WO2024082183A1 - 参数调节方法、装置以及智能终端 - Google Patents

参数调节方法、装置以及智能终端 Download PDF

Info

Publication number
WO2024082183A1
WO2024082183A1 PCT/CN2022/126256 CN2022126256W WO2024082183A1 WO 2024082183 A1 WO2024082183 A1 WO 2024082183A1 CN 2022126256 W CN2022126256 W CN 2022126256W WO 2024082183 A1 WO2024082183 A1 WO 2024082183A1
Authority
WO
WIPO (PCT)
Prior art keywords
parameter set
sub
isp
image
isp parameter
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2022/126256
Other languages
English (en)
French (fr)
Inventor
杨攀
张兴亚
邱守谦
王超
崔泽波
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies Co Ltd
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 Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Priority to PCT/CN2022/126256 priority Critical patent/WO2024082183A1/zh
Priority to CN202280100051.4A priority patent/CN119856204A/zh
Publication of WO2024082183A1 publication Critical patent/WO2024082183A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N23/00Cameras or camera modules comprising electronic image sensors; Control thereof
    • H04N23/60Control of cameras or camera modules

Definitions

  • the present application relates to the field of image processing technology, and in particular to a parameter adjustment method, device and intelligent terminal.
  • image recognition has been widely used in various fields, such as transportation, intelligent security, autonomous driving, mobile terminals, etc.
  • image recognition is the core of autonomous driving.
  • Image quality can directly affect the accuracy of image recognition, and image signal processing (ISP) parameters determine the quality of image.
  • ISP parameters can ultimately determine the accuracy of image recognition.
  • ISP parameters are adjusted manually. That is, engineers adjust ISP parameters according to experience, apply the adjusted ISP parameters to image signal processing, and determine the image quality and subjective effect of the image obtained according to the ISP parameters. If the image obtained by processing with the adjusted ISP parameters does not meet the set indicators of image quality and subjective effect, the engineer adjusts the ISP parameters again based on experience until the image obtained according to the adjusted ISP parameters meets the set indicators of image quality and subjective effect, and the ISP parameters that meet the set indicators are used as the final ISP parameters.
  • the above method requires engineers to repeatedly adjust the ISP parameters, which will result in high labor costs and low efficiency.
  • the present application provides a parameter adjustment method, device and intelligent terminal, which can autonomously adjust ISP parameters without investing a lot of manpower costs, thereby improving the efficiency of ISP parameter adjustment, and can also improve the image recognition rate while ensuring human eye vision.
  • the present application provides a parameter adjustment method, in which a second ISP parameter set is determined based on a training image set and a first image signal processing ISP parameter set, the first ISP parameter set being the ISP parameter set corresponding to the smart terminal, a test image set is processed based on the second ISP parameter set, and a first quality score and a first recognition rate are determined based on the processed test image set, the first quality score indicating the image quality of the processed test image set, and the first recognition rate indicating the image recognition status of the processed test image set, if the first quality score is greater than the first reference quality score and the first recognition rate is greater than the first reference recognition rate, the ISP parameter set corresponding to the smart terminal is adjusted from the first ISP parameter set to the second ISP parameter set, the first reference quality score indicating the image quality of the test image set after being processed by the first ISP parameter set, and the first reference recognition rate indicating the image recognition status of the test image set after being processed by the first ISP parameter set.
  • the present application automatically determines the second ISP parameters based on the training image set and the first ISP parameter set, without manually adjusting the ISP parameters repeatedly, thus reducing the labor cost.
  • the first quality score represents the image quality of the test image set after being processed by the second ISP parameter set
  • the first recognition rate represents the image recognition of the test image set after being processed by the second ISP parameter set
  • the first reference quality score represents the image quality of the test image set after being processed by the first ISP parameter set
  • the first reference recognition rate represents the image recognition of the test image set after being processed by the first ISP parameter set.
  • the first quality score is greater than the first reference quality score, and the first recognition rate is greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter set is better than that of the first ISP parameter set, and the image recognition rate after being processed by the second ISP parameter set is also higher than that of the first ISP parameter set.
  • adjusting the ISP parameter set corresponding to the smart terminal from the first ISP parameter set to the second ISP parameter set can not only ensure that the quality of the image obtained by the smart terminal according to the second ISP parameter set is better than the image quality obtained according to the first ISP parameter set, but also ensure that the recognition rate of the image obtained by the smart terminal according to the second ISP parameter set is also better than the recognition rate of the image obtained according to the first ISP parameter set.
  • the method provided in the present application can not only autonomously adjust ISP parameters and improve the efficiency of ISP parameter adjustment without investing a lot of manpower costs, but also improve the image recognition rate while ensuring human eye vision.
  • the training image set includes N sub-training sets
  • the first ISP parameter set includes M parameters
  • each of the N sub-training sets corresponds to a scene
  • N and M are both integers greater than 1.
  • the following operations are performed for each of the N sub-training sets: one of the sub-training sets is used as a target sub-training set, and a first sub-parameter set corresponding to a target scene in the first ISP parameter set is adjusted to obtain a second sub-parameter set corresponding to a target scene in a second ISP parameter set, the target scene is a scene corresponding to the target sub-training set, the first sub-parameter set includes at least one parameter of the M parameters, and the second sub-parameter set includes at least one parameter of the M parameters.
  • the N sub-training sets have one-to-one corresponding N scenes, each of the N scenes corresponds to a sub-parameter set in the first ISP parameter set, and each sub-parameter set includes at least one parameter of the M parameters.
  • the sub-training set is called a target sub-training set
  • the scene corresponding to the target sub-training set is called a target scene
  • the sub-parameter set corresponding to the target scene in the first ISP parameter is called a first sub-parameter set.
  • the sub-parameter set corresponding to the target scene in the second ISP parameter i.e., the second sub-parameter set
  • the sub-parameter set corresponding to each scene in the second ISP parameter set can be obtained.
  • the implementation process of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set to obtain the second sub-parameter set corresponding to the target scene in the second ISP parameter set includes: adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set to multiple candidate sub-parameter sets, keeping the remaining parameters unchanged to obtain multiple candidate ISP parameter sets, and determining multiple candidate quality scores corresponding to the target sub-training set based on the multiple candidate ISP parameter sets, the multiple candidate quality scores indicating the image quality of the target sub-training set after being processed by the multiple candidate ISP parameter sets, and determining the parameter corresponding to the target scene in the candidate ISP parameter set corresponding to the maximum candidate quality score among the multiple candidate quality scores as the second sub-parameter set.
  • the parameter corresponding to the target scene in the candidate ISP parameter set corresponding to the maximum candidate quality score among multiple candidate quality scores is determined as the second sub-parameter set. In this way, it can be ensured that the image processing effect of the second sub-parameter set finally determined is the best among the multiple candidate ISP parameter sets, further ensuring the adjustment effect of the ISP parameter set.
  • the test image set includes N sub-test sets, which correspond one-to-one to the N sub-training sets.
  • the intelligent terminal may also determine a second quality score, the second quality score indicating the image quality of the target sub-training set after being processed by the first ISP parameter set.
  • the first sub-parameter set corresponding to the target scene in the first ISP parameter set is adjusted in the above manner, the second reference quality score indicating the image quality of the target sub-test set after being processed by the first ISP parameter set, and the target sub-test set is the sub-test set corresponding to the target scene. If the second quality score is greater than or equal to the second reference quality score, the first sub-parameter set corresponding to the target scene in the first ISP parameter set is not adjusted.
  • the second quality score is less than the second reference quality score, it means that the processing effect of the images in the target sub-training set by the first ISP parameter set is not good, and the ISP parameters corresponding to the target scene need to be adjusted, so the step of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set is performed. If the second quality score is greater than or equal to the second reference quality score, it means that the processing effect of the images in the target sub-training set by the first ISP parameter set is good, and the ISP parameters corresponding to the target scene do not need to be adjusted, so the step of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set is not performed.
  • N sub-parameter sets can be obtained. Since each sub-parameter set includes at least one parameter of the M parameters, the union of the N sub-parameter sets may include each parameter of the M parameters, or may only include some parameters of the M parameters. In different cases, the methods for determining the second ISP parameter set based on the N sub-parameter sets are different, which will be introduced below.
  • the union of the N sub-parameter sets includes each parameter in the M parameters.
  • a second ISP parameter set can be determined based on the N sub-parameter sets.
  • the sub-parameter sets corresponding to different scenarios may or may not have intersections.
  • the N sub-parameter sets can be directly merged to obtain a second ISP parameter set.
  • the parameters with intersections are called first-category parameters, and the parameters without intersections are called second-category parameters.
  • the parameter values of the first-category parameters in the N sub-parameter sets are combined to obtain multiple parameter value combinations, and the multiple parameter value combinations are merged with the parameter values of the second-category parameters in the N sub-parameter sets to obtain multiple second ISP parameter sets.
  • the union of the N sub-parameter sets includes some parameters in the M parameters.
  • the second ISP parameter set is determined based on the N sub-parameter sets and the first ISP parameter set.
  • the sub-parameter sets corresponding to different scenarios may or may not have intersections.
  • the N sub-parameter sets can be directly merged to obtain a merged parameter set, and the merged parameter set can be merged again with other parameters in the first ISP parameter set except the merged parameter set to obtain a second ISP parameter set.
  • the parameters with intersections are called first-category parameters, and the parameters without intersections are called second-category parameters.
  • the parameter values of the first-category parameters in the N sub-parameter sets are combined to obtain multiple parameter value combinations, and the multiple parameter value combinations are merged with the parameter values of the second-category parameters in the N sub-parameter sets to obtain multiple merged parameter sets.
  • the merged parameter set is merged again with other parameters in the first ISP parameter set except the merged parameter set to obtain multiple second ISP parameter sets.
  • the smart terminal before determining the second ISP parameter set based on the training image set and the first ISP parameter set, can also obtain an image corresponding to the current environment, process the image corresponding to the current environment based on the first ISP parameter set, and perform scene recognition based on the processed image to obtain a scene recognition result, which indicates whether there is a key scene in the current environment. If the scene recognition result indicates that there is a key scene in the current environment, the image corresponding to the current environment is stored in the training image set.
  • the training image set includes N sub-training sets, each of which corresponds to a scene. If the scene indicated by the scene recognition result is a key scene, the image corresponding to the current environment is stored in the sub-training set of the corresponding scene.
  • the images in the training image set can be images taken by the camera in real time. Since different users have different behavioral habits, the images taken by the camera are also different. Therefore, adjusting the ISP parameter set based on the images taken by the camera in real time can make the final adjusted ISP parameters adaptive.
  • the images in the training image set can also be images obtained by other means.
  • the images in the training image set are images that have not been processed by the ISP parameter set.
  • the adjusted ISP parameter set can be applied to the images that have not been processed by the ISP parameter set, so as to determine the quality of the adjusted ISP parameter set.
  • the number of the second ISP parameter sets may be one or more. If there are multiple second ISP parameter sets, the test image set is processed based on each second ISP parameter set in the multiple second ISP parameter sets, thereby obtaining multiple processed test image sets.
  • the processed test image set may be one or more.
  • the methods for determining the first quality score and the first recognition rate are different, which will be introduced below.
  • the first quality score and the first recognition rate are determined based on the processed test image set.
  • the quality scores and recognition rates of the multiple processed test image sets are determined, the maximum quality score among the quality scores of the multiple processed test image sets is used as the first quality score, and the recognition rate corresponding to the maximum quality score is used as the first recognition rate.
  • the ISP parameter set corresponding to the smart terminal is not adjusted from the first ISP parameter set to the second ISP parameter set, that is, the ISP parameter set corresponding to the smart terminal remains unchanged.
  • the first quality score is greater than the first reference quality score and the first recognition rate is greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter is better than the first ISP parameter, and the image recognition rate after being processed by the second ISP parameter is also higher than the first ISP parameter, so the ISP parameter set corresponding to the smart terminal can be adjusted from the first ISP parameter set to the second ISP parameter set.
  • the first quality score is not greater than the first reference quality score and ⁇ or the first recognition rate is not greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter is not as good as the first ISP parameter, and the image recognition rate after being processed by the second ISP parameter is also lower than the first ISP parameter, so the ISP parameter set corresponding to the smart terminal is not adjusted from the first ISP parameter set to the second ISP parameter set.
  • the first reference recognition rate may be updated to the first recognition rate.
  • the first reference recognition rate is updated to the first recognition rate and the first reference quality score is updated to the first quality score. If the first quality score is not greater than the first reference quality score and/or the first recognition rate is not greater than the first reference recognition rate, the first reference recognition rate and the first reference quality score are kept unchanged. This ensures that the ISP parameter set adjusted subsequently is better than the previous ISP parameter set, thereby further improving the efficiency of ISP parameter adjustment and the recognition rate of the image.
  • a parameter adjustment device which has the function of implementing the parameter adjustment method in the first aspect.
  • the parameter adjustment device includes at least one module, which is used to implement the parameter adjustment method provided in the first aspect.
  • a parameter adjustment device comprising a processor and a memory, the memory being used to store a computer program for executing the parameter adjustment method provided in the first aspect.
  • the processor is configured to execute the computer program stored in the memory to implement the parameter adjustment method described in the first aspect.
  • the parameter adjustment device may further include a communication bus, and the communication bus is used to establish a connection between the processor and the memory.
  • a smart terminal comprising the parameter adjustment device described in the second or third aspect.
  • a computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the parameter adjustment method described in the first aspect.
  • a computer program product comprising instructions is provided, and when the instructions are executed on a computer, the computer executes the steps of the parameter adjustment method described in the first aspect.
  • a computer program is provided, and when the computer program is executed on a computer, the computer executes the parameter adjustment method described in the first aspect.
  • FIG1 is a schematic diagram of the structure of an intelligent vehicle provided by an exemplary embodiment of the present application.
  • FIG2 is a schematic diagram of the structure of a parameter adjustment device provided by an exemplary embodiment of the present application.
  • FIG3 is a flow chart of a parameter adjustment method provided by an exemplary embodiment of the present application.
  • FIG4 is a flow chart of a parameter adjustment method provided by another exemplary embodiment of the present application.
  • FIG. 5 is a schematic diagram of the structure of a parameter adjustment device provided by an exemplary embodiment of the present application.
  • image recognition has been widely used in various fields, such as transportation, intelligent security, autonomous driving, mobile terminal, etc.
  • a smart vehicle can be equipped with a camera to capture the environment around the smart vehicle to obtain images around the smart vehicle, process the images around the smart vehicle based on the ISP parameter set, and then recognize the processed images to determine the scene of the image, objects in the image, etc.
  • the smart vehicle determines people, cars, lane lines, bicycles, traffic signs, etc. in the image.
  • the smart vehicle can realize driving reminders, assisted driving, and autonomous driving based on the identified scenes and objects. Therefore, image recognition is the core of realizing autonomous driving, and image quality can directly affect the accuracy of image recognition, and ISP parameters determine the quality of image quality. In other words, ISP parameters can ultimately determine the accuracy of image recognition.
  • the ISP parameters can be adjusted manually. However, engineers are required to adjust the ISP parameters repeatedly, which will result in high labor costs and low efficiency. After the image is processed by the adjusted ISP parameters, although the human eye perceives the image quality of the image well, the accuracy of image recognition through the intelligent terminal is not good. Therefore, the embodiment of the present application provides a parameter adjustment method, which can realize the autonomous adjustment of ISP parameters without investing a lot of manpower costs, thereby improving the efficiency of ISP parameter adjustment, and can also ensure that the image conforms to the human eye vision and improve the image recognition rate.
  • the method provided by the embodiment of the present application is briefly introduced. Please refer to Figure 1.
  • the intelligent vehicle is equipped with a computing platform, for example, a mobile data center (MDC) and a camera is installed.
  • the computing platform is equivalent to the brain of the intelligent vehicle and can process various types of data.
  • the computing platform is located inside the intelligent vehicle and is powered by the battery of the intelligent vehicle.
  • the computing platform can support the power supply of several (usually several to more than a dozen) cameras to support the operation of the cameras.
  • the camera is a peripheral of the computing platform and is installed on the windshield of the intelligent vehicle or on the outside of the intelligent vehicle.
  • the camera can be used to capture the environment around the smart vehicle to obtain images around the smart vehicle.
  • the images captured by the camera include raw data (RAW) and embedded bitmap data (EBD).
  • the RAW and EBD are serialized by a serializer and transmitted to a deserializer for deserialization through a mobile industry processor interface (MIPI) protocol.
  • the deserialized data is then processed by applying an ISP parameter set to obtain a processed image.
  • the processed image format may be YUV420NV12, YUV420NV21, etc.
  • the processed image is identified by an image perception model/algorithm to determine the scene of the image, and the images of key scenes are stored in a training image set of a memory.
  • a test image set is also built into the memory.
  • the smart vehicle performs image quality evaluation on the training image set and the test image set based on an image quality evaluation system, and then adjusts the ISP parameters based on the image quality evaluation results.
  • the image quality assessment system is located on a system-on-chip (SoC) of a computing platform, and the image quality assessment system includes multiple image quality assessment algorithms, which include but are not limited to clarity, color, noise, white balance, wide dynamic and other image quality assessment algorithms.
  • SoC system-on-chip
  • the method provided in the embodiment of the present application can be executed by any intelligent terminal with image signal processing function, for example, the intelligent terminal can be a personal computer (PC), a mobile phone, a personal digital assistant (PDA), a handheld computer PPC (pocket PC), a tablet computer, a server, a robot, an intelligent driving device, a vehicle-mounted computing platform, etc.
