the unmanned vehicle may include, but is not limited to, a processor and a memory.
  • FIG. 7 is only an example of the unmanned vehicle 7 and does not constitute a limitation on the unmanned vehicle 7. It may include more or fewer components than shown in the figure, or combine certain components, or different components.
  • the components of, for example, can also include input and output devices, network access devices, and so on.
  • the so-called processor 70 may be a central processing unit (Central Processing Unit, CPU), and the processor 70 may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs), and application specific integrated circuits (Application Specific Integrated Circuits). , ASIC), ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the memory 71 may be an internal storage unit of the unmanned vehicle 7 in some embodiments, such as a hard disk or a memory of the unmanned vehicle 7. In other embodiments, the memory 71 may also be an external storage device of the unmanned vehicle 7, for example, a plug-in hard disk or a smart memory card (Smart Media Card, SMC) equipped on the unmanned vehicle 7. Secure Digital (SD) card, Flash Card, etc. Further, the memory 71 may also include both an internal storage unit of the unmanned vehicle 7 and an external storage device. The memory 71 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 71 can also be used to temporarily store data that has been output or will be output.
  • a boot loader BootLoader
  • the embodiments of the present application also provide a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in each of the foregoing method embodiments can be realized.
  • the embodiments of the present application provide a computer program product.
  • the steps in the foregoing method embodiments can be realized when the mobile terminal is executed.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the computer program can be stored in a computer-readable storage medium.
  • the computer program can be stored in a computer-readable storage medium.
  • the steps of the foregoing method embodiments can be implemented.
  • the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file, or some intermediate forms.
  • the computer-readable medium may at least include: any entity or device capable of carrying the computer program code to an unmanned vehicle, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), and a random access memory (RAM). , Random Access Memory), electric carrier signal, telecommunications signal and software distribution medium.
  • ROM read-only memory
  • RAM random access memory
  • Electric carrier signal telecommunications signal and software distribution medium.
  • U disk mobile hard disk, floppy disk or CD-ROM, etc.
  • computer-readable media cannot be electrical carrier signals and telecommunication signals.
  • the disclosed apparatus/network equipment and method may be implemented in other ways.
  • the device/network device embodiments described above are only illustrative.
  • the division of the modules or units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units.
  • components can be combined or integrated into another system, or some features can be omitted or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Automation & Control Theory (AREA)
  • Transportation (AREA)
  • Mechanical Engineering (AREA)
  • Multimedia (AREA)
  • Traffic Control Systems (AREA)

Abstract

一种无人车驾驶决策方法、无人车驾驶决策装置（6）及无人车（7），包括：获取当前决策时刻无人车（7）前方道路的拍摄图像，得到待决策图像（S2101）；将待决策图像输入到训练后的决策模型中得到决策结果，决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值（S202）；将决策结果中最大的决策值对应的候选驾驶动作确定为无人车（7）在下一决策时刻的目标驾驶动作（S203）。有效提高了无人车（7）的驾驶决策能力，进而有效保证了无人车（7）驾驶过程的安全性。

