Image quality parameter adjustment method and device, and vehicle-mounted chip SoC

Abstract

The application provides a kind of image quality parameter adjustment method, device and vehicle-mounted chip SoC, the method includes: obtaining target image after being processed by image signal processor, carries out semantic segmentation to target image, and obtains semantic region mask;According to the image quality index of entity region determined according to semantic region mask, the image quality evaluation parameter of entity region is determined according to the multiple semantic category weights and image quality index corresponding to semantic region mask, the image quality optimization parameter in image signal processor is searched and optimized according to image quality evaluation parameter, and image quality adjustment parameter is obtained;Image quality adjustment parameter is written into image signal processor, to adjust the processing of subsequent acquisition image by image signal processor, so, the automatic adjustment of vehicle-mounted image quality is realized, and the quality and stability of image around vehicle under complex road environment are improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, apparatus and automotive SoC for adjusting image quality parameters. Background Technology

[0002] In existing in-vehicle image processing systems, image quality tuning solutions typically involve engineers manually adjusting parameters such as sharpening and noise reduction in the ISP in laboratory or road scenes. They also create parameter tables based on conditions such as illumination, gain, and exposure time to optimize the in-vehicle image quality.

[0003] However, in the actual operation of vehicle image quality processing systems, weather, illumination, and road environments are complex and variable, making it difficult for preset parameter tables to cover all operating conditions. At the same time, existing solutions typically perform uniform quality evaluation based on the entire frame of the image, failing to distinguish the different image quality (such as sharpness, brightness, and contrast) requirements of different semantic regions such as the sky, road surface, vehicles, and pedestrians. This leads to problems of over-processing (such as over-sharpening and over-denoising) in some areas. Furthermore, due to the coupling relationship between different image processing parameters and the difficulty in coordinating their adjustment, existing adjustment methods are unable to improve image quality for different image regions or suppress image side effects. This results in a decline in the visual presentation quality of key targets, thereby affecting the driver's accuracy in judging the road conditions around the vehicle.

[0004] Therefore, how to improve the quality and stability of images displayed in vehicle-mounted image processing systems under complex road conditions is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a method, apparatus, and automotive SoC for adjusting image quality parameters, aiming to solve the problems of existing automotive image parameter tuning methods being unable to differentiate the image quality (such as sharpness, brightness, contrast, etc.) requirements of different semantic regions in the image, resulting in over-processing of some areas (such as over-sharpening, over-noise reduction), and the coupling relationship between different image processing parameters, making it difficult to coordinate adjustment, which leads to a decrease in the visual presentation quality of key targets, thereby affecting the accuracy of the driver's judgment of the road conditions around the vehicle.

[0006] To address the above issues, the following solutions are provided: In a first aspect, this application provides a method for adjusting image quality parameters, applied to the main processor of an in-vehicle image processing system, the method comprising: Acquire the target image after it has been processed by the image signal processor; The target image is semantically segmented to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions within the road environment in the target image; the entity regions include vehicles, pedestrians, road surface, sky, plants, and buildings; The image quality index corresponding to the entity region is determined based on the semantic region mask; the image quality evaluation parameters corresponding to the entity region are determined based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index. Based on the image quality assessment parameters, the image quality optimization parameters in the image signal processor are searched and optimized to obtain the image quality adjustment parameters; The image quality adjustment parameters are written to the image signal processor to adjust the image signal processor's processing of subsequently acquired images.

[0007] Secondly, this application provides an image quality parameter adjustment device, the device comprising: An image signal processor is used to process acquired images to obtain the target image; A neural network processor is used to perform semantic segmentation on the target image to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions within the road environment in the target image; the entity regions include vehicles, pedestrians, road surface, sky, plants, and buildings; The indicator statistics module is used to determine the image quality indicators corresponding to the entity region based on the semantic region mask. The main processor is configured to determine the image quality evaluation parameters corresponding to the entity region based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index; search and optimize the image quality optimization parameters in the image signal processor based on the image quality evaluation parameters to obtain image quality adjustment parameters; and write the image quality adjustment parameters to the image signal processor to adjust the image signal processor's processing of subsequently acquired images.

[0008] Thirdly, this application provides an in-vehicle chip SoC, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for performing the steps in any of the methods in the first aspect of this application.