  • the intelligent driving device in the present application can include land vehicles, water vehicles, air vehicles, industrial equipment, agricultural equipment, or entertainment equipment, etc.
  • the intelligent driving device can be a vehicle, which is a vehicle in a broad sense, and can be a vehicle (such as a commercial vehicle, a passenger car, a motorcycle, a flying car, a train, etc.), an industrial vehicle (such as a forklift, a trailer, a tractor, etc.), an engineering vehicle (such as an excavator, a bulldozer, a crane, etc.), agricultural equipment (such as a mower, a harvester, etc.), amusement equipment, a toy vehicle, etc.
  • the embodiment of the present application does not specifically limit the type of vehicle.
  • the intelligent driving device can be a vehicle such as an airplane or a ship.
  • the parameter adjustment device can be deployed on an intelligent terminal.
  • the parameter adjustment device includes at least one processor 201, a communication bus 202, a memory 203 and at least one communication interface 204.
  • the processor 201 may be a general-purpose central processing unit (CPU), a network processor (NP), a microprocessor, or may be one or more integrated circuits for implementing the solution of the present application, such as an application-specific integrated circuit (ASIC), a programmable logic device (PLD) or a combination thereof.
  • the above-mentioned PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.
  • the communication bus 202 is used to transmit information between the above components.
  • the communication bus 202 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.
  • the memory 203 may be a read-only memory (ROM), a random access memory (RAM), an electrically erasable programmable read-only memory (EEPROM), an optical disc (including a compact disc read-only memory (CD-ROM), a compressed optical disc, a laser disc, a digital versatile disc, a Blu-ray disc, etc.), a disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.
  • the memory 203 may exist independently and be connected to the processor 201 via the communication bus 202.
  • the memory 203 may also be integrated with the processor 201.
  • the communication interface 204 uses any transceiver-like device for communicating with other devices or communication networks.
  • the communication interface 204 includes a wired communication interface and may also include a wireless communication interface.
  • the wired communication interface may be, for example, an Ethernet interface.
  • the Ethernet interface may be an optical interface, an electrical interface, or a combination thereof.
  • the wireless communication interface may be a wireless local area network (WLAN) interface, a cellular network communication interface, or a combination thereof, etc.
  • WLAN wireless local area network
  • the processor 201 may include one or more CPUs, such as CPU0 and CPU1 shown in FIG2 .
  • the processor 201 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen.
  • the processor 201 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning and intelligent driving.
  • AI artificial intelligence
  • the parameter adjustment device may include multiple processors, such as processor 201 and processor 205 shown in Figure 2. Each of these processors may be a single-core processor or a multi-core processor.
  • the processor here may refer to one or more devices, circuits, and/or processing cores for processing data (such as computer program instructions).
  • the parameter adjustment device may further include an output device 206 and an input device 207.
  • the output device 206 communicates with the processor 201 and may display information in a variety of ways.
  • the output device 206 may be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, or a projector.
  • the input device 207 communicates with the processor 201 and may receive user input in a variety of ways.
  • the input device 207 may be a mouse, a keyboard, a touch screen device, or a sensor device.
  • the memory 203 is used to store the program code 210 for executing the solution of the present application, and the processor 201 can execute the program code 210 stored in the memory 203.
  • the program code 210 may include one or more software modules, and the parameter adjustment device can implement the parameter adjustment method provided in the embodiment of FIG. 3 below through the processor 201 and the program code 210 in the memory 203.
  • FIG3 is a flow chart of a parameter adjustment method provided in an embodiment of the present application.
  • the execution subject of the method is a smart terminal. Please refer to FIG3, the method includes the following steps.
  • Step 301 Determine a second ISP parameter set based on a training image set and a first ISP parameter set, where the first ISP parameter set is an ISP parameter set corresponding to the smart terminal.
  • the training image set includes N sub-training sets
  • the first ISP parameter set includes M parameters
  • each of the N sub-training sets corresponds to a scene
  • N and M are both integers greater than 1.
  • the following operations are performed: one of the sub-training sets is used as a target sub-training set, and a first sub-parameter set corresponding to a target scene in the first ISP parameter set is adjusted to obtain a second sub-parameter set corresponding to a target scene in a second ISP parameter set, the target scene is a scene corresponding to the target sub-training set, the first sub-parameter set includes at least one parameter of the M parameters, and the second sub-parameter set includes at least one parameter of the M parameters.
  • the N sub-training sets have one-to-one corresponding N scenes, each of the N scenes corresponds to a sub-parameter set in the first ISP parameter set, and each sub-parameter set includes at least one parameter of the M parameters.
  • the sub-training set is called a target sub-training set
  • the scene corresponding to the target sub-training set is called a target scene
  • the sub-parameter set corresponding to the target scene in the first ISP parameter is called a first sub-parameter set.
  • the sub-parameter set corresponding to the target scene in the second ISP parameter i.e., the second sub-parameter set
  • the sub-parameter set corresponding to each scene in the second ISP parameter set can be obtained.
  • the intelligent terminal stores a correspondence between the sub-training set and the scene, i.e., a first correspondence, and also stores a correspondence between the scene and the sub-parameter set, i.e., a second correspondence.
  • the scene corresponding to the target sub-training set i.e., the target scene
  • the sub-parameter set corresponding to the target scene i.e., the first sub-parameter set
  • the intelligent terminal can also store the correspondence between the sub-training set, the scene and the sub-parameter set. In this way, based on the target sub-training set, the sub-parameter set corresponding to the target scene, that is, the first sub-parameter set, can be directly determined from the correspondence between the three.
  • the first ISP parameter set is the ISP parameter set currently being applied by the intelligent terminal, and the current adjustment of the ISP parameters may be the first time or not. If the current adjustment of the ISP parameters is the first time, the parameter value corresponding to each parameter in the first ISP parameter set may be set in advance. If the current adjustment of the ISP parameters is not the first time, the parameter value corresponding to each parameter in the first ISP parameter set is obtained after the last adjustment of the ISP parameters.
  • the M parameters included in the first ISP parameter set are various parameters used for image processing.
  • the M parameters may be automatic tone remapping (ATR), dynamic range compression (DRC), gamma correction (GAMMA), raw noise fall (RAWNF), YUV noise fall (YUVNF), color correction matrix (CCM), white balance gain (AWB), etc., which is not limited in the embodiments of the present application.
  • the N scenes corresponding to the above-mentioned N sub-training sets can be key scenes for image recognition.
  • the N scenes are entering and exiting a tunnel during the day, backlighting on a sunny day, traffic intersections at night, traffic lights, viaducts, highways, and the like.
  • the sub-parameter set corresponding to each scene in the N scenes refers to a set of parameters that can affect the scene, and the sub-parameter sets corresponding to different scenes may or may not have intersections.
  • Table 1 the second corresponding relationship is shown in Table 1 below.
  • the sub-parameter set corresponding to the scene of entering and exiting a tunnel during the day includes ATR, DRC, and GAMMA
  • the sub-parameter set corresponding to the scene of backlighting on a sunny day includes DRC and GAMMA
  • the sub-parameter set corresponding to the scene of a traffic intersection at night includes RAWNF and YUVNF
  • the sub-parameter set corresponding to the scene of a traffic light includes CCM and AWB.
  • the sub-parameter sets corresponding to the scene of entering and exiting a tunnel during the day and the scene of backlighting on a sunny day have intersections
  • the sub-parameter sets corresponding to other scenes do not have intersections.
  • the implementation process of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set to obtain the second sub-parameter set corresponding to the target scene in the second ISP parameter set includes: adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set to multiple candidate sub-parameter sets, keeping the remaining parameters unchanged to obtain multiple candidate ISP parameter sets, based on the multiple candidate ISP parameter sets, determining multiple candidate quality scores corresponding to the target sub-training set, the multiple candidate quality scores indicating the image quality of the target sub-training set after being processed by the multiple candidate ISP parameter sets, and determining the parameter corresponding to the target scene in the candidate ISP parameter set corresponding to the maximum candidate quality score among the multiple candidate quality scores as the second sub-parameter set.
  • the intelligent terminal stores a correspondence between parameters, adjustment ranges, and adjustment steps, i.e., a third correspondence.
  • the adjustment range and adjustment step corresponding to each parameter in the first sub-parameter set can be determined from the third correspondence.
  • the parameter is used as the target parameter, and based on the adjustment range and adjustment step corresponding to the target parameter, multiple candidate parameter values corresponding to the target parameter are determined.
  • the multiple candidate parameter values corresponding to each parameter in the first sub-parameter set are combined to obtain multiple candidate sub-parameter sets.
  • multiple candidate parameter values corresponding to the target parameter can be determined according to the following formula (1).
  • Fn refers to the n+1th candidate parameter value among multiple candidate parameter values corresponding to the target parameter
  • min refers to the minimum value in the adjustment range corresponding to the target parameter
  • max refers to the maximum value in the adjustment range corresponding to the target parameter
  • step refers to the adjustment step corresponding to the target parameter
  • N refers to a natural number.
  • the adjustment range of the target parameter is 1 to 1.6 and the adjustment step is 0.2
  • the implementation process of combining multiple candidate parameter values corresponding to each parameter in the first sub-parameter set includes: for any parameter in the first sub-parameter set, arbitrarily selecting a candidate parameter value from the multiple candidate parameter values corresponding to the parameter, selecting a candidate parameter value corresponding to each parameter in the first sub-parameter set in the same manner, and using the candidate parameter value selected for each parameter in the first sub-parameter set as a candidate sub-parameter set.
  • Multiple candidate sub-parameter sets can be obtained in the same manner, and the multiple candidate sub-parameter sets are different.
  • the first sub-parameter set includes two parameters, and the first parameter corresponds to two candidate parameter values, namely 1 and 1.5.
  • the second parameter corresponds to three candidate parameter values, namely 2, 4, and 6.
  • the candidate parameter values corresponding to the two parameters are combined to obtain multiple candidate sub-parameter sets, namely (1, 2), (1, 4), (1, 6), (1.5, 2), (1.5, 4), and (1.5, 6).
  • the implementation process of adjusting the first sub-parameter set corresponding to the target scenario in the first ISP parameter set to multiple candidate sub-parameter sets, and keeping the remaining parameters unchanged to obtain multiple candidate ISP parameter sets includes: for any one of the multiple candidate sub-parameter sets, replacing the first sub-parameter set corresponding to the target scenario in the first ISP parameter set with the candidate sub-parameter set, and keeping the other parameters in the first ISP parameter set except the first sub-parameter set unchanged, thereby obtaining a candidate ISP parameter set.
  • each candidate sub-parameter set in the multiple candidate sub-parameter sets is processed in the above manner, multiple candidate ISP parameter sets can be obtained, and each candidate ISP parameter set corresponds to a candidate sub-parameter set.
  • the implementation process of determining multiple candidate quality scores corresponding to the target sub-training set based on the multiple candidate ISP parameter sets includes: based on the multiple candidate ISP parameter sets, processing each image in the target sub-training set respectively to obtain multiple processed target sub-training sets corresponding one-to-one to the multiple candidate ISP parameter sets; based on the multiple processed target sub-training sets, determining multiple candidate quality scores, and the multiple candidate quality scores corresponding one-to-one to the multiple processed target sub-training sets.
  • the intelligent terminal can determine the image quality score of each image in the processed target sub-training set according to a relevant algorithm to obtain multiple image quality scores, and determine the candidate quality score corresponding to the processed target sub-training set based on the multiple image quality scores. In the same way, multiple candidate quality scores can be obtained, and the multiple candidate quality scores correspond to the multiple processed target sub-training sets one by one.
  • the mode of the multiple image quality scores may be determined, and the mode may be determined as the candidate quality score corresponding to the processed target sub-training set.
  • the average value of the multiple image quality scores may be determined, and the average value may be determined as the candidate quality score corresponding to the processed target sub-training set.
  • the multiple image quality scores may also be processed in other ways to obtain the candidate quality score corresponding to the processed target sub-training set, and the embodiments of the present application are not limited to this.
  • the parameter corresponding to the target scene in the candidate ISP parameter set corresponding to the maximum candidate quality score among multiple candidate quality scores is determined as the second sub-parameter set. In this way, it can be ensured that the image processing effect of the second sub-parameter set finally determined is the best among the multiple candidate ISP parameter sets, further ensuring the adjustment effect of the ISP parameter set.
  • the test image set includes N sub-test sets, which correspond one-to-one to the N sub-training sets.
  • the intelligent terminal may also determine a second quality score, the second quality score indicating the image quality of the target sub-training set after being processed by the first ISP parameter set.
  • the first sub-parameter set corresponding to the target scene in the first ISP parameter set is adjusted in the above manner, the second reference quality score indicating the image quality of the target sub-test set after being processed by the first ISP parameter set, and the target sub-test set is the sub-test set corresponding to the target scene. If the second quality score is greater than or equal to the second reference quality score, the first sub-parameter set corresponding to the target scene in the first ISP parameter set is not adjusted.
  • the second quality score is less than the second reference quality score, it means that the processing effect of the images in the target sub-training set by the first ISP parameter set is not good, and the ISP parameters corresponding to the target scene need to be adjusted, so the step of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set is performed. If the second quality score is greater than or equal to the second reference quality score, it means that the processing effect of the images in the target sub-training set by the first ISP parameter set is good, and the ISP parameters corresponding to the target scene do not need to be adjusted, so the step of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set is not performed.
  • the process of determining the second quality score and the process of determining the second reference quality score are similar to the process of determining the candidate quality score. For details, please refer to the corresponding content above, which will not be repeated here.
  • test image set is set in advance, and can be adjusted according to different needs in different situations.
  • the first sub-parameter set can be directly adjusted according to the above method to obtain the second sub-parameter set.
  • target sub-training set processed by the first ISP parameter set can also be scored to determine the processing effect of the first ISP parameter set on the images in the target sub-training set, and then determine whether the first sub-parameter set needs to be adjusted.
  • N sub-parameter sets can be obtained. Since each sub-parameter set includes at least one parameter of the M parameters, the union of the N sub-parameter sets may include each parameter of the M parameters, or may only include some parameters of the M parameters. In different cases, the methods for determining the second ISP parameter set based on the N sub-parameter sets are different, which will be introduced below.
  • the union of the N sub-parameter sets includes each parameter in the M parameters.
  • a second ISP parameter set can be determined based on the N sub-parameter sets.
  • the sub-parameter sets corresponding to different scenarios may or may not have intersections.
  • the N sub-parameter sets can be directly merged to obtain a second ISP parameter set.
  • the parameters with intersections are called first-category parameters, and the parameters without intersections are called second-category parameters.
  • the parameter values of the first-category parameters in the N sub-parameter sets are combined to obtain multiple parameter value combinations, and the multiple parameter value combinations are merged with the parameter values of the second-category parameters in the N sub-parameter sets to obtain multiple second ISP parameter sets.
  • the parameter value is merged with the parameter value of the second type of parameters in the N sub-parameter sets to obtain a second ISP parameter set.
  • each parameter value of the first type of parameters is merged with the parameter value of the second type of parameters in the N sub-parameter sets to obtain multiple second ISP parameter sets.
  • each parameter value combination is merged with the parameter value of the second-category parameters in the N sub-parameter sets to obtain a second ISP parameter set.
  • each parameter value combination is merged with the parameter value of the second-category parameters in the N sub-parameter sets to obtain multiple second ISP parameter sets.
  • the implementation method of combining multiple parameter values corresponding to each parameter in the first category of parameters is similar to the implementation method of combining multiple candidate parameter values corresponding to each parameter in the first sub-parameter set mentioned above. Please refer to the relevant content above and will not be repeated here.
  • the union of the N sub-parameter sets includes some parameters in the M parameters.
  • the second ISP parameter set is determined based on the N sub-parameter sets and the first ISP parameter set.
  • the sub-parameter sets corresponding to different scenarios may or may not have intersections.
  • the N sub-parameter sets can be directly merged to obtain a merged parameter set, and the merged parameter set can be merged again with other parameters in the first ISP parameter set except the merged parameter set to obtain a second ISP parameter set.
  • the parameters with intersections are called first-category parameters, and the parameters without intersections are called second-category parameters.
  • the parameter values of the first-category parameters in the N sub-parameter sets are combined to obtain multiple parameter value combinations, and the multiple parameter value combinations are merged with the parameter values of the second-category parameters in the N sub-parameter sets to obtain multiple merged parameter sets.
  • the merged parameter set is merged again with other parameters in the first ISP parameter set except the merged parameter set to obtain multiple second ISP parameter sets.
  • the second ISP parameter set based on the training image set and the first ISP parameter set before determining the second ISP parameter set based on the training image set and the first ISP parameter set, it is also possible to obtain an image corresponding to the current environment, process the image corresponding to the current environment based on the first ISP parameter set, and perform scene recognition based on the processed image to obtain a scene recognition result, which indicates whether there is a key scene in the current environment. If the scene recognition result indicates that there is a key scene in the current environment, the image corresponding to the current environment is stored in the training image set.
  • the smart terminal has a camera that can capture the environment around the smart terminal to obtain an image corresponding to the current environment. In this way, the smart terminal can obtain an image corresponding to the current environment.
  • the recognition can be performed through a neural network model, that is, the processed image is input into the neural network model to obtain the scene name output by the neural network model, that is, the scene recognition result.
  • the neural network model Before using the neural network model for recognition, the neural network model needs to be trained. That is, multiple sample images and the scene name corresponding to each image are obtained, the image is used as the input of the neural network model, and the scene name is used as the output of the neural network model to train the neural network model.
  • the smart terminal stores multiple key scenes, so that the smart terminal can determine whether the scene indicated by the scene recognition result exists in the multiple key scenes. If the scene indicated by the scene recognition result exists in the multiple key scenes, the scene indicated by the scene recognition result is determined to be a key scene, and the image corresponding to the current environment is stored in the training image set. If the scene indicated by the scene recognition result does not exist in the multiple key scenes, the scene indicated by the scene recognition result is determined not to be a key scene, and the image corresponding to the current environment is not stored in the training image set.