Description

无人车驾驶决策方法、无人车驾驶决策装置及无人车
本申请要求于2019年11月14日提交中国专利局、申请号为201911113303.3、发明名称为“无人车驾驶决策方法、无人车驾驶决策装置及无人车”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请属于无人车技术领域，尤其涉及一种无人车驾驶决策方法、无人车驾驶决策装置及无人车。
背景技术
无人驾驶汽车(简称无人车)是通过车载传感系统感知道路环境，根据道路环境自动生成驾驶决策，并根据驾驶决策控制车辆到达预定目的地的智能汽车。无人车集自动控制、体系结构、人工智能、视觉计算等众多技术于一体，是计算机科学、模式识别和智能控制技术高度发展的产物，具有广阔的发展前景。
现有的无人车驾驶决策方法主要是通过建立自车或他车的运动学动态模型，并将该模型作为决策模型来对驾驶动作进行预测。但是在实际应用中，存在许多时变不确定性因素，因而无法准确地建立他车或自车的运动学动态模型，进而影响了无人车的驾驶决策能力，无法保障无人车行驶过程的安全性。
技术问题
本申请实施例的目的之一在于：提供了一种无人车驾驶决策方法、无人车驾驶决策装置及无人车，可以解决现有技术中无人车驾驶决策能力较低、导致无人车行驶过程的安全性较低的问题。
技术解决方案
为解决上述技术问题，本申请实施例采用的技术方案是：
第一方面，本申请实施例提供了一种无人车驾驶决策方法，包括：
获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；
将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值；
将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。
在第一方面的一种可能的实现方式中，在将所述待决策图像输入到训练后的决策模型中得到决策结果之前，所述方法还包括：
获取多幅训练图像，其中，所述训练图像为所述无人车前方道路的拍摄图像；
基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息；
利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型，其中，每组训练数据中包括一幅训练图像和所述训练图像对应的驾驶信息。
在第一方面的一种可能的实现方式中，所述训练图像对应的驾驶信息包括目标驾驶动作、所述目标驾驶动作对应的环境回报值和动作图像。
所述基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息，包括：
对于每幅训练图像，将所述训练图像输入到所述预设的决策模型中得到第一输出结果，所述第一输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第一输出值；
将所述第一输出结果中最大的第一输出值对应的候选驾驶动作确定为所述无人车的目标驾驶动作；
在所述无人车执行所述目标驾驶动作之后，获取所述无人车前方道路的拍摄图像得到动作图像，并计算所述目标驾驶动作对应的环境回报值。
在第一方面的一种可能的实现方式中，所述计算所述目标驾驶动作对应的环境回报值，包括：
获取所述无人车执行所述目标驾驶动作之后的驾驶速度，并根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值；
获取所述无人车执行所述目标驾驶动作之后的驾驶状态，并确定所述驾驶状态对应的预设碰撞回报值，所述驾驶状态包括碰撞和非碰撞；
根据所述目标驾驶动作计算减速回报值；
根据所述行驶回报值、所述预设碰撞回报值和所述减速回报值，计算所述目标驾驶动作对应的环境回报值。
在第一方面的一种可能的实现方式中，所述根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值，包括：
通过公式
Figure PCTCN2019120559-appb-000001
计算所述行驶回报值；
其中，R velocity为所述行驶回报值，v t为所述无人车执行所述目标驾驶动作之后的驾驶速度，v min为所述最小限速，v max为所述最大限速。
在第一方面的一种可能的实现方式中，所述根据所述目标驾驶动作计算减速回报值，包括：
若所述目标驾驶动作是减速，则通过公式
Figure PCTCN2019120559-appb-000002
计算所述减速回报值，其中，R stop为所述减速回报值，d为剩余驾驶距离，K为第一预设值，所述剩余驾驶距离为所述无人车执行所述目标驾驶动作之后的位置与预设目标位置之间的距离；
若所述目标驾驶动作不是减速，则将所述减速回报值设置为第二预设值。
在第一方面的一种可能的实现方式中，在所述利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型的步骤中，进行第i次迭代训练的过程包括：
将第i组训练数据中的训练图像输入到第i-1次迭代训练后的决策模型中得到第二输出结果；
将所述第i组训练数据中的动作图像输入到参考模型中得到第三输出结果；
利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练；
其中，当i≤N时，所述参考模型为预设的决策模型；当i>N时，所述参考模型为第i-1次迭代训练后的决策模型；N为正整数。
第二方面，本申请实施例提供了一种无人车驾驶决策装置，其特征在于，包括：
待决策图像获取单元，用于获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；
决策单元，用于将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值；
结果确定单元，用于将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。
第三方面，本申请实施例提供了一种无人车，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如上述第一方面中任一项所述的无人车驾驶决策方法。
第四方面，本申请实施例提供了一种计算机可读存储介质，本申请实施例提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如上述第一方面中任一项所述的无人车驾驶决策方法。
第五方面，本申请实施例提供了一种计算机程序产品，当计算机程序产品在终端设备上运行时，使得终端设备执行上述第一方面中任一项所述的无人车驾驶决策方法。
有益效果