[0009] Fourthly, this application provides a computer storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in any method of the first aspect of this application.

[0010] Fifthly, this application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in any method of the first aspect of this application. The computer program product may be a software installation package.

[0011] By implementing the embodiments of this application, the following beneficial effects are achieved: This application provides a method, apparatus, and automotive SoC for adjusting image quality parameters, applied to the main processor of an automotive image processing system. The method involves: acquiring a target image processed by an image signal processor; performing semantic segmentation on the target image to obtain a semantic region mask; the semantic region mask corresponding to the mask of entity regions within the road environment of the target image; entity regions including vehicles, pedestrians, road surface, sky, plants, and buildings; determining image quality indicators corresponding to the entity regions based on the semantic region mask; determining image quality evaluation parameters corresponding to the entity regions based on the weights of multiple semantic categories corresponding to the semantic region mask and the image quality indicators; then, searching and optimizing image quality optimization parameters in the image signal processor based on the image quality evaluation parameters to obtain image quality adjustment parameters; and finally, writing the image quality adjustment parameters into the image signal processor to adjust the processing of subsequently acquired images. Thus, by distinguishing different entity regions in the target image through semantic region masks and fusing and evaluating the image quality indicators of each entity region based on the weights corresponding to different semantic categories, the image quality evaluation process can be linked to the visual needs of different regions, avoiding the problem of insufficient or excessive image quality enhancement in some regions due to a uniform evaluation of the entire frame. At the same time, the image quality optimization parameters are searched and optimized based on the fused image quality evaluation parameters, enabling different image quality optimization processes to be coordinated and adjusted based on the image quality evaluation results, thereby improving the visual performance of the image while reducing the generation of image side effects. Furthermore, by writing the optimized image quality adjustment parameters into the image signal processor, the image signal processor can adjust the subsequent image processing process according to the optimization results, realizing the automatic adjustment of vehicle image quality parameters and improving the quality and stability of images around vehicles in complex road environments. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1This is an architecture diagram of an in-vehicle image processing system provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an automotive chip SoC provided in an embodiment of this application; Figure 3 This is a flowchart illustrating a method for adjusting image quality parameters provided in an embodiment of this application; Figure 4 This is a schematic flowchart illustrating a method for calculating an image side effect penalty index, provided in an embodiment of this application. Figure 5 This is a schematic diagram illustrating how edge halo parameters are determined based on brightness profile features, as provided in an embodiment of this application. Figure 6 This is a flowchart of an iterative adjustment process for ISP parameters based on feedback optimization, provided in an embodiment of this application. Figure 7 This is a flowchart illustrating an automatic adjustment of image quality parameters provided in an embodiment of this application; Figure 8 This is a block diagram of an image quality parameter adjustment method provided in an embodiment of this application; Figure 9 This is a block diagram of the functional modules of an image quality parameter adjustment device provided in an embodiment of this application. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0015] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0016] It should be understood that the character " / " in this document indicates that the objects before and after it are in an "or" relationship. In the embodiments of this application, "multiple" refers to two or more.

[0017] In the embodiments of this application, "at least one item" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items, which means one or more, and "multiple" means two or more.

[0018] In this application, the term "connection" refers to various connection methods, such as direct connection or indirect connection, to achieve communication between devices. This application does not impose any limitations on this.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] The following is combined Figure 1 The system architecture of an image quality parameter adjustment method according to an embodiment of this application is described. Figure 1 This is an architecture diagram of an in-vehicle image processing system provided in an embodiment of this application. The in-vehicle image processing system 100 includes a camera module 110 and an in-vehicle chip 120.

[0021] The camera module 110 is used to acquire image data of the environment surrounding the vehicle and send the acquired image data to the image signal processing module 121 in the vehicle chip 120 for processing. In some embodiments, the camera module 110 can be located at the side, rear, or front of the vehicle to acquire image information of target objects including vehicles, pedestrians, roads, buildings, and environmental backgrounds, and output the image information to the vehicle chip 120 to enable the vehicle image processing system to display the environment surrounding the vehicle.