  • the training image set includes N sub-training sets, and each sub-training set corresponds to a scene. If the scene indicated by the scene recognition result is a key scene, the image corresponding to the current environment is stored in the sub-training set of the corresponding scene.
  • the images in the training image set can be images taken by the camera in real time. Since different users have different behavioral habits, the images taken by the camera are also different. Therefore, adjusting the ISP parameter set based on the images taken by the camera in real time can make the final adjusted ISP parameters adaptive.
  • the images in the training image set can also be images obtained by other means, and the embodiments of the present application are not limited to this.
  • the intelligent terminal can also determine in real time the number of images in each sub-training set in the training image set. If the number of images in each sub-training set in the training image set reaches the image number threshold, the above step 301 can be executed. Of course, if there is a sub-training set in the training image set that reaches the image number threshold, the intelligent terminal can also process the sub-training set first to obtain the sub-parameter set corresponding to the sub-training set in the second ISP parameter set until all sub-training sets in the training image set are processed. In other words, the intelligent terminal can process each sub-training set only after all sub-training sets in the training image set reach the image number threshold. It is also possible to directly process a sub-training set when the number of images in a sub-training set reaches the image number threshold without waiting for the number of images in other sub-training sets to reach the image number threshold.
  • the image quantity threshold is set in advance, and the image quantity threshold corresponding to each sub-training set can be the same or different, and can be adjusted according to different requirements in different situations.
  • the embodiment of the present application does not limit this.
  • the images in the training image set are images that have not been processed by the ISP parameter set.
  • the adjusted ISP parameter set can be applied to the images that have not been processed by the ISP parameter set, so as to determine the quality of the adjusted ISP parameter set.
  • Step 302 Process the test image set based on the second ISP parameter set.
  • the number of the second ISP parameter sets may be one or more. If there are multiple second ISP parameter sets, the test image set is processed based on each second ISP parameter set in the multiple second ISP parameter sets, thereby obtaining multiple processed test image sets.
  • a test image set can be obtained through a recharge interface, which can be expressed as: recharge(&test_type, &scene, &camera_type, &FOV, &picture_path).
  • recharge represents the recharge interface
  • &test_type is used to determine whether it is a training image set or a test image set, where 1 represents a training image set, 0 represents a test image set
  • &scene represents a scene, such as entering a tunnel, inside a tunnel, an overpass, backlighting, a highway, a desert, snowy days, rainy days, etc.
  • &camera_type represents the camera type
  • &FOV represents the camera wide-angle type, for example, close range (FOV120), medium range (FOV60), long range (FOV30), fisheye (FOV180), etc.
  • &picture_path represents the recharged image path directory.
  • Step 303 Determine a first quality score and a first recognition rate based on the processed test image set, where the first quality score indicates the image quality of the processed test image set, and the first recognition rate indicates the image recognition status of the processed test image set.
  • the processed test image set may be one or more.
  • the methods for determining the first quality score and the first recognition rate are different, which will be introduced below.
  • the first quality score and the first recognition rate are determined based on the processed test image set.
  • the average of the quality scores of the N subtest sets can be determined as the first quality score.
  • each subtest set corresponds to a weight. The quality scores of the N subtest sets are multiplied by their respective corresponding weights and then added to obtain the first quality score.
  • the process of determining the quality score of each sub-test set in the processed test image set is similar to the process of determining the candidate quality score of the processed target sub-training set mentioned above. Please refer to the corresponding content in the above text and will not be repeated here.
  • the mode of the recognition rates of the N subtest sets may be determined, and the mode may be determined as the first recognition rate.
  • the average value of the recognition rates of the N subtest sets may be determined, and the average value may be determined as the first recognition rate.
  • the first recognition rate may also be determined in other ways, and the embodiments of the present application do not limit this.
  • the quality scores and recognition rates of the multiple processed test image sets are determined, the maximum quality score among the quality scores of the multiple processed test image sets is used as the first quality score, and the recognition rate corresponding to the maximum quality score is used as the first recognition rate.
  • Step 304 If the first quality score is greater than the first reference quality score and the first recognition rate is greater than the first reference recognition rate, the ISP parameter set corresponding to the smart terminal is adjusted from the first ISP parameter set to the second ISP parameter set, the first reference quality score indicates the image quality of the test image set after being processed by the first ISP parameter set, and the first reference recognition rate indicates the image recognition status of the test image set after being processed by the first ISP parameter set.
  • the process of determining the first reference quality score and the first reference recognition rate is similar to the process of determining the first quality score and the first recognition rate in the first case of step 303. Please refer to the corresponding content above and will not be repeated here.
  • the ISP parameter set corresponding to the smart terminal is not adjusted from the first ISP parameter set to the second ISP parameter set, that is, the ISP parameter set corresponding to the smart terminal remains unchanged.
  • the first quality score is greater than the first reference quality score and the first recognition rate is greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter is better than the first ISP parameter, and the image recognition rate after being processed by the second ISP parameter is also higher than the first ISP parameter, so the ISP parameter set corresponding to the smart terminal can be adjusted from the first ISP parameter set to the second ISP parameter set.
  • the first quality score is not greater than the first reference quality score and ⁇ or the first recognition rate is not greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter is not as good as the first ISP parameter, and the image recognition rate after being processed by the second ISP parameter is also lower than the first ISP parameter, so the ISP parameter set corresponding to the smart terminal is not adjusted from the first ISP parameter set to the second ISP parameter set.
  • one second ISP parameter set may be determined through the above step 301, or multiple second ISP parameter sets may be determined.
  • the ISP parameter set corresponding to the smart terminal may be directly adjusted from the first ISP parameter set to the second ISP parameter set.
  • the ISP parameter set corresponding to the smart terminal may be adjusted from the first ISP parameter set to the second ISP parameter set corresponding to the first quality score.
  • the first reference recognition rate may be updated to the first recognition rate.
  • the first reference recognition rate is updated to the first recognition rate and the first reference quality score is updated to the first quality score. If the first quality score is not greater than the first reference quality score and/or the first recognition rate is not greater than the first reference recognition rate, the first reference recognition rate and the first reference quality score are kept unchanged. This ensures that the ISP parameter set adjusted subsequently is better than the previous ISP parameter set, thereby further improving the efficiency of ISP parameter adjustment and the recognition rate of the image.
  • the smart terminal can trigger an update prompt, which is used to prompt the user to update the ISP parameter set corresponding to the smart terminal. If the smart terminal receives an update instruction triggered by the user, it means that the user agrees to update the ISP parameter set corresponding to the smart terminal. At this time, the smart terminal adjusts the corresponding ISP parameter set from the first ISP parameter set to the second ISP parameter set.
  • the smart terminal can directly adjust the corresponding ISP parameter set from the first ISP parameter set to the second ISP parameter set, or ask the user whether to update the ISP parameter set. Only when the user agrees to update, the smart terminal will adjust the corresponding ISP parameter set from the first ISP parameter set to the second ISP parameter set.
  • the above content is to determine whether the first quality score is greater than the first reference quality score and whether the first recognition rate is greater than the first reference recognition rate after the first quality score and the first recognition rate are determined.
  • the first quality score can also be determined first, and when the first quality score is greater than the first reference quality score, the first recognition rate can be determined to determine whether the first recognition rate is greater than the first reference recognition rate.
  • the embodiments of the present application do not limit this.
  • the camera captures the environment around the smart vehicle to obtain an image corresponding to the current environment, and then the computing platform processes the image corresponding to the current environment based on the first ISP parameter set, and recognizes the processed image. Based on the scene recognition result obtained by recognition, the image corresponding to the current environment is stored in a training image set.
  • a second ISP parameter set is determined based on the training image set, and then the images in the test image set are processed based on the second ISP parameter set to determine a first quality score and a first recognition rate of the test image set.
  • an ISP parameter update prompt is triggered to remind the user to update the ISP parameters.
  • the first ISP parameter set is adjusted to the second ISP parameter set.
  • the embodiment of the present application automatically determines the second ISP parameter based on the training image set and the first ISP parameter set, and does not require manual repeated adjustment of the ISP parameter, thereby reducing the labor cost.
  • the first quality score represents the image quality of the test image set after being processed by the second ISP parameter set
  • the first recognition rate represents the image recognition of the test image set after being processed by the second ISP parameter set
  • the first reference quality score represents the image quality of the test image set after being processed by the first ISP parameter set
  • the first reference recognition rate represents the image recognition of the test image set after being processed by the first ISP parameter set.
  • the first quality score is greater than the first reference quality score, and the first recognition rate is greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter set is better than that of the first ISP parameter set, and the image recognition rate after being processed by the second ISP parameter set is also higher than that of the first ISP parameter set.
  • adjusting the ISP parameter set corresponding to the smart terminal from the first ISP parameter set to the second ISP parameter set can not only ensure that the quality of the image obtained by the smart terminal according to the second ISP parameter set is better than the image quality obtained according to the first ISP parameter set, but also ensure that the recognition rate of the image obtained by the smart terminal according to the second ISP parameter set is also better than the recognition rate of the image obtained according to the first ISP parameter set. That is to say, the method provided in the present application can not only autonomously adjust the ISP parameters without investing a lot of manpower costs, improve the efficiency of ISP parameter adjustment, but also improve the recognition rate of images on the basis of ensuring human eye vision.
  • the method provided in the embodiment of the present application can also update the first reference recognition rate, or the first reference recognition rate and the first reference quality score, so that the ISP parameter set adjusted subsequently can be better than the previous ISP parameter set, thereby further improving the efficiency of ISP parameter adjustment and the recognition rate of images.
  • the method provided in the embodiment of the present application can also capture images in real time, and then adjust the ISP parameter set based on the real-time captured image, that is, different users capture different images, and then the ISP parameter set adjusted based on the captured image is different, therefore, the final adjusted ISP parameters are adaptive, and can adjust better ISP parameters for different users for the user.
  • FIG5 is a schematic diagram of the structure of a parameter adjustment device provided in an embodiment of the present application, and the parameter adjustment device can be implemented as part or all of the intelligent terminal by software, hardware or a combination of both.
  • the device includes: a first determination module 501, a first processing module 502, a second determination module 503 and an adjustment module 504.
  • the first determination module 501 is used to determine the second ISP parameter set based on the training image set and the first image signal processing ISP parameter set, where the first ISP parameter set is the ISP parameter set corresponding to the smart terminal.
  • the detailed implementation process refers to the corresponding content in the above embodiments, which will not be repeated here.
  • the first processing module 502 is used to process the test image set based on the second ISP parameter set.
  • the detailed implementation process refers to the corresponding content in the above embodiments, which will not be repeated here.
  • the second determination module 503 is used to determine a first quality score and a first recognition rate based on the processed test image set, wherein the first quality score indicates the image quality of the processed test image set, and the first recognition rate indicates the image recognition of the processed test image set.
  • the detailed implementation process refers to the corresponding content in the above embodiments, which will not be repeated here.
  • the adjustment module 504 is used to adjust the ISP parameter set corresponding to the intelligent terminal from the first ISP parameter set to the second ISP parameter set if the first quality score is greater than the first reference quality score and the first recognition rate is greater than the first reference recognition rate, wherein the first reference quality score indicates the image quality of the test image set after being processed by the first ISP parameter set, and the first reference recognition rate indicates the image recognition of the test image set after being processed by the first ISP parameter set.
  • the detailed implementation process refers to the corresponding contents in the above-mentioned embodiments, which will not be repeated here.
  • the training image set includes N sub-training sets
  • the first ISP parameter set includes M parameters
  • each of the N sub-training sets corresponds to a scene
  • N and M are both integers greater than 1;
  • the first determining module 501 is specifically used for:
  • the target sub-training set Take one of the sub-training sets as the target sub-training set, adjust the first sub-parameter set corresponding to the target scene in the first ISP parameter set, so as to obtain the second sub-parameter set corresponding to the target scene in the second ISP parameter set, where the target scene is the scene corresponding to the target sub-training set, and the first sub-parameter set includes at least one parameter among the M parameters.
  • the first determining module 501 is specifically configured to:
  • the parameters corresponding to the target scene in the candidate ISP parameter set corresponding to the maximum candidate quality score among the multiple candidate quality scores are determined as the second sub-parameter set.
  • the test image set includes N sub-test sets, and the N sub-test sets correspond one-to-one to the N sub-training sets;
  • the first determining module 501 is specifically used for:
  • the step of adjusting the first sub-parameter set corresponding to the target scene in the first ISP parameter set is executed, and the second reference quality score indicates the image quality of the target sub-test set after being processed by the first ISP parameter set, and the target sub-test set is the sub-test set corresponding to the target scene.
  • the device further comprises:
  • An updating module is used to update the first reference recognition rate to the first recognition rate if the first quality score is greater than the first reference quality score and the first recognition rate is greater than the first reference recognition rate, or to update the first reference recognition rate to the first recognition rate and the first reference quality score to the first quality score.
  • the device further comprises:
  • An acquisition module is used to obtain an image corresponding to the current environment
  • a second processing module configured to process an image corresponding to the current environment based on the first ISP parameter set
  • a recognition module configured to perform scene recognition based on the processed image to obtain a scene recognition result, wherein the scene recognition result indicates whether a key scene exists in the current environment;
  • the storage module is used to store the image corresponding to the current environment into the training image set if the scene recognition result indicates that there is a key scene in the current environment.
  • the embodiment of the present application automatically determines the second ISP parameter based on the training image set and the first ISP parameter set, without manually adjusting the ISP parameter repeatedly, thereby reducing the labor cost.
  • the first quality score represents the image quality of the test image set after being processed by the second ISP parameter set
  • the first recognition rate represents the image recognition of the test image set after being processed by the second ISP parameter set
  • the first reference quality score represents the image quality of the test image set after being processed by the first ISP parameter set
  • the first reference recognition rate represents the image recognition of the test image set after being processed by the first ISP parameter set.
  • the first quality score is greater than the first reference quality score, and the first recognition rate is greater than the first reference recognition rate, it means that the image quality after being processed by the second ISP parameter set is better than that of the first ISP parameter set, and the image recognition rate after being processed by the second ISP parameter set is also higher than that of the first ISP parameter set.
  • adjusting the ISP parameter set corresponding to the smart terminal from the first ISP parameter set to the second ISP parameter set can not only ensure that the quality of the image obtained by the smart terminal according to the second ISP parameter set is better than the image quality obtained according to the first ISP parameter set, but also ensure that the recognition rate of the image obtained by the smart terminal according to the second ISP parameter set is also better than the recognition rate of the image obtained according to the first ISP parameter set. That is to say, the method provided in the present application can not only autonomously adjust the ISP parameters without investing a lot of manpower costs, improve the efficiency of ISP parameter adjustment, but also improve the recognition rate of images on the basis of ensuring human eye vision.
  • the method provided in the embodiment of the present application can also update the first reference recognition rate, or the first reference recognition rate and the first reference quality score, so that the ISP parameter set adjusted subsequently can be better than the previous ISP parameter set, thereby further improving the efficiency of ISP parameter adjustment and the recognition rate of images.
  • the method provided in the embodiment of the present application can also capture images in real time, and then adjust the ISP parameter set based on the real-time captured image, that is, different users capture different images, and then the ISP parameter set adjusted based on the captured image is different, therefore, the final adjusted ISP parameters are adaptive, and can adjust better ISP parameters for different users.
  • the parameter adjustment device provided in the above embodiment only uses the division of the above functional modules as an example when performing parameter adjustment.
  • the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.
  • the parameter adjustment device provided in the above embodiment and the parameter adjustment method embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.
  • the computer program product includes one or more computer instructions.
  • the computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device.
  • the computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions can be transmitted from a website site, computer, server or data center by wired (for example: coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example: infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.
  • the computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that includes one or more available media integrated.
  • the available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a digital versatile disc (DVD)), or a semiconductor medium (e.g., a solid state disk (SSD)).
  • the computer-readable storage medium mentioned in the embodiment of the present application may be a non-volatile storage medium, in other words, a non-transient storage medium.
  • the information including but not limited to user device information, user personal information, etc.
  • data including but not limited to data for analysis, stored data, displayed data, etc.
  • signals involved in the embodiments of the present application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.
  • the training image set, test image set and first ISP parameter set involved in the embodiments of the present application are all obtained with full authorization.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