本申请实施例通过获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个 候选驾驶动作、以及各个候选驾驶动作对应的决策值；训练后的决策模型能够较准确地反映出无人车前方道路环境与决策结果之间的映射关系，避免了时变不确定性因素造成的影响；将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。通过上述方法，有效提高了无人车的驾驶决策能力，进而保证了无人车驾驶过程的安全性。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例提供的无人车系统示意图；
图2是本申请一实施例提供的无人车驾驶决策方法的流程示意图；
图3是本申请一实施例提供的无人车驾驶决策模型的训练方法的流程示意图；
图4是本申请一实施例提供的驾驶信息的获取方法的流程示意图；
图5是本申请一实施例提供的迭代训练过程的示意图；
图6是本申请一实施例提供的无人车驾驶决策装置的结构框图；
图7是本申请一实施例提供的无人车的结构示意图。
本发明的实施方式
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、技术之类的具体细节，以便透彻理解本申请实施例。然而，本领域的技术人员应当清楚，在没有这些具体细节的其它实施例中也可以实现本申请。在其它情况中，省略对众所周知的系统、装置、电路以及方法的详细说明，以免不必要的细节妨碍本申请的描述。
应当理解，当在本申请说明书和所附权利要求书中使用时，术语“包括”指示所描述特征、整体、步骤、操作、元素和/或组件的存在，但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解，在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合，并且包括这些组合。
另外，在本申请说明书和所附权利要求书的描述中，术语“第一”、“第二”、“第三”等仅用于区分描述，而不能理解为指示或暗示相对重要性。
在本申请说明书中描述的参考“一个实施例”或“一些实施例”等意味着在本申请的 一个或多个实施例中包括结合该实施例描述的特定特征、结构或特点。由此，在本说明书中的不同之处出现的语句“在一个实施例中”、“在一些实施例中”、“在其他一些实施例中”、“在另外一些实施例中”等不是必然都参考相同的实施例，而是意味着“一个或多个但不是所有的实施例”，除非是以其他方式另外特别强调。
先介绍本申请实施例的一个应用场景。参见图1，为本申请实施例提供的无人车系统示意图。如图1所示，无人车上可以设置有处理器10、控制器11以及用于拍摄无人车前方道路图像的摄像装置12。处理器10中设置有本申请实施例提供的无人车驾驶决策装置。其中，无人车驾驶决策装置用于执行本申请实施例提供的无人车驾驶决策方法。
在无人车的决策过程中，无人车驾驶决策装置可以先通过控制器11控制摄像装置12获取当前决策时刻无人车前方道路的拍摄图像得到待决策图像，然后利用本申请实施例提供的无人车驾驶决策方法和获取到的待决策图像进行决策，得到无人车在下一决策时刻的目标驾驶动作，然后可以将目标驾驶动作反馈给无人车的控制器11，由控制器11控制无人车执行目标驾驶动作。
在决策过程之前，还包括决策模型的训练过程。无人车驾驶决策装置可以通过控制器11控制摄像装置12获取多幅训练图像，并利用本申请实施例提供的无人车驾驶决策模型的训练方法和获取到的多幅训练图像对预设的决策模型进行迭代训练，得到训练后的决策模型。
图2示出了本申请一实施例提供的无人车驾驶决策方法的流程示意图，作为示例而非限定，所述方法可以包括以下步骤：
S201，获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像。
S202，将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值。
其中，决策模型的训练过程可参见图3实施例中的方法。
S203，将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。
本实施例中，决策时刻的间隔可以由人为设定。例如，可以设置为每隔3s做一次决策，那么当前决策时刻的下一决策时刻为，当前决策时刻过3s后的时刻。
在一个实施例中，在步骤S202，将所述待决策图像输入到训练后的决策模型中得到决策结果之前，所述方法还包括决策模型的训练过程。
参见图3，为本申请一实施例提供的无人车驾驶决策模型的训练方法的流程示意图，作为示例而非限定，无人车驾驶决策模型的训练方法可以包括以下步骤：
S301，获取多幅训练图像，其中，所述训练图像为所述无人车前方道路的拍摄图像。
在实际应用中，为了保证训练数据的丰富性，避免训练数据的偶然性和训练数据之间的关联性，应尽量选取不同时段内、不同环境下的训练图像。
S302，基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息。
可选的，预设的决策网络可以为卷积神经网络。
示例性的，预设的决策网络可以由6层卷积层、3层池化层以及2层全连接层构成。
S303，利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型，其中，每组训练数据中包括一幅训练图像和所述训练图像对应的驾驶信息。
在每次的迭代训练的过程中，利用一组训练数据进行训练。为了保证迭代训练的有效性，每次迭代训练选取的训练数据均不同。
在一个实施例中，所述训练图像对应的驾驶信息包括目标驾驶动作、所述目标驾驶动作对应的环境回报值和动作图像。
参见图4，为本申请实施例提供的驾驶信息的获取方法的流程示意图。如图4所示，步骤S302中基于预设的决策模型，获取一幅训练图像对应的驾驶信息的方法，可以包括以下步骤：
S401，将所述训练图像输入到所述预设的决策模型中得到第一输出结果，所述第一输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第一输出值。
驾驶候选驾驶动作可以包括缓慢驾驶、较快加速、缓慢减速、较快减速、紧急刹车等。
S402，将所述第一输出结果中最大的第一输出值对应的候选驾驶动作确定为所述无人车的目标驾驶动作。
S403，在所述无人车执行所述目标驾驶动作之后，获取所述无人车前方道路的拍摄图像得到动作图像，并计算所述目标驾驶动作对应的环境回报值。
在本申请实施例中，S403中计算所述目标驾驶动作对应的环境回报值，可以包括：
S41，获取所述无人车执行所述目标驾驶动作之后的驾驶速度，并根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值。