[0022] The vehicle-mounted chip 120 is used to perform image quality analysis and image quality parameter adjustment processing on the image data output by the camera module 110 to obtain image processing parameters that meet the image quality requirements. Specifically, the vehicle-mounted chip 120 may include an image signal processing module 121, an NPU 122, and a control processing module 123. The image signal processing module 121 performs image signal processing on the image data acquired by the camera module 110 to obtain an output image after ISP processing. The image signal processing module 121 may include an ISP parameter control unit 12, which receives the parameter adjustment results output by the control processing module 123 and adjusts the image processing parameters in the image signal processing module 121 according to the parameter adjustment results. The NPU 122 performs semantic recognition processing on the image output by the image signal processing module 121 to obtain corresponding semantic segmentation results. The NPU 122 includes a semantic segmentation unit 22, which can perform semantic segmentation on low-resolution images after downsampling, frame extraction, or cropping to obtain mask information corresponding to different semantic regions. The control processing module 123 analyzes the current image quality based on the semantic segmentation results output by the NPU 122 and the image data output by the image signal processing module 121, and generates control information for adjusting ISP parameters based on the analysis results. The control processing module 123 includes an index statistics unit 31, a parameter search unit 32, and a mask mapping unit 33. The mask mapping unit 33 maps the semantic region mask output by the semantic segmentation unit 22 to the corresponding location region of the image output by the image signal processing module 121, or to a preset image statistics region, thereby determining the image evaluation region corresponding to different semantic categories. The index statistics unit 31 performs statistical analysis on image quality-related indicators based on the different semantic regions determined by the mask mapping unit 33. The parameter search unit 32 determines the comprehensive evaluation result corresponding to the current image based on the image quality indicators output by the index statistics unit, and searches for target ISP parameter combinations based on the comprehensive evaluation result. In one possible embodiment, the parameter search unit 32 can employ a search optimization algorithm to iteratively adjust the sharpening and noise reduction parameters while satisfying parameter safety boundaries, and send the determined target parameters to the ISP parameter control unit 12.

[0023] As can be seen, through the architecture of the aforementioned vehicle-mounted image processing system 100, the camera module 110 can acquire images of the vehicle's surrounding environment, and these images are collaboratively processed by the image signal processing module 121, NPU 122, and control processing module 123 in the vehicle-mounted chip 120. This enables a closed-loop image quality parameter adjustment process based on semantic region recognition, image quality index statistics, and parameter search optimization. The vehicle-mounted image processing system 100 can employ differentiated evaluation strategies for different semantic regions, improving image quality in key areas such as vehicles, pedestrians, and roads while effectively suppressing image side effects such as noise, halo, jagged edges, and artifacts. This enhances the image display performance of the vehicle-mounted image processing system in complex driving environments.

[0024] The following is combined Figure 2 The vehicle-mounted chip in the embodiments of this application will be described. Figure 2 This is a schematic diagram of the structure of an automotive chip SoC provided in an embodiment of this application, as shown below. Figure 2 As shown, the vehicle-mounted chip SoC includes a processor 210, a memory 220, a communication interface 230, and one or more programs 221. The processor 210 is connected to the memory 220 and the communication interface 230 via an internal communication bus.

[0025] The processor 210 is primarily used to execute some or all of the steps described in any of the methods in the embodiments of the image quality parameter adjustment method described below. The processor 210 may be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), or any combination thereof. It may implement or execute various exemplary logic blocks and units described in conjunction with the disclosure of this application. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The communication unit may be a communication interface, transceiver, transceiver circuit, etc., and the storage unit may be a memory.

[0026] The one or more programs 221 are stored in the memory 220 and configured to be executed by the processor 210. The one or more programs 221 include instructions for performing any step in the above method embodiments.

[0027] The memory 220 may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0028] It is understood that the automotive SoC may include more or fewer structural elements than those shown in the block diagram above, and this is not limited thereto. It is also understood that the automotive SoC may incorporate elements such as... Figure 1 The architecture of the in-vehicle image processing system.

[0029] After understanding the software and hardware architecture of this application, the following will be combined with... Figure 3 This application describes a method for adjusting an image quality parameter according to an embodiment. Figure 3 This is a flowchart illustrating a method for adjusting image quality parameters according to an embodiment of this application. The method is applied to the main processor of an in-vehicle image processor system, and specifically includes the following steps: Step S301: Obtain the target image after processing by the image signal processor.