本申请公开了一种参数调节方法、装置以及智能终端,属于图像处理技术领域。所述方法包括:基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集,基于第二ISP参数集对测试图像集进行处理,基于处理后的测试图像集确定第一质量分数和第一识别率,若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,则将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。本申请提供的方法不仅能够在不投入大量的人力成本的基础上,自主调节ISP参数,提高ISP参数调节的效率,还能够在保证人眼视觉的基础上,提高图像的识别率。

Description

参数调节方法、装置以及智能终端 技术领域
本申请涉及图像处理技术领域,特别涉及一种参数调节方法、装置以及智能终端。
背景技术
目前,图像识别已经广泛应用于各个领域之中,例如,交通领域、智能安防领域、自动驾驶领域、手机终端领域等等。以自动驾驶为例,图像识别是实现自动驾驶的核心,图像质量能够直接影响图像识别的准确率,而图像信号处理(image signal processing,ISP)参数决定了图像质量的优劣。也就是说,ISP参数最终能够决定图像识别的准确率。
在相关技术中,通过人工进行ISP参数的调节。也即是,工程师按照经验对ISP参数进行调节,将调节后的ISP参数应用到图像信号处理中,并确定根据该ISP参数得到的图像的图像质量以及主观效果。如果通过调节后的ISP参数处理得到的图像不满足图像质量和主观效果的设定指标,则工程师再次依据经验调节ISP参数,直至根据该调节后的ISP参数得到的图像满足图像质量和主观效果的设定指标,将满足设定指标的ISP参数作为最终的ISP参数。然而,上述方法需要工程师反复调节ISP参数,这会导致投入的人力成本较高、效率较低。
发明内容
本申请提供了一种参数调节方法、装置以及智能终端,能够在不投入大量的人力成本的基础上,自主调节ISP参数,提高ISP参数调节的效率,还能够在保证人眼视觉的基础上,提高图像的识别率。
第一方面,本申请提供一种参数调节方法,在该方法中,基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集,所述第一ISP参数集为智能终端对应的ISP参数集,基于所述第二ISP参数集对测试图像集进行处理,基于处理后的测试图像集确定第一质量分数和第一识别率,所述第一质量分数指示处理后的测试图像集的图像质量,所述第一识别率指示处理后的测试图像集的图像识别情况,若所述第一质量分数大于第一参考质量分数且所述第一识别率大于第一参考识别率,则将所述智能终端对应的ISP参数集从所述第一ISP参数集调整为所述第二ISP参数集,所述第一参考质量分数指示经过所述第一ISP参数集处理后的所述测试图像集的图像质量,所述第一参考识别率指示经过所述第一ISP参数集处理后的所述测试图像集的图像识别情况。
本申请是基于训练图像集和第一ISP参数集来自动地确定第二ISP参数,无需人工反复调节ISP参数,降低了人力成本。而且,由于该第一质量分数表征经过第二ISP参数集处理后的测试图像集的图像质量,第一识别率表征经过第二ISP参数集处理后的测试图像集的图像识别情况,第一参考质量分数表征经过第一ISP参数集处理后的测试图像集的图像质量,第一参考识别率表征经过第一ISP参数集处理后的测试图像集的图像识别情况。因此,在第一质量分数大于第一参考质量分数,且第一识别率大于第一参考识别率的情况下,说明经过 第二ISP参数集处理后的图像质量优于第一ISP参数集,并且经过第二ISP参数集处理后的图像识别率也高于第一ISP参数集。在这种情况下,将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集,不仅能够保证智能终端根据第二ISP参数集得到的图像的质量优于根据第一ISP参数集得到的图像质量,还能够保证智能终端根据第二ISP参数集得到的图像的识别率也优于根据第一ISP参数集得到的图像的识别率。也就是说,本申请提供的方法不仅能够在不投入大量的人力成本的基础上,自主调节ISP参数,提高ISP参数调节的效率,还能够在保证人眼视觉的基础上,提高图像的识别率。
可选地,训练图像集包括N个子训练集,第一ISP参数集包括M个参数,该N个子训练集中的每个子训练集对应一个场景,N和M均为大于1的整数。对于该N个子训练集中的每个子训练集均执行以下操作:将其中一个子训练集作为目标子训练集,调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集,目标场景为目标子训练集对应的场景,第一子参数集包括该M个参数中的至少一个参数,第二子参数集包括该M个参数中的至少一个参数。
也即是,该N个子训练集具有一一对应的N个场景,该N个场景中的每个场景对应第一ISP参数集中的一个子参数集,每个子参数集包括该M个参数中的至少一个参数。对于该N个子训练集中的任一子训练集,将该子训练集称为目标子训练集,将目标子训练集对应的场景称为目标场景,将第一ISP参数中目标场景对应的子参数集称为第一子参数集,对第一子参数集中各个参数对应的参数值进行调节,能够得到第二ISP参数中目标场景对应的子参数集,即第二子参数集。按照相同的方式对第一ISP参数集中每个场景对应的子参数集进行调节,能够得到第二ISP参数集中每个场景对应的子参数集。
可选地,调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集的实现过程包括:将第一ISP参数集中目标场景对应的第一子参数集调整为多个候选子参数集,将其余参数保持不变,以得到多个候选ISP参数集,基于该多个候选ISP参数集,确定目标子训练集对应的多个候选质量分数,该多个候选质量分数指示经过该多个候选ISP参数集处理后的目标子训练集的图像质量,将该多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中目标场景对应的参数,确定为第二子参数集。
由于候选质量分数越大,说明ISP参数集对图像的处理效果越好,因此,将多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中目标场景对应的参数,确定为第二子参数集。这样,能够保证最终确定的第二子参数集对于图像的处理效果是该多个候选ISP参数集中最好的,进一步保证ISP参数集的调节效果。
可选地,测试图像集包括N个子测试集,该N个子测试集与N个子训练集一一对应。智能终端在调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集之前,还可以确定第二质量分数,第二质量分数指示经过第一ISP参数集处理后的目标子训练集的图像质量,若第二质量分数小于第二参考质量分数,则按照上述方式调节第一ISP参数集中目标场景对应的第一子参数集,第二参考质量分数指示经过第一ISP参数集处理后的目标子测试集的图像质量,目标子测试集为目标场景对应的子测试集。若第二质量分数大于或等于第二参考质量分数,则不调节第一ISP参数集中目标场景对应的第一子参数集。
若第二质量分数小于第二参考质量分数,说明通过第一ISP参数集对目标子训练集中的图像的处理效果不佳,需要对目标场景对应的ISP参数进行调节,因此执行调节第一ISP参数集中目标场景对应的第一子参数集的步骤。若第二质量分数大于或等于第二参考质量分数,说明通过第一ISP参数集对目标子训练集中的图像的处理效果较好,并不需要对目标场景对应的ISP参数进行调节,因此不执行调节第一ISP参数集中目标场景对应的第一子参数集的步骤。
按照上述方式确定出第二ISP参数集中每个场景对应的子参数集之后,能够得到N个子参数集。由于每个子参数集包括M个参数中的至少一个参数,该N个子参数集的并集可能包括该M个参数中的每个参数,也可能只包括该M个参数中的部分参数。在不同的情况下,基于该N个子参数集确定第二ISP参数集的方式不同,接下来将分别进行介绍。
第一种情况,该N个子参数集的并集包括该M个参数中的每个参数。在这种情况下,可以基于该N个子参数集确定第二ISP参数集。
基于上文描述,不同场景对应的子参数集可能存在交集,也可能不存在交集。在这些场景对应的子参数集不存在交集的情况下,可以直接将该N个子参数集进行合并,以得到第二ISP参数集。在这些场景对应的子参数集存在交集的情况下,将存在交集的这些参数称为第一类参数,将不存在交集的参数称为第二类参数,将该N个子参数集中第一类参数的参数值进行组合,以得到多种参数值组合,将该多种参数值组合与该N个子参数集中第二类参数的参数值进行合并,以得到多个第二ISP参数集。
第二种情况,该N个子参数集的并集包括该M个参数中的部分参数。在这种情况下,基于该N个子参数集和第一ISP参数集确定第二ISP参数集。
与上文同理,不同场景对应的子参数集可能存在交集,也可能不存在交集。在这些场景对应的子参数集不存在交集的情况下,可以直接将该N个子参数集进行合并,以得到合并参数集,将该合并参数集与第一ISP参数集中除合并参数集之外的其他参数进行再次合并,以得到第二ISP参数集。在这些场景对应的子参数集存在交集的情况下,将存在交集的这些参数称为第一类参数,将不存在交集的参数称为第二类参数,将该N个子参数集中第一类参数的参数值进行组合,以得到多种参数值组合,将该多种参数值组合与该N个子参数集中第二类参数的参数值进行合并,以得到多个合并参数集。对于该多个合并参数集中的每个合并参数集,将该合并参数集与第一ISP参数集中除该合并参数集之外的其他参数进行再次合并,能够得到多个第二ISP参数集。
可选地,基于训练图像集和第一ISP参数集确定第二ISP参数集之前,智能终端还能够获取当前环境对应的图像,基于第一ISP参数集处理当前环境对应的图像,基于处理后的图像进行场景识别,以得到场景识别结果,该场景识别结果指示当前环境中是否存在关键场景,若该场景识别结果指示当前环境中存在关键场景,则将当前环境对应的图像存储至训练图像集中。
在实际应用中,由于摄像头自身存在一些缺陷或者摄像头在拍摄时的光学条件不同,这会导致摄像头拍摄的图像的质量较差,难以较好地还原拍摄现场的细节。因此,需要对摄像头拍摄的图像进行后期处理,即对拍摄的图像应用第一ISP参数集进行处理,才能保证经过处理后的图像能够较好地还原拍摄现场的细节,得到质量较好的图像,进而保证场景识别结果的准确性。
基于上文描述,训练图像集包括N个子训练集,每个子训练集对应一个场景。若该场景识别结果所指示的场景为关键场景,则将当前环境对应的图像存储至对应场景的子训练集中。也就是说,训练图像集中的图像可以是摄像头实时拍摄的图像。由于不同用户的行为习惯不同,进而摄像头所拍摄的图像也不同,因此,基于摄像头实时拍摄的图像进行ISP参数集的调节,能够使最终调节的ISP参数具有自适应性。当然,训练图像集中的图像也可以是通过其他方式获得的图像。
需要说明的是,上述训练图像集中的图像是未经过ISP参数集处理之后的图像。这样,在后续调节ISP参数时,能够将调节之后的ISP参数集应用至该未经ISP参数集处理的图像中,以便确定该调节之后的ISP参数集的优劣。
基于上文描述,第二ISP参数集的数量可能为一个,也可能为多个。若存在多个第二ISP参数集,则基于该多个第二ISP参数集中的每个第二ISP参数集,分别对测试图像集进行处理,从而得到多个处理后的测试图像集。
与上文同理,处理后的测试图像集可能为一个,也可能为多个。在不同的情况下,确定第一质量分数和第一识别率的方式不同,接下来将分别进行介绍。
第一种情况,处理后的测试图像集为一个。此时,基于该处理后的测试图像集确定第一质量分数和第一识别率。
确定该处理后的测试图像集包括的N个子测试集中每个子测试集的质量分数,基于该N个子测试集的质量分数,确定第一质量分数。确定该处理后的测试图像集包括的N个子测试集中每个子测试集的识别率,基于该N个子测试集的识别率,确定第一识别率。
第二种情况,处理后的测试图像集为多个。确定该多个处理后的测试图像集的质量分数和识别率,将该多个处理后的测试图像集的质量分数中的最大质量分数作为第一质量分数,将与该最大质量分数对应的识别率作为第一识别率。
对于该多个处理后的测试图像集中的每个测试图像集,确定该处理后的测试图像集包括的N个子测试集中每个子测试集的质量分数,基于该N个子测试集的质量分数,确定该处理后的测试图像集的质量分数。确定该处理后的测试图像集包括的N个子测试集中每个子测试集的识别率,基于该N个子测试集的识别率,确定该处理后的测试图像集识别率。
若第一质量分数不大于第一参考质量分数和\或第一识别率不大于第一参考识别率,则不将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。即保持智能终端对应的ISP参数集不变。
若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,说明经过第二ISP参数处理后的图像质量优于第一ISP参数,并且经过第二ISP参数处理后的图像识别率也高于第一ISP参数,因此可以将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。若第一质量分数不大于第一参考质量分数和\或第一识别率不大于第一参考识别率,说明经过第二ISP参数处理后的图像质量不如第一ISP参数,并且经过第二ISP参数处理后的图像识别率也低于第一ISP参数,因此不将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。
可选地,若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,还可以将第一参考识别率更新为第一识别率。或者,将第一参考识别率更新为第一识别率并且将第一参考质量分数更新为第一质量分数。若第一质量分数不大于第一参考质量分数和\或第 一识别率不大于第一参考识别率,则保持第一参考识别率和第一参考质量分数不变。这样可以保证后续调节的ISP参数集比上一个ISP参数集更优,从而进一步提高ISP参数调节的效率以及图像的识别率。
第二方面,提供了一种参数调节装置,所述参数调节装置具有实现上述第一方面中参数调节方法的功能。所述参数调节装置包括至少一个模块,该至少一个模块用于实现上述第一方面所提供的参数调节方法。
第三方面,提供了一种参数调节装置,所述参数调节装置包括处理器和存储器,所述存储器用于存储执行上述第一方面所提供的参数调节方法的计算机程序。所述处理器被配置为用于执行所述存储器中存储的计算机程序,以实现上述第一方面所述的参数调节方法。
可选地,所述参数调节装置还可以包括通信总线,该通信总线用于该处理器与存储器之间建立连接。
第四方面,提供了一种智能终端,所述智能终端包括上述第二或第三方面所述的参数调节装置。
第五方面,提供了一种计算机可读存储介质,所述存储介质内存储有指令,当所述指令在计算机上运行时,使得计算机执行上述第一方面所述的参数调节方法。