可选的，可以通过公式
Figure PCTCN2019120559-appb-000003
计算所述行驶回报值。
其中，R velocity为所述行驶回报值，v t为所述无人车执行所述目标驾驶动作之后的驾驶速度，v min为所述最小限速，v max为所述最大限速。
S42，获取所述无人车执行所述目标驾驶动作之后的驾驶状态，并确定所述驾驶状态对应的预设碰撞回报值，所述驾驶状态包括碰撞和非碰撞。
示例性的，当驾驶状态为碰撞时，预设碰撞回报值为-10；当驾驶状态为非碰撞时，预设碰撞回报值为0。实际应用中，可以根据实验经验定义预设碰撞回报值，在此不做具体限定。
S43，根据所述目标驾驶动作计算减速回报值。
可选的，S43可以包括：
I、若所述目标驾驶动作是减速，则通过公式
Figure PCTCN2019120559-appb-000004
计算所述减速回报值，其中，R stop为所述减速回报值，d为剩余驾驶距离，K为第一预设值，所述剩余驾驶距离为所述无人车执行所述目标驾驶动作之后的位置与预设目标位置之间的距离。
II、若所述目标驾驶动作不是减速，则将所述减速回报值设置为第二预设值。
实际应用中，也可把刹车动作归属于减速动作中，即当目标驾驶动作是减速或刹车时，均需要通过公式
Figure PCTCN2019120559-appb-000005
计算所述减速回报值；而其他目标驾驶动作则对应第二预设值。
S44，根据所述行驶回报值、所述预设碰撞回报值和所述减速回报值，计算所述目标驾驶动作对应的环境回报值。
可选的，可以将行驶回报值、预设碰撞回报值和减速回报值直接相加得到环境回报值。
当然，也可以将行驶回报值、预设碰撞回报值和减速回报值进行加权求和得到环境回报值。具体的计算方式不做限定。
在一个实施例中，参见图5，为本申请实施例提供的迭代训练过程的示意图。如图5所示，步骤S303中，进行第i次迭代训练的过程包括：
S501，将第i组训练数据中的训练图像输入到第i-1次迭代训练后的决策模型中得到第二输出结果。
其中，第二输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第二输出值。
S502，将所述第i组训练数据中的动作图像输入到参考模型中得到第三输出结果。
其中，第三输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第三输出值。第三输出结果中的候选驾驶动作与第二输出结果中的候选驾驶动作相同。
S503，利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练。
其中，当i≤N时，所述参考模型为预设的决策模型；当i>N时，所述参考模型为第i-1次迭代训练后的决策模型；N为正整数。
在本实施例中，引入了参考模型。示例性的，驾驶N为3，当进行前3次迭代训练时，参考模型均为预设的决策模型，即初始定义的决策模型。当进行第4次迭代训练时，参考模型为进行了第1次迭代训练后的决策模型；当进行第5次迭代训练时，参考模型为进行了第2次迭代训练后的决策模型；以此类推。实际上，参考模型为与当前迭代训练过程中的决策模型相差了N次迭代训练过程的决策模型。利用这样的训练方法，可以有效保证迭代训练过程的稳定性。
在步骤S503中，利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练，可以包括以下步骤：
A、获取第二输出结果中与第i组训练数据中的目标驾驶动作对应的第二输出值得到第一计算数值，获取第三输出结果中最大的第三输出值得到第二计算数值，并利用第一计算数值、第二计算数值和第i组训练数据中的环境回报值计算第i-1次迭代训练后的决策模型的误差值。
在本申请实施例中，可以通过下式计算误差值：
L(θ)＝E[(TargetQ-Q(s,a；θ)) 2]；
其中，L(θ)为误差值，Q(s,a；θ)为第一计算数值，s表示训练图像，a表示目标驾驶动作，θ表示第i-1次迭代训练后的决策模型的权值；式中：
TargetQ＝r+γmax a′Q(s′,a′；θ′)；
其中，r为环境回报值，γ为折扣因子，max a′Q(s′,a′；θ′)为第二计算数值，s′表示动作图像，a′表示第二计算数值对应的候选驾驶动作，θ′表示预设的决策模型的权值。
B、若误差值小于或等于预设阈值，则将第i-1次迭代训练后的决策模型作为训练后的决策模型。
C、若误差值大于预设阈值，则对第i-1次迭代训练后的决策模型进行权值更新，并在更新后继续进行下一次迭代训练。
实际应用中，可以利用梯度下降法进行权值更新。
本申请实施例通过获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值；训练后的决策模型能够较准确地反映出无人车前方道路环境与决策结果之间的映射关系，避免了时变不确定性因素造成的影响；将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。通过上述方法，有效提高了无人车的驾驶决策能力，进而保证了无人车驾驶过程的安全性。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
对应于上文实施例所述的无人车驾驶决策方法，图6示出了本申请实施例提供的无人车驾驶决策装置的结构框图，为了便于说明，仅示出了与本申请实施例相关的部分。
参照图6，该装置包括：
待决策图像获取单元61，用于获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像。
决策单元62，用于将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值。
结果确定单元63，用于将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。
可选的，装置6还包括：
训练图像获取单元64，用于在将所述待决策图像输入到训练后的决策模型中得到决策结果之前，获取多幅训练图像，其中，所述训练图像为所述无人车前方道路的拍摄图像。
驾驶信息获取单元65，用于基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息。
决策模型训练单元66，用于利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型，其中，每组训练数据中包括一幅训练图像和所述训练图像对应的驾驶信息。
可选的，所述训练图像对应的驾驶信息包括目标驾驶动作、所述目标驾驶动作对应的环境回报值和动作图像。
可选的，驾驶信息获取单元65包括：