[0030] The target image can be an image frame output by an image acquisition device (camera) in an electronic rearview mirror system (Camera-Monitor System, CMS) after capturing road environment images and performing image processing operations by an image signal processor (ISP). Simultaneously, the image signal processor can also be used to perform image enhancement processing on the acquired raw image data, including at least one or more of sharpening, noise reduction, and edge enhancement processing.

[0031] Specifically, the camera device in the CMS continuously acquires raw images of the road environment surrounding the vehicle and inputs these raw images to the image signal processor. The image signal processor processes the raw images according to the currently configured image processing parameters to obtain the target image. These image processing parameters may include sharpening parameters, noise reduction parameters, and other parameters used to control image quality.

[0032] It is worth noting that this embodiment does not directly perform image quality analysis based on the original image captured by the camera device, but rather acquires the target image after it has been processed by the current image signal processor. On the one hand, since the target image can reflect the actual image enhancement effect produced by the current image signal processing parameters, quality evaluation based on the target image can accurately obtain the actual impact of the current sharpening and noise reduction parameters. On the other hand, by performing subsequent quality index calculations on the target image and adjusting the image processing parameters in the image signal processor based on the calculation results, the image signal processor can automatically adjust the image processing effect according to changes in the actual road environment.

[0033] Step S302: Perform semantic segmentation on the target image to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions in the road environment of the target image; the entity regions include vehicles, pedestrians, road surface, sky, plants, and buildings.

[0034] The semantic region mask represents the pixel distribution corresponding to different semantic categories in the target image. That is, by semantically labeling each pixel in the target image, the pixel positions belonging to different entity regions are determined. The semantic categories can be pre-set according to the display requirements of the CMS. The semantic region mask can be represented in binary mask form or in category label matrix form, with different mask labels corresponding to entity regions of different semantic categories. In other words, the semantic region mask is a two-dimensional pixel matrix matching the size of the original image. Each pixel carries a category label and confidence weight, which is used to divide different object / scene regions in the image and mark which semantic object each pixel belongs to (e.g., sky, road surface, vehicle, pedestrian, plant, building, etc.).

[0035] Specifically, firstly, the target image is input into a pre-deployed semantic segmentation model. The semantic segmentation model extracts features and identifies categories of road environment information in the target image, obtaining the semantic segmentation result corresponding to the target image. The semantic segmentation model can be deployed in the Neural Processing Unit (NPU) of the onboard processor to utilize edge hardware acceleration capabilities to complete semantic region recognition of the target image. In some embodiments, to reduce the computational resource consumption during real-time CMS operation, scaling, frame extraction, or region cropping can be performed on the target image before semantic segmentation to generate a low-resolution semantic segmentation input map. Semantic segmentation processing is then performed based on this low-resolution input map. Subsequently, the obtained low-resolution semantic region result is mapped to the corresponding original image region of the target image to obtain a semantic region mask matching the target image. Further, after obtaining the semantic region mask, the processor can determine the location range corresponding to different entity regions in the target image based on the semantic region mask.

[0036] As can be seen, by introducing semantic region masks, this embodiment can provide a basis for region division for subsequent adaptive optimization of ISP parameters, thereby improving the matching degree between the image quality adjustment process and the actual driving scenario.

[0037] Step S303: Determine the image quality index corresponding to the entity region based on the semantic region mask; determine the image quality evaluation parameters corresponding to the entity region based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index.

[0038] Among them, the image quality index characterizes the change in image quality of the entity region after processing by the image signal processor. The image quality index may include the image quality gain index characterizing the image detail enhancement effect, and the side effect penalty index used to characterize the undesirable visual effects generated during the image enhancement process.