第六方面,提供了一种包含指令的计算机程序产品,当所述指令在计算机上运行时,使得计算机执行上述第一方面所述的参数调节方法的步骤。或者说,提供了一种计算机程序,当所述计算机程序在计算机上运行时,使得计算机执行上述第一方面所述的参数调节方法。
上述第二方面至第六方面所获得的技术效果与第一方面中对应的技术手段获得的技术效果近似,在这里不再赘述。
附图说明
图1是本申请一个示例性实施例提供的智能车辆的结构示意图;
图2是本申请一个示例性实施例提供的参数调节装置的结构示意图;
图3是本申请一个示例性实施例提供的参数调节方法的流程图;
图4是本申请另一个示例性实施例提供的参数调节方法的流程图;
图5是本申请一个示例性实施例提供的参数调节装置的结构示意图。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚,下面将结合附图对本申请实施方式作进一步地详细描述。
在对本申请实施例提供的参数调节方法进行详细地解释说明之前,先对本申请实施例涉及的应用场景进行介绍。
目前,图像识别已经广泛应用于各个领域之中,例如,交通领域、智能安防领域、自动驾驶领域、手机终端领域等等。以自动驾驶为例,智能车辆可以安装摄像头,通过该摄像头对智能车辆周围的环境进行拍摄,以得到智能车辆周围的图像,基于ISP参数集对智能车辆周围的图像进行处理,进而对处理后的图像进行识别,以确定该图像的场景、图像中的物体等等。例如,智能车辆确定图像中的人、汽车、车道线、自行车、交通标志等等。这样,智能车辆能够基于识别出的场景和物体,实现行车提醒、辅助驾驶和自动驾驶等功能。所以,图像识别是实现自动驾驶的核心,而图像质量能够直接影响图像识别的准确率,ISP参数决定了图像质量的优劣。也就是说,ISP参数最终能够决定图像识别的准确率。
当前,可以通过人工进行ISP参数的调节。但是需要工程师反复调节ISP参数,这会导致投入的人力成本较高、效率较低,并且通过调节后的ISP参数进行图像处理后,虽然人眼对该图像的图像质量感知较好,但通过智能终端进行图像识别的准确率不佳。所以,本申请实施例提供了一种参数调节方法,能够在不投入大量人力成本的基础上,实现自主调节ISP参数,从而提高ISP参数调节的效率,还能够保证图像在符合人眼视觉的基础上,提高图像的识别率。接下来以应用于自动驾驶领域为例,对本申请实施例提供的方法进行简单介绍。请参考图1,智能车辆上搭载了计算平台,例如,移动数据中心(mobile data center,MDC)并且安装了摄像头,该计算平台相当智能车辆的大脑,能够处理各类数据。计算平台位于智能车辆内部,通过智能车辆的电池进行供电,计算平台可支持对若干(通常为几个到十几个不等)的摄像头进行供电,支撑摄像头工作。摄像头属于计算平台的外设,安装在智能车辆的挡风玻璃处,或者安装在智能车辆外部。通过该摄像头可以对智能车辆周围的环境进行拍摄,以得到智能车辆周围的图像,该摄像头拍摄得到的图像包括原始数据(raw data,RAW)以及嵌入式位图数据(embedded bitmap data,EBD),将RAW和EBD经过加串器进行加串,并通过移动行业处理器接口(mobile industry processor interface,MIPI)协议传输给解串器进行解串,进而对解串后的数据应用ISP参数集进行图像处理,以得到处理后的图像,该处理后的图像格式可以是YUV420NV12、YUV420NV21等等,然后,通过图像感知模型/算法对处理后的图像进行识别,以确定该图像的场景,将关键场景的图像存储到存储器的训练图像集中,并且存储器中还内置有测试图像集,智能车辆基于图像质量测评系统,对训练图像集以及测试图像集进行图像质量测评,进而基于图像质量测评结果对ISP参数进行调节,在确定出更优的ISP参数后,能够提示用户更新ISP参数。其中,图像质量评测系统位于计算平台的系统级芯片(system on chip,SoC)上,图像质量评测系统包含多个图像质量评测算法,该多个图像质量评测算法包括但不限于清晰度,色彩,噪声,白平衡,宽动态等图像质量评测算法。
本申请实施例提供的方法可以由任何一种具备图像信号处理功能的智能终端来执行,比如,该智能终端可以为个人计算机(personal computer,PC)、手机、个人数字助手(personal digital assistant,PDA)、掌上电脑PPC(pocket PC)、平板电脑、服务器、机器人、智能驾驶设备、车载计算平台等。本申请中的智能驾驶设备可以包括陆上交通工具、水上交通工具、空中交通工具、工业设备、农业设备、或娱乐设备等。例如智能驾驶设备可以为车辆,该车辆为广义概念上的车辆,可以是交通工具(如商用车、乘用车、摩托车、飞行车、火车等),工业车辆(如:叉车、挂车、牵引车等),工程车辆(如挖掘机、推土车、吊车等),农用设备(如割草机、收割机等),游乐设备,玩具车辆等,本申请实施例对车辆的类型不作具体限定。 再如,智能驾驶设备可以为飞机、或轮船等交通工具。
需要说明的是,本申请实施例描述的应用场景和执行主体是为了更加清楚的说明本申请实施例的技术方案,并不构成对于本申请实施例提供的技术方案的限定,本领域普通技术人员可知,随着新应用场景和智能终端的出现,本申请实施例提供的技术方案对于类似的技术问题,同样适用。
请参考图2,图2是根据本申请实施例示出的一种参数调节装置的结构示意图。该参数调节装置可以部署在智能终端上。该参数调节装置包括至少一个处理器201、通信总线202、存储器203以及至少一个通信接口204。
处理器201可以是一个通用中央处理器(central processing unit,CPU)、网络处理器(network processor,NP)、微处理器、或者可以是一个或多个用于实现本申请方案的集成电路,例如,专用集成电路(application-specific integrated circuit,ASIC),可编程逻辑器件(programmable logic device,PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device,CPLD)、现场可编程逻辑门阵列(field-programmable gate array,FPGA)、通用阵列逻辑(generic array logic,GAL)或其任意组合。
通信总线202用于在上述组件之间传送信息。通信总线202可以分为地址总线、数据总线、控制总线等。为便于表示,图中仅用一条粗线表示,但并不表示仅有一根总线或一种类型的总线。
存储器203可以是只读存储器(read-only memory,ROM),也可以是随机存取存储器(random access memory,RAM),也可以是电可擦可编程只读存储器(electrically erasable programmable read-only memory,EEPROM)、光盘(包括只读光盘(compact disc read-only memory,CD-ROM)、压缩光盘、激光盘、数字通用光盘、蓝光光盘等)、磁盘存储介质或者其它磁存储设备,或者是能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其它介质,但不限于此。存储器203可以是独立存在,并通过通信总线202与处理器201相连接。存储器203也可以和处理器201集成在一起。
通信接口204使用任何收发器一类的装置,用于与其它设备或通信网络通信。通信接口204包括有线通信接口,还可以包括无线通信接口。其中,有线通信接口例如可以为以太网接口。以太网接口可以是光接口、电接口或其组合。无线通信接口可以为无线局域网(wireless local area networks,WLAN)接口、蜂窝网络通信接口或其组合等。
在具体实现中,作为一种实施例,处理器201可以包括一个或多个CPU,如图2中所示的CPU0和CPU1。处理器201可以集成有图像处理器(graphics processing unit,GPU),GPU用于负责显示屏所需要显示的内容的渲染和绘制。一些实施例中,处理器201还可以包括人工智能(artificial intelligence,AI)处理器,该AI处理器用于处理有关机器学习、智能驾驶的计算操作。
在具体实现中,作为一种实施例,参数调节装置可以包括多个处理器,如图2中所示的处理器201和处理器205。这些处理器中的每个可以是一个单核处理器,也可以是一个多核处理器。这里的处理器可以指一个或多个设备、电路、和/或用于处理数据(如计算机程序指令)的处理核。
在具体实现中,作为一种实施例,参数调节装置还可以包括输出设备206和输入设备207。 输出设备206和处理器201通信,可以以多种方式来显示信息。例如,输出设备206可以是液晶显示器(liquid crystal display,LCD)、发光二级管(light emitting diode,LED)显示设备、阴极射线管(cathode ray tube,CRT)显示设备或投影仪(projector)等。输入设备207和处理器201通信,可以以多种方式接收用户的输入。例如,输入设备207可以是鼠标、键盘、触摸屏设备或传感设备等。
在一些实施例中,存储器203用于存储执行本申请方案的程序代码210,处理器201可以执行存储器203中存储的程序代码210。该程序代码210中可以包括一个或多个软件模块,该参数调节装置可以通过处理器201以及存储器203中的程序代码210,来实现下文图3实施例提供的参数调节方法。
图3是本申请实施例提供的一种参数调节方法的流程图。该方法的执行主体为智能终端。请参考图3,该方法包括如下步骤。
步骤301:基于训练图像集和第一ISP参数集,确定第二ISP参数集,该第一ISP参数集为智能终端对应的ISP参数集。
在一些实施例中,训练图像集包括N个子训练集,第一ISP参数集包括M个参数,该N个子训练集中的每个子训练集对应一个场景,N和M均为大于1的整数。对于该N个子训练集中的每个子训练集均执行以下操作:将其中一个子训练集作为目标子训练集,调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集,目标场景为目标子训练集对应的场景,第一子参数集包括该M个参数中的至少一个参数,第二子参数集包括该M个参数中的至少一个参数。
也即是,该N个子训练集具有一一对应的N个场景,该N个场景中的每个场景对应第一ISP参数集中的一个子参数集,每个子参数集包括该M个参数中的至少一个参数。对于该N个子训练集中的任一子训练集,将该子训练集称为目标子训练集,将目标子训练集对应的场景称为目标场景,将第一ISP参数中目标场景对应的子参数集称为第一子参数集,对第一子参数集中各个参数对应的参数值进行调节,能够得到第二ISP参数中目标场景对应的子参数集,即第二子参数集。按照相同的方式对第一ISP参数集中每个场景对应的子参数集进行调节,能够得到第二ISP参数集中每个场景对应的子参数集。
可选地,智能终端存储有子训练集与场景的对应关系,即第一对应关系,并且还存储有场景与子参数集的对应关系,即第二对应关系。这样,从第一对应关系中能够确定目标子训练集对应的场景,即目标场景,进而从第二对应关系中能够确定目标场景对应的子参数集,即第一子参数集。
为了使参数调节过程更加便捷、高效,智能终端也可以存储子训练集、场景与子参数集三者之间的对应关系。这样,可以基于目标子训练集,从该三者之间的对应关系中直接确定目标场景对应的子参数集,即第一子参数集。
第一ISP参数集是智能终端当前正在应用的ISP参数集,而且,当前可能是首次进行ISP参数的调节,也可能是非首次进行ISP参数的调节。如果当前是首次进行ISP参数的调节,那么,第一ISP参数集中每个参数对应的参数值可以是事先设置的。如果当前是非首次进行ISP参数的调节,那么,第一ISP参数集中每个参数对应的参数值是上一次进行ISP参数调节后得到的。
第一ISP参数集包括的M个参数是用来进行图像处理的各种参数。作为一种示例,该M个参数可以是自动色调再映射(automatic tone reproduction,ATR)、动态范围压缩(dynamic range compression,DRC)、伽马校正(GAMMA)、raw域降噪(raw noise fall,RAWNF)、YUV域降噪(YUV noise fall,YUVNF)、颜色校正矩阵(color correction matrix,CCM)、白平衡增益(auto white balance,AWB)等等,本申请实施例对此不做限定。
上述N个子训练集对应的N个场景可以为图像识别的关键场景。以自动驾驶为例,该N个场景为白天出入隧道、晴天逆光、夜间交通路口、红绿灯、高架桥、高速路等等。其中,该N个场景中每个场景对应的子参数集是指能够对该场景产生影响的参数的集合,不同场景对应的子参数集可能存在交集,也可能不存在交集。比如,第二对应关系如下表1所示,在表1中,白天出入隧道场景对应的子参数集包括ATR、DRC和GAMMA,晴天逆光场景对应的子参数集包括DRC和GAMMA,夜间交通路口场景对应的子参数集包括RAWNF和YUVNF,红绿灯场景对应的子参数集包括CCM和AWB。其中,白天出入隧道场景和晴天逆光场景对应的子参数集存在交集,其他场景对应的子参数集不存在交集。
表1
场景 子参数集
白天出入隧道 ATR、DRC、GAMMA
晴天逆光 DRC、GAMMA
夜间交通路口 RAWNF、YUVNF
红绿灯 CCM、AWB
在一些实施例中,调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集的实现过程包括:将第一ISP参数集中目标场景对应的第一子参数集调整为多个候选子参数集,将其余参数保持不变,以得到多个候选ISP参数集,基于该多个候选ISP参数集,确定目标子训练集对应的多个候选质量分数,该多个候选质量分数指示经过该多个候选ISP参数集处理后的目标子训练集的图像质量,将该多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中目标场景对应的参数,确定为第二子参数集。
可选地,智能终端存储有参数、调整范围与调整步长之间的对应关系,即第三对应关系。这样,从第三对应关系中能够确定第一子参数集中每个参数对应的调整范围和调整步长。对于第一子参数集中的任意一个参数,将该参数作为目标参数,基于目标参数对应的调整范围和调整步长,确定目标参数对应的多个候选参数值。按照相同的方式确定出第一子参数集中每个参数对应的多个候选参数值之后,将第一子参数集中每个参数对应的多个候选参数值进行组合,以得到多个候选子参数集。
作为一种示例,可以基于目标参数对应的调整范围和调整步长,按照如下公式(1),确定目标参数对应的多个候选参数值。
F n+1=min+n×step
Figure PCTCN2022126256-appb-000001