第一输出结果模块，用于对于每幅训练图像，将所述训练图像输入到所述预设的决策模型中得到第一输出结果，所述第一输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第一输出值。
目标动作模块，用于将所述第一输出结果中最大的第一输出值对应的候选驾驶动作确定为所述无人车的目标驾驶动作。
计算模块，用于在所述无人车执行所述目标驾驶动作之后，获取所述无人车前方道路的拍摄图像得到动作图像，并计算所述目标驾驶动作对应的环境回报值。
可选的，计算模块包括：
速度获取子模块，用于获取所述无人车执行所述目标驾驶动作之后的驾驶速度，并根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值。
状态获取子模块，用于获取所述无人车执行所述目标驾驶动作之后的驾驶状态，并确定所述驾驶状态对应的预设碰撞回报值，所述驾驶状态包括碰撞和非碰撞。
第一计算子模块，用于根据所述目标驾驶动作计算减速回报值。
第二计算子模块，用于根据所述行驶回报值、所述预设碰撞回报值和所述减速回报值，计算所述目标驾驶动作对应的环境回报值。
可选的，速度获取子模块，还用于通过公式
Figure PCTCN2019120559-appb-000006
计算所述行驶回报值。
其中，R velocity为所述行驶回报值，v t为所述无人车执行所述目标驾驶动作之后的驾驶 速度，v min为所述最小限速，v max为所述最大限速
可选的，第一计算子模块，还用于若所述目标驾驶动作是减速，则通过公式
Figure PCTCN2019120559-appb-000007
计算所述减速回报值，其中，R stop为所述减速回报值，d为剩余驾驶距离，K为第一预设值，所述剩余驾驶距离为所述无人车执行所述目标驾驶动作之后的位置与预设目标位置之间的距离；若所述目标驾驶动作不是减速，则将所述减速回报值设置为第二预设值。
可选的，决策模型训练单元66包括：
第二输出结果模块，用于将第i组训练数据中的训练图像输入到第i-1次迭代训练后的决策模型中得到第二输出结果。
第三输出结果模块，用于将所述第i组训练数据中的动作图像输入到参考模型中得到第三输出结果。
迭代训练模块，用于利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练。
其中，当i≤N时，所述参考模型为预设的决策模型；当i>N时，所述参考模型为第i-1次迭代训练后的决策模型；N为正整数。
对应于上文实施例所述的无人车驾驶决策方法，图6示出了本申请实施例提供的无人车驾驶决策装置的结构框图，为了便于说明，仅示出了与本申请实施例相关的部分。
需要说明的是，上述装置/单元之间的信息交互、执行过程等内容，由于与本申请方法实施例基于同一构思，其具体功能及带来的技术效果，具体可参见方法实施例部分，此处不再赘述。
另外，图6所示的装置可以是内置于现有的终端设备内的软件单元、硬件单元、或软硬结合的单元，也可以作为独立的挂件集成到所述终端设备中，还可以作为独立的终端设备存在。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。实施例中的各功能单元、模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中，上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。另外，各功能单元、模块的具体名称也只是为了便于相互区分，并不用于限制本申请的保护范围。上述系统中单元、模块的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
图7为本申请一实施例提供的无人车的结构示意图。如图7所示，该实施例的无人车7 包括：至少一个处理器70(图7中仅示出一个)处理器、存储器71以及存储在所述存储器71中并可在所述至少一个处理器70上运行的计算机程序72，所述处理器70执行所述计算机程序72时实现上述任意各个无人车驾驶决策方法实施例中的步骤。
该无人车可包括，但不仅限于，处理器、存储器。本领域技术人员可以理解，图7仅仅是无人车7的举例，并不构成对无人车7的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件，例如还可以包括输入输出设备、网络接入设备等。
所称处理器70可以是中央处理单元(Central Processing Unit，CPU)，该处理器70还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
所述存储器71在一些实施例中可以是所述无人车7的内部存储单元，例如无人车7的硬盘或内存。所述存储器71在另一些实施例中也可以是所述无人车7的外部存储设备，例如所述无人车7上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。进一步地，所述存储器71还可以既包括所述无人车7的内部存储单元也包括外部存储设备。所述存储器71用于存储操作系统、应用程序、引导装载程序(BootLoader)、数据以及其他程序等，例如所述计算机程序的程序代码等。所述存储器71还可以用于暂时地存储已经输出或者将要输出的数据。
本申请实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现可实现上述各个方法实施例中的步骤。
本申请实施例提供了一种计算机程序产品，当计算机程序产品在移动终端上运行时，使得移动终端执行时实现可实现上述各个方法实施例中的步骤。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请实现上述实施例方法中的全部或部分流程，可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一计算机可读存储介质中，该计算机程序在被处理器执行时，可实现上述各个方法实施例的步骤。其中，所述计算机程序包括计算机程序代码，所述计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。所述计算机可读介质至少可以包括：能够将计算机程序代码携带到无人车的任何实体或装置、记录介质、计算机存储器、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、电载波信号、电信信号以及软件分发介质。例如U盘、移动硬盘、磁碟或者光盘等。在某些司法管辖区，根据立法和专利实践，计算机可读介质不可以是电载波信号和 电信信号。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