[0039] Specifically, the target image is first divided into different entity regions, and pixel sets corresponding to vehicles, pedestrians, road surfaces, sky, plants, and buildings are obtained respectively. Then, image quality metrics are calculated for the pixel sets corresponding to each entity region. Specifically, the image quality gain metric for each entity region can be determined by analyzing the brightness gradient changes, texture distribution features, and image changes between consecutive frames within the entity region. Simultaneously, the degree of side effects generated during image enhancement can be determined by detecting brightness transition features and edge step change features after edge enhancement within the entity region, thus obtaining a side effect penalty metric. Furthermore, since different entity regions have varying importance and image quality requirements in the vehicle driving scenario, the image quality metrics for different entity regions are weighted and fused based on multiple semantic category weight parameters corresponding to the semantic region mask to obtain image quality evaluation parameters for each entity region. The semantic category weight parameters can be determined based on information such as the semantic category of the entity region, region area, semantic segmentation confidence, and the current driving state of the vehicle. For example, when the vehicle area is within the driver's primary observation range, or when the vehicle target is close, the weight parameters corresponding to the vehicle area can be increased, making the image optimization process focus more on the vehicle outline and edge details. When the sky area occupies a large proportion, the weight related to sky area sharpness enhancement can be reduced, and the contribution related to noise suppression evaluation can be increased to avoid obvious noise or artifacts in the sky area due to over-sharpening. In some embodiments, a comprehensive image quality evaluation function can be constructed based on sharpness gain indicators and side effect penalty indicators. By inputting the image quality indicators corresponding to different semantic categories into the evaluation function and combining them with the corresponding semantic category weight parameters, image quality evaluation parameters used to evaluate the processing effect of the current ISP parameters can be obtained.

[0040] It can be seen that by jointly evaluating the image quality gains and side effect penalties, we can avoid problems such as noise enhancement, edge halo, and jagged edges caused by simply increasing the sharpening intensity, thereby improving the accuracy of subsequent ISP sharpening and noise reduction parameter co-optimization.

[0041] In an optional embodiment, determining the image quality index corresponding to the entity region based on the semantic region mask includes: determining an image quality gain index corresponding to the entity region based on the semantic region mask; the image quality gain index characterizes the degree of image quality improvement of the entity region after image processing; determining a side effect penalty index corresponding to the entity region based on the semantic region mask; the side effect penalty index characterizes the degree of image side effects generated by image enhancement processing in the entity region; and performing image quality calculation on the entity region based on the image quality gain index and the side effect penalty index to obtain the image quality index corresponding to the entity region.

[0042] Image quality gains metrics can include gains in sharpness, brightness, exposure, contrast, and dynamic range. Contrast gains represent the global and local contrast of entity regions, as well as the degree of tonal variation in subject regions. Dynamic range gains represent the effective dynamic range of entity regions and the degree of detail retention in bright and dark areas. Side effect penalties represent the degree of undesirable image changes generated during image enhancement processing, i.e., the visual quality loss introduced by the image signal processor when improving image sharpness. Side effect penalties can include metrics such as halo intensity, aliasing intensity, and noise enhancement. Furthermore, different semantic categories corresponding to different entity regions have different emphases on image quality gains and side effect penalties; therefore, differentiated image quality evaluation emphases can be set for different semantic categories. For example, for vehicle and pedestrian areas, the weights of image quality gain metrics such as sharpness gain, dynamic range gain, and local contrast can be increased, while the weights of the penalty metrics for edge halo and jaggedness can be increased; for road surface areas, the weights of the dynamic range gain metrics corresponding to lane lines, curbs, and obstacle edges can be increased, while constraints on noise and compression block magnification can be imposed; for sky areas, the weight of sharpness gain can be reduced, while the evaluation weights corresponding to brightness gain and exposure gain metrics can be increased to suppress chromatic noise and compression blocks; for vegetation areas, higher texture detail gains can be allowed, while penalties for moiré patterns, false color, and excessive graininess can be increased; for building areas, the weight of sharpness gain for the continuity of straight edge lines can be increased, while the penalty weights for jaggedness and edge halo can be increased.