其中,在上述公式(1)中,F n是指目标参数对应的多个候选参数值中第n+1个候选参数值,min是指目标参数对应的调整范围中的最小值,max是指目标参数对应的调整范围中的 最大值,step是指目标参数对应的调整步长,N是指自然数。
例如,目标参数对应的调整范围为1至1.6,调整步长为0.2,则目标参数对应的多个候选参数值中第一个候选参数值为1+0=1,第二个候选参数值为1+0.2=1.2,第三个候选参数值为1+0.4=1.4,第四个候选参数值为1+0.6=1.6。
可选地,将第一子参数集中每个参数对应的多个候选参数值进行组合的实现过程包括:对于第一子参数集中的任意一个参数,从该参数对应的多个候选参数值中任意选择一个候选参数值,按照相同的方式能够选择出第一子参数集中每个参数对应的一个候选参数值,将为第一子参数集中每个参数选择出的候选参数值作为一个候选子参数集。按照相同的方式能够得到多个候选子参数集,该多个候选子参数集不同。
例如,第一子参数集中包括两个参数,第一个参数对应2个候选参数值,分别为1和1.5。第二个参数对应3个候选参数值,分别为2、4和6。将这两个参数对应的候选参数值进行组合,得到的多个候选子参数集分别为(1,2)、(1,4)、(1,6)、(1.5,2)、(1.5,4)和(1.5,6)。
可选地,将第一ISP参数集中目标场景对应的第一子参数集调整为多个候选子参数集,将其余参数保持不变,以得到多个候选ISP参数集的实现过程包括:对于该多个候选子参数集中的任意一个候选子参数集,将第一ISP参数集中目标场景对应的第一子参数集替换为该候选子参数集,将第一ISP参数集中除第一子参数集之外的其他参数保持不变,从而得到一个候选ISP参数集。对于该多个候选子参数集中的每个候选子参数集,都按照上述方式进行处理之后,能够得到多个候选ISP参数集,每个候选ISP参数集对应一个候选子参数集。
在一些实施例中,基于该多个候选ISP参数集,确定目标子训练集对应的多个候选质量分数的实现过程包括:基于该多个候选ISP参数集,分别对目标子训练集中的各个图像进行处理,以得到与该多个候选ISP参数集一一对应的多个处理后的目标子训练集,基于该多个处理后的目标子训练集,确定多个候选质量分数,该多个候选质量分数与该多个处理后的目标子训练集一一对应。
对于该多个处理后的目标子训练集中的任意一个处理后的目标子训练集,智能终端能够按照相关算法,确定该处理后的目标子训练集中每个图像的图像质量分数,以得到多个图像质量分数,基于该多个图像质量分数,确定该处理后的目标子训练集对应的候选质量分数。按照相同的方式能够得到多个候选质量分数,该多个候选质量分数与该多个处理后的目标子训练集一一对应。
在一些实施例中,可以确定该多个图像质量分数的众数,将该众数确定为处理后的目标子训练集对应的候选质量分数。在另一些实施例中,可以确定该多个图像质量分数的平均值,将该平均值确定为处理后的目标子训练集对应的候选质量分数。当然,在实际应用中,还能够通过其他的方式对该多个图像质量分数进行处理,以得到处理后的目标子训练集对应的候选质量分数,本申请实施例对此不做限定。
由于候选质量分数越大,说明ISP参数集对图像的处理效果越好,因此,将多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中目标场景对应的参数,确定为第二子参数集。这样,能够保证最终确定的第二子参数集对于图像的处理效果是该多个候选ISP参数集中最好的,进一步保证ISP参数集的调节效果。
在一些实施例中,测试图像集包括N个子测试集,该N个子测试集与N个子训练集一 一对应。智能终端在调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集之前,还可以确定第二质量分数,第二质量分数指示经过第一ISP参数集处理后的目标子训练集的图像质量,若第二质量分数小于第二参考质量分数,则按照上述方式调节第一ISP参数集中目标场景对应的第一子参数集,第二参考质量分数指示经过第一ISP参数集处理后的目标子测试集的图像质量,目标子测试集为目标场景对应的子测试集。若第二质量分数大于或等于第二参考质量分数,则不调节第一ISP参数集中目标场景对应的第一子参数集。
若第二质量分数小于第二参考质量分数,说明通过第一ISP参数集对目标子训练集中的图像的处理效果不佳,需要对目标场景对应的ISP参数进行调节,因此执行调节第一ISP参数集中目标场景对应的第一子参数集的步骤。若第二质量分数大于或等于第二参考质量分数,说明通过第一ISP参数集对目标子训练集中的图像的处理效果较好,并不需要对目标场景对应的ISP参数进行调节,因此不执行调节第一ISP参数集中目标场景对应的第一子参数集的步骤。
其中,确定第二质量分数的过程以及确定第二参考质量分数的过程,均与上述确定候选质量分数的过程类似,具体请参考上文中对应的内容,此处不再赘述。
需要说明的是,测试图像集是事先设置的,而且在不同的情况下,还可以按照不同的需求进行调整。另外,对于目标子训练集来说,可以直接按照上述方法对第一子参数集进行调节以得到第二子参数集。当然,也可以在调节之前,对经过第一ISP参数集处理后的目标子训练集进行图像打分,以判断第一ISP参数集对目标子训练集中的图像的处理效果,进而确定是否需要对第一子参数集进行调节。
按照上述方式确定出第二ISP参数集中每个场景对应的子参数集之后,能够得到N个子参数集。由于每个子参数集包括M个参数中的至少一个参数,该N个子参数集的并集可能包括该M个参数中的每个参数,也可能只包括该M个参数中的部分参数。在不同的情况下,基于该N个子参数集确定第二ISP参数集的方式不同,接下来将分别进行介绍。
第一种情况,该N个子参数集的并集包括该M个参数中的每个参数。在这种情况下,可以基于该N个子参数集确定第二ISP参数集。
基于上文描述,不同场景对应的子参数集可能存在交集,也可能不存在交集。在这些场景对应的子参数集不存在交集的情况下,可以直接将该N个子参数集进行合并,以得到第二ISP参数集。在这些场景对应的子参数集存在交集的情况下,将存在交集的这些参数称为第一类参数,将不存在交集的参数称为第二类参数,将该N个子参数集中第一类参数的参数值进行组合,以得到多种参数值组合,将该多种参数值组合与该N个子参数集中第二类参数的参数值进行合并,以得到多个第二ISP参数集。
在第一类参数的数量为1的情况下,对于第一类参数的多个参数值中的任意一个参数值,将该参数值与该N个子参数集中第二类参数的参数值进行合并,得到一个第二ISP参数集。按照相同的方式,将第一类参数的每个参数值均与该N个子参数集中第二类参数的参数值进行合并,能够得到多个第二ISP参数集。
在第一类参数的数量大于1的情况下,将第一类参数中每个参数对应的多个参数值进行组合,以得到多种参数值组合,对于该多种参数值组合中的任意一种参数值组合,将该参数值组合与该N个子参数集中第二类参数的参数值进行合并,以得到一个第二ISP参数集,按 照相同的方式,将每种参数值组合与该N个子参数集中第二类参数的参数值进行合并,能够得到多个第二ISP参数集。
其中,将第一类参数中每个参数对应的多个参数值进行组合的实现方式,与上述将第一子参数集中每个参数对应的多个候选参数值进行组合的实现方式类似,请参考上文相关内容,此处不再赘述。
第二种情况,该N个子参数集的并集包括该M个参数中的部分参数。在这种情况下,基于该N个子参数集和第一ISP参数集确定第二ISP参数集。
与上文同理,不同场景对应的子参数集可能存在交集,也可能不存在交集。在这些场景对应的子参数集不存在交集的情况下,可以直接将该N个子参数集进行合并,以得到合并参数集,将该合并参数集与第一ISP参数集中除合并参数集之外的其他参数进行再次合并,以得到第二ISP参数集。在这些场景对应的子参数集存在交集的情况下,将存在交集的这些参数称为第一类参数,将不存在交集的参数称为第二类参数,将该N个子参数集中第一类参数的参数值进行组合,以得到多种参数值组合,将该多种参数值组合与该N个子参数集中第二类参数的参数值进行合并,以得到多个合并参数集。对于该多个合并参数集中的每个合并参数集,将该合并参数集与第一ISP参数集中除该合并参数集之外的其他参数进行再次合并,能够得到多个第二ISP参数集。
在一些实施例中,基于训练图像集和第一ISP参数集确定第二ISP参数集之前,还能够获取当前环境对应的图像,基于第一ISP参数集处理当前环境对应的图像,基于处理后的图像进行场景识别,以得到场景识别结果,该场景识别结果指示当前环境中是否存在关键场景,若该场景识别结果指示当前环境中存在关键场景,则将当前环境对应的图像存储至训练图像集中。
可选地,智能终端具有摄像头,该摄像头能够对智能终端周围的环境进行拍摄,以得到当前环境对应的图像。这样,智能终端能够获取当前环境对应的图像。
在实际应用中,由于摄像头自身存在一些缺陷或者摄像头在拍摄时的光学条件不同,这会导致摄像头拍摄的图像的质量较差,难以较好地还原拍摄现场的细节。因此,需要对摄像头拍摄的图像进行后期处理,即对拍摄的图像应用第一ISP参数集进行处理,才能保证经过处理后的图像能够较好地还原拍摄现场的细节,得到质量较好的图像,进而保证场景识别结果的准确性。
其中,基于处理后的图像进行场景识别,以得到场景识别结果场的方式包括多种。作为一种示例,可以通过神经网络模型进行识别,也即是,将处理后的图像输入至神经网络模型中,以得到神经网络模型输出的场景名称,即场景识别结果。
在用神经网络模型进行识别之前需要进行神经网络模型的训练。也即是,获取多个样本图像,以及每个图像对应的场景名称,将图像作为神经网络模型的输入,将场景名称作为神经网络模型的输出,对神经网络模型进行训练。
在一些实施例中,智能终端存储有多个关键场景,这样智能终端能够确定该多个关键场景中是否存在该场景识别结果所指示的场景。若该多个关键场景中存在该场景识别结果所指示的场景,则确定该场景识别结果所指示的场景为关键场景,并将当前环境对应的图像存储至训练图像集中。若该多个关键场景中不存在该场景识别结果所指示的场景,则确定该场景识别结果所指示的场景不为关键场景,不将当前环境对应的图像存储至训练图像集中。
基于上文描述,训练图像集包括N个子训练集,每个子训练集对应一个场景。若该场景识别结果所指示的场景为关键场景,则将当前环境对应的图像存储至对应场景的子训练集中。也就是说,训练图像集中的图像可以是摄像头实时拍摄的图像。由于不同用户的行为习惯不同,进而摄像头所拍摄的图像也不同,因此,基于摄像头实时拍摄的图像进行ISP参数集的调节,能够使最终调节的ISP参数具有自适应性。当然,训练图像集中的图像也可以是通过其他方式获得的图像,本申请实施例对此不做限定。
可选地,在训练图像集中的图像为摄像头实时拍摄的情况下,智能终端还可以实时确定训练图像集中每个子训练集的图像数量。若训练图像集中每个子训练集的图像数量都达到图像数量阈值,则可以执行上述步骤301。当然,若训练图像集中存在达到图像数量阈值的子训练集,智能终端也能够先对该子训练集进行处理,以得到第二ISP参数集中该子训练集对应的子参数集,直到训练图像集中的所有子训练集都处理完成后。也就是说,智能终端可以在训练图像集中的所有子训练集都达到图像数量阈值之后,才对每个子训练集进行处理。也可以在某个子训练集的图像数量达到图像数量阈值时,直接对该子训练集进行处理,而无需等待其他子训练集的图像数量都达到图像数量阈值。
其中,图像数量阈值是事先设置的,每个子训练集所对应的图像数量阈值可以相同,也可以不同,而且在不同的情况下,还可以按照不同的需求进行调整。本申请实施例对此不做限定。
需要说明的是,上述训练图像集中的图像是未经过ISP参数集处理之后的图像。这样,在后续调节ISP参数时,能够将调节之后的ISP参数集应用至该未经ISP参数集处理的图像中,以便确定该调节之后的ISP参数集的优劣。
步骤302:基于第二ISP参数集对测试图像集进行处理。
基于上文描述,第二ISP参数集的数量可能为一个,也可能为多个。若存在多个第二ISP参数集,则基于该多个第二ISP参数集中的每个第二ISP参数集,分别对测试图像集进行处理,从而得到多个处理后的测试图像集。
作为一种示例,可以通过回灌接口获取测试图像集,该回灌接口可表示为:recharge(&test_type,&scene,&camera_type,&FOV,&picture_path)。其中,recharge表示回灌接口,&test_type用于判定是训练图像集还是测试图像集,其中1表示训练图像集,0表示测试图像集,&scene表示场景,例如进隧道、隧道内、天桥、逆光、高速公路、沙漠、雪天、雨天等等,&camera_type表示摄像头类型,&FOV表示摄像头广角类型,例如,近距(FOV120)、中距(FOV60)、远距(FOV30)、鱼眼(FOV180)等等,&picture_path表示回灌的图像路径目录。
步骤303:基于处理后的测试图像集确定第一质量分数和第一识别率,第一质量分数指示处理后的测试图像集的图像质量,第一识别率指示处理后的测试图像集的图像识别情况。
与上文同理,处理后的测试图像集可能为一个,也可能为多个。在不同的情况下,确定第一质量分数和第一识别率的方式不同,接下来将分别进行介绍。
第一种情况,处理后的测试图像集为一个。此时,基于该处理后的测试图像集确定第一质量分数和第一识别率。
确定该处理后的测试图像集包括的N个子测试集中每个子测试集的质量分数,基于该N个子测试集的质量分数,确定第一质量分数。确定该处理后的测试图像集包括的N个子测试 集中每个子测试集的识别率,基于该N个子测试集的识别率,确定第一识别率。
在一些实施例中,可以将该N个子测试集的质量分数的平均值,确定为第一质量分数。在另一些实施例中,每个子测试集对应一个权重。将该N个子测试集的质量分数分别乘以各自对应的权重之后再相加,以得到第一质量分数。
其中,确定该处理后的测试图像集中每个子测试集的质量分数的过程,与上述确定处理后的目标子训练集的候选质量分数的过程类似,请参考上文中的对应内容,此处不再赘述。
在一些实施例中,可以确定该N个子测试集的识别率的众数,将该众数确定为第一识别率。在另一些实施例中,可以确定该N个子测试集的识别率的平均值,将该平均值确定为第一识别率。当然,在实际应用中,还能够通过其他的方式确定第一识别率,本申请实施例对此不做限定。
第二种情况,处理后的测试图像集为多个。确定该多个处理后的测试图像集的质量分数和识别率,将该多个处理后的测试图像集的质量分数中的最大质量分数作为第一质量分数,将与该最大质量分数对应的识别率作为第一识别率。
对于该多个处理后的测试图像集中的每个测试图像集,确定该处理后的测试图像集包括的N个子测试集中每个子测试集的质量分数,基于该N个子测试集的质量分数,确定该处理后的测试图像集的质量分数。确定该处理后的测试图像集包括的N个子测试集中每个子测试集的识别率,基于该N个子测试集的识别率,确定该处理后的测试图像集识别率。
步骤304:若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,则将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集,第一参考质量分数指示经过第一ISP参数集处理后的测试图像集的图像质量,第一参考识别率指示经过第一ISP参数集处理后的测试图像集的图像识别情况。