在本申请所提供的实施例中，应该理解到，所揭露的装置/网络设备和方法，可以通过其它的方式实现。例如，以上所描述的装置/网络设备实施例仅仅是示意性的，例如，所述模块或单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通讯连接可以是通过一些接口，装置或单元的间接耦合或通讯连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种无人车驾驶决策方法，其特征在于，包括：
    获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；
    将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值；
    将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。
  2. 如权利要求1所述的无人车驾驶决策方法，其特征在于，在将所述待决策图像输入到训练后的决策模型中得到决策结果之前，所述方法还包括：
    获取多幅训练图像，其中，所述训练图像为所述无人车前方道路的拍摄图像；
    基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息；
    利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型，其中，每组训练数据中包括一幅训练图像和所述训练图像对应的驾驶信息。
  3. 如权利要求2所述的无人车驾驶决策方法，其特征在于，所述训练图像对应的驾驶信息包括目标驾驶动作、所述目标驾驶动作对应的环境回报值和动作图像；
    所述基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息，包括：
    对于每幅训练图像，将所述训练图像输入到所述预设的决策模型中得到第一输出结果，所述第一输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第一输出值；
    将所述第一输出结果中最大的第一输出值对应的候选驾驶动作确定为所述无人车的目标驾驶动作；
    在所述无人车执行所述目标驾驶动作之后，获取所述无人车前方道路的拍摄图像得到动作图像，并计算所述目标驾驶动作对应的环境回报值。
  4. 如权利要求3所述的无人车驾驶决策方法，其特征在于，所述计算所述目标驾驶动作对应的环境回报值，包括：
    获取所述无人车执行所述目标驾驶动作之后的驾驶速度，并根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值；
    获取所述无人车执行所述目标驾驶动作之后的驾驶状态，并确定所述驾驶状态对应的预设碰撞回报值，所述驾驶状态包括碰撞和非碰撞；
    根据所述目标驾驶动作计算减速回报值；
    根据所述行驶回报值、所述预设碰撞回报值和所述减速回报值，计算所述目标驾驶动作对应的环境回报值。
  5. 如权利要求4所述的无人车驾驶决策方法，其特征在于，所述根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值，包括：
    通过公式
    Figure PCTCN2019120559-appb-100001
    计算所述行驶回报值；
    其中，R velocity为所述行驶回报值，v t为所述无人车执行所述目标驾驶动作之后的驾驶速度，v min为所述最小限速，v max为所述最大限速。
  6. 如权利要求4所述的无人车驾驶决策方法，其特征在于，所述根据所述目标驾驶动作计算减速回报值，包括：
    若所述目标驾驶动作是减速，则通过公式
    Figure PCTCN2019120559-appb-100002
    计算所述减速回报值，其中，R stop为所述减速回报值，d为剩余驾驶距离，K为第一预设值，所述剩余驾驶距离为所述无人车执行所述目标驾驶动作之后的位置与预设目标位置之间的距离；
    若所述目标驾驶动作不是减速，则将所述减速回报值设置为第二预设值。
  7. 如权利要求2所述的无人车驾驶决策模型的训练方法，其特征在于，在所述利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型的步骤中，进行第i次迭代训练的过程包括：
    将第i组训练数据中的训练图像输入到第i-1次迭代训练后的决策模型中得到第二输出结果；
    将所述第i组训练数据中的动作图像输入到参考模型中得到第三输出结果；
    利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练；
    其中，当i≤N时，所述参考模型为预设的决策模型；当i>N时，所述参考模型为第i-1次迭代训练后的决策模型；N为正整数。
  8. 一种无人车驾驶决策装置，其特征在于，包括：
    待决策图像获取单元，用于获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；
    决策单元，用于将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值；
    结果确定单元，用于将所述决策结果中最大的决策值对应的候选驾驶动作确定为 所述无人车在下一决策时刻的目标驾驶动作。
  9. 如权利要求8所述的无人车驾驶决策装置，其特征在于，所述装置还包括：
    训练图像获取单元，用于在将所述待决策图像输入到训练后的决策模型中得到决策结果之前，获取多幅训练图像，其中，所述训练图像为所述无人车前方道路的拍摄图像；
    驾驶信息获取单元，用于基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息；
    决策模型训练单元，用于利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型，其中，每组训练数据中包括一幅训练图像和所述训练图像对应的驾驶信息。
  10. 如权利要求9所述的无人车驾驶决策装置，其特征在于，所述训练图像对应的驾驶信息包括目标驾驶动作、所述目标驾驶动作对应的环境回报值和动作图像；
    所述驾驶信息获取单元包括：
    第一输出结果模块，用于对于每幅训练图像，将所述训练图像输入到所述预设的决策模型中得到第一输出结果，所述第一输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第一输出值；