其中,第一参考质量分数和第一参考识别率的确定过程,与步骤303的第一种情况中确定第一质量分数和第一识别率的过程类似,请参考上文中对应的内容,此处不再赘述。
若第一质量分数不大于第一参考质量分数和\或第一识别率不大于第一参考识别率,则不将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。即保持智能终端对应的ISP参数集不变。
若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,说明经过第二ISP参数处理后的图像质量优于第一ISP参数,并且经过第二ISP参数处理后的图像识别率也高于第一ISP参数,因此可以将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。若第一质量分数不大于第一参考质量分数和\或第一识别率不大于第一参考识别率,说明经过第二ISP参数处理后的图像质量不如第一ISP参数,并且经过第二ISP参数处理后的图像识别率也低于第一ISP参数,因此不将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。
基于上文描述,通过上述步骤301可能会确定出一个第二ISP参数集,也可能会确定出多个第二ISP参数集。在确定出一个第二ISP参数集的情况下,可以直接将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。在确定出多个第二ISP参数集的情况下,可以将智能终端对应的ISP参数集从第一ISP参数集调整为第一质量分数对应的第二ISP参数集。
可选地,若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,还 可以将第一参考识别率更新为第一识别率。或者,将第一参考识别率更新为第一识别率并且将第一参考质量分数更新为第一质量分数。若第一质量分数不大于第一参考质量分数和\或第一识别率不大于第一参考识别率,则保持第一参考识别率和第一参考质量分数不变。这样可以保证后续调节的ISP参数集比上一个ISP参数集更优,从而进一步提高ISP参数调节的效率以及图像的识别率。
在实际应用中,若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,智能终端能够触发更新提示,该更新提示用于提示用户可以对智能终端对应的ISP参数集进行更新,若智能终端接收到用户触发的更新指令,说明用户同意对智能终端对应的ISP参数集进行更新,此时,智能终端将对应的ISP参数集从第一ISP参数集调整为第二ISP参数集。
也就是说,在第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率的情况下,智能终端可以直接将智能终端将对应的ISP参数集从第一ISP参数集调整为第二ISP参数,也可以询问用户是否更新ISP参数集,在用户同意更新的情况下,才将智能终端将对应的ISP参数集从第一ISP参数集调整为第二ISP参数。
上述内容是确定出第一质量分数和第一识别率之后,再确定第一质量分数是否大于第一参考质量分数,以及第一识别率是否大于第一参考识别率。实际应用中,也可以先确定第一质量分数,在第一质量分数大于第一参考质量分数的情况下,再确定第一识别率,从而确定第一识别率是否大于第一参考识别率。本申请实施例对此不做限定。
接下来以智能车辆为例,对本申请实施例提供的一种参数调节方法进行介绍。
请参考图4,摄像头对智能车辆周围的环境进行拍摄,以得到当前环境对应的图像,进而计算平台基于第一ISP参数集对该当前环境对应的图像进行处理,并且对该处理后的图像进行识别,基于识别得到的场景识别结果,将该当前环境对应的图像存储至训练图像集中,当训练图像集中的图像达到图像数量阈值后,基于该训练图像集,确定第二ISP参数集,进而基于该第二ISP参数集对测试图像集中的图像进行处理,以确定该测试图像集的第一质量分数和第一识别率,在第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率的情况下,触发ISP参数更新提示,以提醒用户可以对ISP参数进行更新,在用户同意对ISP参数进行更新的情况下将第一ISP参数集调整为第二ISP参数集。
本申请实施例是基于训练图像集和第一ISP参数集来自动地确定第二ISP参数,无需人工反复调节ISP参数,降低了人力成本。而且,由于该第一质量分数表征经过第二ISP参数集处理后的测试图像集的图像质量,第一识别率表征经过第二ISP参数集处理后的测试图像集的图像识别情况,第一参考质量分数表征经过第一ISP参数集处理后的测试图像集的图像质量,第一参考识别率表征经过第一ISP参数集处理后的测试图像集的图像识别情况。因此,在第一质量分数大于第一参考质量分数,且第一识别率大于第一参考识别率的情况下,说明经过第二ISP参数集处理后的图像质量优于第一ISP参数集,并且经过第二ISP参数集处理后的图像识别率也高于第一ISP参数集。在这种情况下,将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集,不仅能够保证智能终端根据第二ISP参数集得到的图像的质量优于根据第一ISP参数集得到的图像质量,还能够保证智能终端根据第二ISP参数集得到的图像的识别率也优于根据第一ISP参数集得到的图像的识别率。也就是说,本申请提供的方法不仅能够在不投入大量的人力成本的基础上,自主调节ISP参数,提高ISP参数 调节的效率,还能够在保证人眼视觉的基础上,提高图像的识别率。此外,本申请实施例提供的方法还可以对第一参考识别率,或者,第一参考识别率和第一参考质量分数进行更新,这样能够保证后续调节的ISP参数集比上一个ISP参数集更优,从而进一步提高ISP参数调节的效率以及图像的识别率。并且,在本申请实施例提供的方法还可以实时拍摄图像,进而基于该实时拍摄的图像,对ISP参数集进行调节,也就是说,不同的用户所拍摄的图像不同,进而基于该拍摄的图像所调节的ISP参数集就有所不同,因此,最终调节的ISP参数具有自适应性,能够为不同的用户调节对该用户来说更好的ISP参数。
图5是本申请实施例提供的一种参数调节装置的结构示意图,该参数调节装置可以由软件、硬件或者两者的结合实现成为智能终端的部分或者全部。参见图5,该装置包括:第一确定模块501、第一处理模块502、第二确定模块503和调整模块504。
第一确定模块501,用于基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集,该第一ISP参数集为智能终端对应的ISP参数集。详细实现过程参考上述各个实施例中对应的内容,此处不再赘述。
第一处理模块502,用于基于第二ISP参数集对测试图像集进行处理。详细实现过程参考上述各个实施例中对应的内容,此处不再赘述。
第二确定模块503,用于基于处理后的测试图像集确定第一质量分数和第一识别率,该第一质量分数指示处理后的测试图像集的图像质量,该第一识别率指示处理后的测试图像集的图像识别情况。详细实现过程参考上述各个实施例中对应的内容,此处不再赘述。
调整模块504,用于若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,则将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集,该第一参考质量分数指示经过第一ISP参数集处理后的测试图像集的图像质量,该第一参考识别率指示经过第一ISP参数集处理后的测试图像集的图像识别情况。详细实现过程参考上述各个实施例中对应的内容,此处不再赘述。
可选地,训练图像集包括N个子训练集,第一ISP参数集包括M个参数,N个子训练集中的每个子训练集对应一个场景,N和M均为大于1的整数;
第一确定模块501具体用于:
对于N个子训练集中的每个子训练集,均执行以下操作:
将其中一个子训练集作为目标子训练集,调节第一ISP参数集中目标场景对应的第一子参数集,以得到第二ISP参数集中目标场景对应的第二子参数集,该目标场景为目标子训练集对应的场景,该第一子参数集包括M个参数中的至少一个参数。
可选地,第一确定模块501具体用于:
将第一ISP参数集中目标场景对应的第一子参数集调整为多个候选子参数集,将其余参数保持不变,以得到多个候选ISP参数集;
基于多个候选ISP参数集,确定目标子训练集对应的多个候选质量分数,该多个候选质量分数指示经过多个候选ISP参数集处理后的目标子训练集的图像质量;
将多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中目标场景对应的参数,确定为第二子参数集。
可选地,测试图像集包括N个子测试集,N个子测试集与N个子训练集一一对应;
第一确定模块501具体用于:
确定第二质量分数,该第二质量分数指示经过第一ISP参数集处理后的目标子训练集的图像质量;
若第二质量分数小于第二参考质量分数,则执行调节第一ISP参数集中目标场景对应的第一子参数集的步骤,该第二参考质量分数指示经过第一ISP参数集处理后的目标子测试集的图像质量,该目标子测试集为目标场景对应的子测试集。
可选地,该装置还包括:
更新模块,用于若第一质量分数大于第一参考质量分数且第一识别率大于第一参考识别率,则将第一参考识别率更新为第一识别率,或者,将第一参考识别率更新为第一识别率,并且将第一参考质量分数更新为第一质量分数。
可选地,该装置还包括:
获取模块,用于获取当前环境对应的图像;
第二处理模块,用于基于第一ISP参数集处理当前环境对应的图像;
识别模块,用于基于处理后的图像进行场景识别,以得到场景识别结果,该场景识别结果指示当前环境中是否存在关键场景;
存储模块,用于若场景识别结果指示当前环境中存在关键场景,则将当前环境对应的图像存储至训练图像集中。
本申请实施例是基于训练图像集和第一ISP参数集来自动地确定第二ISP参数,无需人工反复调节ISP参数,降低了人力成本。而且,由于该第一质量分数表征经过第二ISP参数集处理后的测试图像集的图像质量,第一识别率表征经过第二ISP参数集处理后的测试图像集的图像识别情况,第一参考质量分数表征经过第一ISP参数集处理后的测试图像集的图像质量,第一参考识别率表征经过第一ISP参数集处理后的测试图像集的图像识别情况。因此,在第一质量分数大于第一参考质量分数,且第一识别率大于第一参考识别率的情况下,说明经过第二ISP参数集处理后的图像质量优于第一ISP参数集,并且经过第二ISP参数集处理后的图像识别率也高于第一ISP参数集。在这种情况下,将智能终端对应的ISP参数集从第一ISP参数集调整为第二ISP参数集,不仅能够保证智能终端根据第二ISP参数集得到的图像的质量优于根据第一ISP参数集得到的图像质量,还能够保证智能终端根据第二ISP参数据得到的图像的识别率也优于根据第一ISP参数集得到的图像的识别率。也就是说,本申请提供的方法不仅能够在不投入大量的人力成本的基础上,自主调节ISP参数,提高ISP参数调节的效率,还能够在保证人眼视觉的基础上,提高图像的识别率。此外,本申请实施例提供的方法还可以对第一参考识别率,或者,第一参考识别率和第一参考质量分数进行更新,这样能够保证后续调节的ISP参数集比上一个ISP参数集更优,从而进一步提高ISP参数调节的效率以及图像的识别率。并且,在本申请实施例提供的方法还可以实时拍摄图像,进而基于该实时拍摄的图像,对ISP参数集进行调节,也就是说,不同的用户所拍摄的图像不同,进而基于该拍摄的图像所调节的ISP参数集就有所不同,因此,最终调节的ISP参数具有自适应性,能够为不同的用户调节对该用户来说更好的ISP参数。
需要说明的是:上述实施例提供的参数调节装置在进行参数调节时,仅以上述各功能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模块完成,即将装置的内部结构划分成不同的功能模块,以完成以上描述的全部或者部分功能。另 外,上述实施例提供的参数调节装置与参数调节方法实施例属于同一构思,其具体实现过程详见方法实施例,这里不再赘述。
在上述实施例中,可以全部或部分地通过软件、硬件、固件或者其任意结合来实现。当使用软件实现时,可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行所述计算机指令时,全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络或其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中,或者从一个计算机可读存储介质向另一个计算机可读存储介质传输,例如,所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如:同轴电缆、光纤、数据用户线(digital subscriber line,DSL))或无线(例如:红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质,或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质(例如:软盘、硬盘、磁带)、光介质(例如:数字通用光盘(digital versatile disc,DVD))或半导体介质(例如:固态硬盘(solid state disk,SSD))等。值得注意的是,本申请实施例提到的计算机可读存储介质可以为非易失性存储介质,换句话说,可以是非瞬时性存储介质。
应当理解的是,本文提及的“多个”是指两个或两个以上。在本申请实施例的描述中,除非另有说明,“/”表示或的意思,例如,A/B可以表示A或B;本文中的“和/或”仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。另外,为了便于清楚描述本申请实施例的技术方案,在本申请实施例中,采用了“第一”、“第二”等字样对功能和作用基本相同的相同项或相似项进行区分。本领域技术人员可以理解“第一”、“第二”等字样并不对数量和执行次序进行限定,并且“第一”、“第二”等字样也并不限定一定不同。
需要说明的是,本申请实施例所涉及的信息(包括但不限于用户设备信息、用户个人信息等)、数据(包括但不限于用于分析的数据、存储的数据、展示的数据等)以及信号,均为经用户授权或者经过各方充分授权的,且相关数据的收集、使用和处理需要遵守相关国家和地区的相关法律法规和标准。例如,本申请实施例中涉及到的训练图像集、测试图像集和第一ISP参数集都是在充分授权的情况下获取的。
以上所述为本申请提供的实施例,并不用以限制本申请,凡在本申请的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本申请的保护范围之内。

Claims (16)

  1. 一种参数调节方法,其特征在于,所述方法包括:
    基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集,所述第一ISP参数集为智能终端对应的ISP参数集;
    基于所述第二ISP参数集对测试图像集进行处理;
    基于处理后的测试图像集确定第一质量分数和第一识别率,所述第一质量分数指示处理后的测试图像集的图像质量,所述第一识别率指示处理后的测试图像集的图像识别情况;
    若所述第一质量分数大于第一参考质量分数且所述第一识别率大于第一参考识别率,则将所述智能终端对应的ISP参数集从所述第一ISP参数集调整为所述第二ISP参数集,所述第一参考质量分数指示经过所述第一ISP参数集处理后的所述测试图像集的图像质量,所述第一参考识别率指示经过所述第一ISP参数集处理后的所述测试图像集的图像识别情况。
  2. 如权利要求1所述的方法,其特征在于,所述训练图像集包括N个子训练集,所述第一ISP参数集包括M个参数,所述N个子训练集中的每个子训练集对应一个场景,N和M均为大于1的整数;
    所述基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集,包括:
    对于所述N个子训练集中的每个子训练集,均执行以下操作:
    将其中一个子训练集作为目标子训练集,调节所述第一ISP参数集中目标场景对应的第一子参数集,以得到所述第二ISP参数集中所述目标场景对应的第二子参数集,所述目标场景为所述目标子训练集对应的场景,所述第一子参数集包括所述M个参数中的至少一个参数。
  3. 如权利要求2所述的方法,其特征在于,所述调节所述第一ISP参数集中目标场景对应的第一子参数集,以得到所述第二ISP参数集中所述目标场景对应的第二子参数集,包括:
    将所述第一ISP参数集中所述目标场景对应的第一子参数集调整为多个候选子参数集,将其余参数保持不变,以得到多个候选ISP参数集;
    基于所述多个候选ISP参数集,确定所述目标子训练集对应的多个候选质量分数,所述多个候选质量分数指示经过所述多个候选ISP参数集处理后的所述目标子训练集的图像质量;
    将所述多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中所述目标场景对应的参数,确定为所述第二子参数集。
  4. 如权利要求2或3所述的方法,其特征在于,所述测试图像集包括N个子测试集,所述N个子测试集与所述N个子训练集一一对应;
    所述调节所述第一ISP参数集中目标场景对应的第一子参数集,以得到所述第二ISP参数集中所述目标场景对应的第二子参数集之前,所述方法还包括:
    确定第二质量分数,所述第二质量分数指示经过所述第一ISP参数集处理后的所述目标子训练集的图像质量;
    若所述第二质量分数小于第二参考质量分数,则执行所述调节所述第一ISP参数集中目 标场景对应的第一子参数集的步骤,所述第二参考质量分数指示经过所述第一ISP参数集处理后的目标子测试集的图像质量,所述目标子测试集为所述目标场景对应的子测试集。
  5. 如权利要求1-4任一所述的方法,其特征在于,所述方法还包括:
    若所述第一质量分数大于所述第一参考质量分数且所述第一识别率大于所述第一参考识别率,则将所述第一参考识别率更新为所述第一识别率,或者,将所述第一参考识别率更新为所述第一识别率,并且将所述第一参考质量分数更新为所述第一质量分数。
  6. 如权利要求1-5任一所述的方法,其特征在于,所述基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集之前,所述方法还包括:
    获取当前环境对应的图像;
    基于所述第一ISP参数集处理所述当前环境对应的图像;
    基于处理后的图像进行场景识别,以得到场景识别结果,所述场景识别结果指示所述当前环境中是否存在关键场景;
    若所述场景识别结果指示所述当前环境中存在所述关键场景,则将所述当前环境对应的图像存储至所述训练图像集中。
  7. 一种参数调节装置,其特征在于,所述装置包括:
    第一确定模块,用于基于训练图像集和第一图像信号处理ISP参数集,确定第二ISP参数集,所述第一ISP参数集为智能终端对应的ISP参数集;
    第一处理模块,用于基于所述第二ISP参数集对测试图像集进行处理;
    第二确定模块,用于基于处理后的测试图像集确定第一质量分数和第一识别率,所述第一质量分数指示处理后的测试图像集的图像质量,所述第一识别率指示处理后的测试图像集的图像识别情况;
    调整模块,用于若所述第一质量分数大于第一参考质量分数且所述第一识别率大于第一参考识别率,则将所述智能终端对应的ISP参数集从所述第一ISP参数集调整为所述第二ISP参数集,所述第一参考质量分数指示经过所述第一ISP参数集处理后的所述测试图像集的图像质量,所述第一参考识别率指示经过所述第一ISP参数集处理后的所述测试图像集的图像识别情况。
  8. 如权利要求7所述的装置,其特征在于,所述训练图像集包括N个子训练集,所述第一ISP参数集包括M个参数,所述N个子训练集中的每个子训练集对应一个场景,N和M均为大于1的整数;
    所述第一确定模块具体用于:
    对于所述N个子训练集中的每个子训练集,均执行以下操作:
    将其中一个子训练集作为目标子训练集,调节所述第一ISP参数集中目标场景对应的第一子参数集,以得到所述第二ISP参数集中所述目标场景对应的第二子参数集,所述目标场景为所述目标子训练集对应的场景,所述第一子参数集包括所述M个参数中的至少一个参数。
  9. 如权利要求8所述的装置,其特征在于,所述第一确定模块具体用于:
    将所述第一ISP参数集中所述目标场景对应的第一子参数集调整为多个候选子参数集,将其余参数保持不变,以得到多个候选ISP参数集;
    基于所述多个候选ISP参数集,确定所述目标子训练集对应的多个候选质量分数,所述多个候选质量分数指示经过所述多个候选ISP参数集处理后的所述目标子训练集的图像质量;
    将所述多个候选质量分数中的最大候选质量分数对应的候选ISP参数集中所述目标场景对应的参数,确定为所述第二子参数集。
  10. 如权利要求8或9所述的装置,其特征在于,所述测试图像集包括N个子测试集,所述N个子测试集与所述N个子训练集一一对应;
    所述第一确定模块具体用于:
    确定第二质量分数,所述第二质量分数指示经过所述第一ISP参数集处理后的所述目标子训练集的图像质量;
    若所述第二质量分数小于第二参考质量分数,则执行所述调节所述第一ISP参数集中目标场景对应的第一子参数集的步骤,所述第二参考质量分数指示经过所述第一ISP参数集处理后的目标子测试集的图像质量,所述目标子测试集为所述目标场景对应的子测试集。
  11. 如权利要求7-10任一所述的装置,其特征在于,所述装置还包括:
    更新模块,用于若所述第一质量分数大于所述第一参考质量分数且所述第一识别率大于所述第一参考识别率,则将所述第一参考识别率更新为所述第一识别率,或者,将所述第一参考识别率更新为所述第一识别率,并且将所述第一参考质量分数更新为所述第一质量分数。
  12. 如权利要求7-11任一所述的装置,其特征在于,所述装置还包括:
    获取模块,用于获取当前环境对应的图像;
    第二处理模块,用于基于所述第一ISP参数集处理所述当前环境对应的图像;
    识别模块,用于基于处理后的图像进行场景识别,以得到场景识别结果,所述场景识别结果指示所述当前环境中是否存在关键场景;
    存储模块,用于若所述场景识别结果指示所述当前环境中存在所述关键场景,则将所述当前环境对应的图像存储至所述训练图像集中。
  13. 一种参数调节装置,其特征在于,所述参数调节装置包括存储器和处理器,所述存储器用于存储计算机程序,所述处理器被配置为用于执行所述存储器中存储的计算机程序,以实现权利要求1-6任一项所述方法。
  14. 一种智能终端,其特征在于,所述智能终端包括权利要求7-13中任意一项所述的参数调节装置。
  15. 一种计算机可读存储介质,其特征在于,所述存储介质内存储有指令,当所述指令在所述计算机上运行时,使得所述计算机执行权利要求1-6任一所述的方法。
  16. 一种计算机程序,其特征在于,所述计算机程序包括指令,当所述指令在所述计算机上运行时,使得所述计算机执行权利要求1-6任一项所述方法。
PCT/CN2022/126256 2022-10-19 2022-10-19 参数调节方法、装置以及智能终端 Ceased WO2024082183A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2022/126256 WO2024082183A1 (zh) 2022-10-19 2022-10-19 参数调节方法、装置以及智能终端
CN202280100051.4A CN119856204A (zh) 2022-10-19 2022-10-19 参数调节方法、装置以及智能终端

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/126256 WO2024082183A1 (zh) 2022-10-19 2022-10-19 参数调节方法、装置以及智能终端

Publications (1)

Publication Number Publication Date
WO2024082183A1 true WO2024082183A1 (zh) 2024-04-25

Family

ID=90736517

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/126256 Ceased WO2024082183A1 (zh) 2022-10-19 2022-10-19 参数调节方法、装置以及智能终端

Country Status (2)

Country Link
CN (1) CN119856204A (zh)
WO (1) WO2024082183A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119559077A (zh) * 2025-01-24 2025-03-04 人民中科(北京)智能技术有限公司 基于多智能体强化学习下的isp调参方法和设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160284095A1 (en) * 2015-03-27 2016-09-29 Edmond Chalom Machine learning of real-time image capture parameters
WO2019152534A1 (en) * 2018-01-30 2019-08-08 Qualcomm Incorporated Systems and methods for image signal processor tuning
WO2019152499A1 (en) * 2018-01-30 2019-08-08 Qualcomm Incorporated Systems and methods for image signal processor tuning using a reference image
WO2019233341A1 (zh) * 2018-06-08 2019-12-12 Oppo广东移动通信有限公司 图像处理方法、装置、计算机可读存储介质和计算机设备
CN113705597A (zh) * 2021-03-05 2021-11-26 腾讯科技(北京)有限公司 一种图像处理方法、装置、计算机设备以及可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160284095A1 (en) * 2015-03-27 2016-09-29 Edmond Chalom Machine learning of real-time image capture parameters
WO2019152534A1 (en) * 2018-01-30 2019-08-08 Qualcomm Incorporated Systems and methods for image signal processor tuning
WO2019152499A1 (en) * 2018-01-30 2019-08-08 Qualcomm Incorporated Systems and methods for image signal processor tuning using a reference image
WO2019233341A1 (zh) * 2018-06-08 2019-12-12 Oppo广东移动通信有限公司 图像处理方法、装置、计算机可读存储介质和计算机设备
CN113705597A (zh) * 2021-03-05 2021-11-26 腾讯科技(北京)有限公司 一种图像处理方法、装置、计算机设备以及可读存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119559077A (zh) * 2025-01-24 2025-03-04 人民中科(北京)智能技术有限公司 基于多智能体强化学习下的isp调参方法和设备

Also Published As

Publication number Publication date
CN119856204A (zh) 2025-04-18

Similar Documents

Publication Publication Date Title
US12033373B2 (en) Dynamic image recognition model updates
CN111114554B (zh) 行驶轨迹预测方法、装置、终端及存储介质
CN110610463A (zh) 一种图像增强方法及装置
CN114385662B (zh) 路网更新方法、装置、存储介质及电子设备
EP4468724A1 (en) Image processing method and apparatus, terminal device and storage medium
CN110579359A (zh) 自动驾驶失效场景库的优化方法及系统、服务器及介质
CN108648495A (zh) 一种智能实时显示公交车拥挤程度的方法及系统
CN116343158B (zh) 车道线检测模型的训练方法、装置、设备及存储介质
CN117593624B (zh) 检测模型蒸馏方法、目标检测方法、装置及电子设备
CN113246991A (zh) 用于无人车的远程驾驶端的数据传输方法和装置
CN113421191A (zh) 图像处理方法、装置、设备及存储介质
CN111127358A (zh) 图像处理方法、装置及存储介质
WO2024082183A1 (zh) 参数调节方法、装置以及智能终端
CN113705427B (zh) 基于车规级芯片SoC的疲劳驾驶监测预警方法及系统
CN114998764A (zh) 基于深度学习及目标检测的架空线路图像检测方法及系统
CN108288388A (zh) 一种智能交通监控系统
CN116363619A (zh) 一种注意力机制的车道线检测方法、系统、设备及介质
CN111160206A (zh) 一种交通环境元素视觉感知方法及装置
CN113409405B (zh) 评估相机标定位置的方法、装置、设备和存储介质
CN115424242A (zh) 基于改进的yolov3交通标志识别方法、设备和介质
CN120472406A (zh) 目标识别方法、装置、控制器、车辆及存储介质
US20240212097A1 (en) Image enhancement targeted at addressing degradations caused by environmental conditions
CN106960188A (zh) 天气图像分类方法及装置
CN111950468A (zh) 一种基于卷积神经网络的交通状况检测方法、系统及存储介质
CN114494518B (zh) 车道线的生成方法、装置、设备和存储介质

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: 22962373

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202280100051.4

Country of ref document: CN

WWP Wipo information: published in national office

Ref document number: 202280100051.4

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22962373

Country of ref document: EP

Kind code of ref document: A1