    目标动作模块，用于将所述第一输出结果中最大的第一输出值对应的候选驾驶动作确定为所述无人车的目标驾驶动作；
    计算模块，用于在所述无人车执行所述目标驾驶动作之后，获取所述无人车前方道路的拍摄图像得到动作图像，并计算所述目标驾驶动作对应的环境回报值。
  11. 如权利要求10所述的无人车驾驶决策装置，其特征在于，所述计算模块包括：
    速度获取子模块，用于获取所述无人车执行所述目标驾驶动作之后的驾驶速度，并根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值；
    状态获取子模块，用于获取所述无人车执行所述目标驾驶动作之后的驾驶状态，并确定所述驾驶状态对应的预设碰撞回报值，所述驾驶状态包括碰撞和非碰撞；
    第一计算子模块，用于根据所述目标驾驶动作计算减速回报值；
    第二计算子模块，用于根据所述行驶回报值、所述预设碰撞回报值和所述减速回报值，计算所述目标驾驶动作对应的环境回报值。
  12. 如权利要求9所述的无人车驾驶决策装置，其特征在于，所述决策模型训练单元包括：
    第二输出结果模块，用于将第i组训练数据中的训练图像输入到第i-1次迭代训练 后的决策模型中得到第二输出结果；
    第三输出结果模块，用于将所述第i组训练数据中的动作图像输入到参考模型中得到第三输出结果；
    迭代训练模块，用于利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练；
    其中，当i≤N时，所述参考模型为预设的决策模型；当i>N时，所述参考模型为第i-1次迭代训练后的决策模型；N为正整数。
  13. 一种无人车，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如下方法：
    获取当前决策时刻所述无人车前方道路的拍摄图像，得到待决策图像；
    将所述待决策图像输入到训练后的决策模型中得到决策结果，所述决策结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的决策值；
    将所述决策结果中最大的决策值对应的候选驾驶动作确定为所述无人车在下一决策时刻的目标驾驶动作。
  14. 如权利要求13所述的无人车，其特征在于，在将所述待决策图像输入到训练后的决策模型中得到决策结果之前，所述方法还包括：
    获取多幅训练图像，其中，所述训练图像为所述无人车前方道路的拍摄图像；
    基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息；
    利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型，其中，每组训练数据中包括一幅训练图像和所述训练图像对应的驾驶信息。
  15. 如权利要求14所述的无人车，其特征在于，所述训练图像对应的驾驶信息包括目标驾驶动作、所述目标驾驶动作对应的环境回报值和动作图像；
    所述基于预设的决策模型，分别获取每幅训练图像对应的驾驶信息，包括：
    对于每幅训练图像，将所述训练图像输入到所述预设的决策模型中得到第一输出结果，所述第一输出结果包括多个候选驾驶动作、以及各个候选驾驶动作对应的第一输出值；
    将所述第一输出结果中最大的第一输出值对应的候选驾驶动作确定为所述无人车的目标驾驶动作；
    在所述无人车执行所述目标驾驶动作之后，获取所述无人车前方道路的拍摄图像得到动作图像，并计算所述目标驾驶动作对应的环境回报值。
  16. 如权利要求15所述的无人车，其特征在于，所述计算所述目标驾驶动作对应的环境回报值，包括：
    获取所述无人车执行所述目标驾驶动作之后的驾驶速度，并根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值；
    获取所述无人车执行所述目标驾驶动作之后的驾驶状态，并确定所述驾驶状态对应的预设碰撞回报值，所述驾驶状态包括碰撞和非碰撞；
    根据所述目标驾驶动作计算减速回报值；
    根据所述行驶回报值、所述预设碰撞回报值和所述减速回报值，计算所述目标驾驶动作对应的环境回报值。
  17. 如权利要求16所述的无人车，其特征在于，所述根据所述驾驶速度、预设的最小限速和预设的最大限速计算行驶回报值，包括：
    通过公式
    Figure PCTCN2019120559-appb-100003
    计算所述行驶回报值；
    其中，R velocity为所述行驶回报值，v t为所述无人车执行所述目标驾驶动作之后的驾驶速度，v min为所述最小限速，v max为所述最大限速。
  18. 如权利要求16所述的无人车，其特征在于，所述根据所述目标驾驶动作计算减速回报值，包括：
    若所述目标驾驶动作是减速，则通过公式
    Figure PCTCN2019120559-appb-100004
    计算所述减速回报值，其中，R stop为所述减速回报值，d为剩余驾驶距离，K为第一预设值，所述剩余驾驶距离为所述无人车执行所述目标驾驶动作之后的位置与预设目标位置之间的距离；
    若所述目标驾驶动作不是减速，则将所述减速回报值设置为第二预设值。
  19. 如权利要求14所述的无人车，其特征在于，在所述利用各组训练数据对所述预设的决策模型进行迭代训练得到训练后的决策模型的步骤中，进行第i次迭代训练的过程包括：
    将第i组训练数据中的训练图像输入到第i-1次迭代训练后的决策模型中得到第二输出结果；
    将所述第i组训练数据中的动作图像输入到参考模型中得到第三输出结果；
    利用所述第二输出结果和所述第三输出结果对第i-1次迭代训练后的决策模型进行训练；
    其中，当i≤N时，所述参考模型为预设的决策模型；当i>N时，所述参考模型为 第i-1次迭代训练后的决策模型；N为正整数。
  20. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至7任一项所述的方法。
PCT/CN2019/120559 2019-11-14 2019-11-25 无人车驾驶决策方法、无人车驾驶决策装置及无人车 WO2021093011A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911113303.3 2019-11-14
CN201911113303.3A CN111079533B (zh) 2019-11-14 2019-11-14 无人车驾驶决策方法、无人车驾驶决策装置及无人车

Publications (1)

Publication Number Publication Date
WO2021093011A1 true WO2021093011A1 (zh) 2021-05-20

Family

ID=70310991

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/120559 WO2021093011A1 (zh) 2019-11-14 2019-11-25 无人车驾驶决策方法、无人车驾驶决策装置及无人车

Country Status (2)

Country Link
CN (1) CN111079533B (zh)
WO (1) WO2021093011A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114115292A (zh) * 2021-12-21 2022-03-01 上海智蕙林医疗科技有限公司 行驶控制方法、系统、存储介质和计算机设备
CN114261400A (zh) * 2022-01-07 2022-04-01 京东鲲鹏(江苏)科技有限公司 一种自动驾驶决策方法、装置、设备和存储介质
CN115771506A (zh) * 2022-11-17 2023-03-10 清华大学 基于乘员风险认知的车辆行驶策略的确定方法和装置

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111731326B (zh) * 2020-07-02 2022-06-21 知行汽车科技（苏州）有限公司 避障策略确定方法、装置及存储介质
CN112418237A (zh) * 2020-12-07 2021-02-26 苏州挚途科技有限公司 车辆驾驶决策方法、装置及电子设备
CN113635896B (zh) * 2021-07-31 2023-06-20 华为技术有限公司 一种驾驶行为确定方法及其相关设备

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100168998A1 (en) * 2008-12-26 2010-07-01 Toyota Jidosha Kabushiki Kaisha Driving assistance device and driving assistance method
CN107392189A (zh) * 2017-09-05 2017-11-24 百度在线网络技术（北京）有限公司 用于确定无人车的驾驶行为的方法和装置
CN108062569A (zh) * 2017-12-21 2018-05-22 东华大学 一种基于红外和雷达的无人车驾驶决策方法
CN108791302A (zh) * 2018-06-25 2018-11-13 大连大学 驾驶员行为建模系统
CN108891421A (zh) * 2018-06-25 2018-11-27 大连大学 一种构建驾驶策略的方法
CN109131348A (zh) * 2018-07-24 2019-01-04 大连理工大学 一种基于生成式对抗网络的智能车驾驶决策方法
CN109747659A (zh) * 2018-11-26 2019-05-14 北京汽车集团有限公司 车辆驾驶的控制方法和装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100168998A1 (en) * 2008-12-26 2010-07-01 Toyota Jidosha Kabushiki Kaisha Driving assistance device and driving assistance method
CN107392189A (zh) * 2017-09-05 2017-11-24 百度在线网络技术（北京）有限公司 用于确定无人车的驾驶行为的方法和装置
CN108062569A (zh) * 2017-12-21 2018-05-22 东华大学 一种基于红外和雷达的无人车驾驶决策方法
CN108791302A (zh) * 2018-06-25 2018-11-13 大连大学 驾驶员行为建模系统
CN108891421A (zh) * 2018-06-25 2018-11-27 大连大学 一种构建驾驶策略的方法
CN109131348A (zh) * 2018-07-24 2019-01-04 大连理工大学 一种基于生成式对抗网络的智能车驾驶决策方法
CN109747659A (zh) * 2018-11-26 2019-05-14 北京汽车集团有限公司 车辆驾驶的控制方法和装置

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114115292A (zh) * 2021-12-21 2022-03-01 上海智蕙林医疗科技有限公司 行驶控制方法、系统、存储介质和计算机设备
CN114261400A (zh) * 2022-01-07 2022-04-01 京东鲲鹏(江苏)科技有限公司 一种自动驾驶决策方法、装置、设备和存储介质
CN115771506A (zh) * 2022-11-17 2023-03-10 清华大学 基于乘员风险认知的车辆行驶策略的确定方法和装置

Also Published As

Publication number Publication date
CN111079533A (zh) 2020-04-28
CN111079533B (zh) 2023-04-07

Similar Documents

Publication Publication Date Title
WO2021093011A1 (zh) 无人车驾驶决策方法、无人车驾驶决策装置及无人车
CN108038474B (zh) 人脸检测方法、卷积神经网络参数的训练方法、装置及介质
WO2020207190A1 (zh) 一种三维信息确定方法、三维信息确定装置及终端设备
CN109583332B (zh) 人脸识别方法、人脸识别系统、介质及电子设备
WO2018166114A1 (zh) 图片识别的方法、系统、电子装置及介质
CN110245579B (zh) 人流密度预测方法及装置、计算机设备及可读介质
CN111797893A (zh) 一种神经网络的训练方法、图像分类系统及相关设备
CN112084856A (zh) 一种人脸姿态检测方法、装置、终端设备及存储介质
CN112085701B (zh) 一种人脸模糊度检测方法、装置、终端设备及存储介质
CN110263628B (zh) 障碍物检测方法、装置、电子设备以及存储介质
CN111612841A (zh) 目标定位方法及装置、移动机器人及可读存储介质
US11568212B2 (en) Techniques for understanding how trained neural networks operate
CN110686676A (zh) 机器人重定位方法、装置及机器人
CN112232506A (zh) 网络模型训练方法、图像目标识别方法、装置和电子设备
WO2023138537A1 (zh) 一种图像处理方法、装置、终端设备及存储介质
CN107993239B (zh) 一种计算单目图像的深度次序的方法和装置
CN111731326A (zh) 避障策略确定方法、装置及存储介质
CN110060264B (zh) 神经网络训练方法、视频帧处理方法、装置及系统
CN113239746B (zh) 电动车检测方法、装置、终端设备及计算机可读存储介质
CN111627029B (zh) 图像实例分割结果的获取方法及装置
CN112966687A (zh) 图像分割模型训练方法、装置及通信设备
CN112488054A (zh) 一种人脸识别方法、装置、终端设备及存储介质
CN111967406A (zh) 人体关键点检测模型生成方法、系统、设备和存储介质
CN114821513B (zh) 一种基于多层网络的图像处理方法及装置、电子设备
CN112749727A (zh) 本地服务器、影像辨识系统及其更新方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19952208

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 19.09.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19952208

Country of ref document: EP

Kind code of ref document: A1