[0043] Specifically, firstly, the pixel range corresponding to each entity region in the target image is determined based on the semantic region mask, and image features within each entity region are extracted. Since the image content features corresponding to different entity regions are different, calculating the image quality index corresponding to each region based on the semantic region mask can avoid mutual interference between image features of different semantic regions. Next, the image quality gain index is determined based on the pixel information corresponding to the entity region. For example, the contrast gain index can be determined based on the pixel brightness distribution and local brightness changes of the entity region, and the dynamic range gain index can be determined based on the pixel distribution of bright and dark areas and the effective dynamic range of the entity region. Among them, the contrast gain index can be determined based on the global contrast, local contrast, and the degree of hierarchical change of the subject region; the dynamic range gain index can be determined based on the difference between preset high and low quantile brightness values ​​to determine the effective dynamic range, and the degree of dynamic range improvement can be determined in combination with the preservation of details in bright and dark areas. The contrast gain index and the dynamic range gain index correspond to different image quality evaluation dimensions, and can be calculated separately from the sharpness gain index and other gain indices such as exposure and brightness. Further, the image quality gain index corresponding to the entity region is determined based on the gain indices such as sharpness, exposure, brightness, contrast, and dynamic range. Then, the side effect penalty index corresponding to the entity region is determined based on the semantic region mask. Specifically, edge regions that produce significant changes after image enhancement can be detected within the entity region, and the brightness change features corresponding to the edge regions can be extracted. For example, brightness change curves on both sides of the edge are obtained along the edge gradient direction. Based on the peak and valley changes in the brightness curves and the background brightness difference on both sides of the edge, overshoot and undershoot phenomena generated during edge enhancement are determined, and then the edge halo intensity is calculated. Simultaneously, for oblique edges present within the entity region, the aliasing intensity can be determined based on the degree of step change in the edge position with the image scanning direction. Furthermore, the degree of noise enhancement can be determined based on brightness and chromaticity changes in flat regions, and the corresponding side effect penalty degree can be determined based on abnormal changes such as moiré patterns, false colors, and compression blocks in the image. Finally, the image quality index corresponding to the entity region is calculated based on the sharpness gain index and the side effect penalty index. In some embodiments, the image quality gain index can be used as a positive evaluation term, and the side effect penalty index as a negative constraint term, and the image quality index can be obtained through weighted fusion.

[0044] As can be seen, this embodiment breaks down the image quality evaluation process into two dimensions: image quality benefit evaluation index and side effect penalty evaluation index. This allows the image quality index to not only reflect the image detail enhancement effect, but also to constrain the undesirable visual phenomena generated during the image enhancement process. As a result, the obtained image quality index is more in line with the comprehensive requirements of drivers in vehicle image processing systems for image recognizability and display stability.

[0045] In an optional embodiment, determining the image quality gain index corresponding to the entity region based on the semantic region mask includes: extracting region image data corresponding to the entity region from the target image based on the semantic region mask; calculating the image quality gain parameter corresponding to the entity region based on the region image data; the image quality gain parameter includes a sharpness gain parameter, a brightness gain parameter, and an exposure gain parameter; generating a sharpness gain index corresponding to the entity region in the target image based on the sharpness gain parameter; and determining an exposure-brightness combined gain index corresponding to the entity region based on the brightness gain parameter and the exposure gain parameter; and determining the image quality gain index corresponding to the entity region based on the sharpness gain index and the exposure-brightness combined gain index.

[0046] The regional image data may include image data such as brightness distribution, brightness quantiles, edge features, local structure, and distribution of bright and dark areas within the entity region. The brightness gain parameter characterizes the improvement in the brightness distribution of the entity region relative to a preset brightness range; the exposure gain parameter characterizes the exposure state of the entity region, including improvements in overexposure, underexposure, shadow visibility, and highlight transition. It should be noted that increased brightness in the entity region does not necessarily indicate improved image quality. When increased brightness leads to increased noise in shadows, decreased local contrast, or loss of detail in highlights, the corresponding exposure gain may decrease. Therefore, a joint evaluation based on the brightness gain parameter and the exposure gain parameter can be used to obtain the comprehensive exposure-brightness gain index.

[0047] Specifically, firstly, region image data corresponding to entity regions is extracted from the target image based on a semantic region mask, so that subsequent calculations of image quality gain parameters are performed on the corresponding entity regions, rather than being uniformly calculated based on the entire target image. Then, brightness gain parameters and exposure gain parameters are determined based on the brightness information corresponding to the entity regions, and an exposure-brightness comprehensive gain index corresponding to the entity regions is determined based on these parameters. For example, the brightness distribution can be determined based on information such as the mean brightness, median brightness, brightness quantile, and logarithmic brightness of pixels within the entity regions, and the brightness distribution is compared with a preset target brightness range to obtain the brightness gain parameters. Simultaneously, exposure gain parameters can be determined based on information such as the shadow cutoff ratio, highlight cutoff ratio, near-black area ratio, near-white area ratio, shadow visibility, and highlight retention within the entity regions. Then, a sharpness gain index corresponding to the entity regions is generated based on the sharpness gain parameters; and an exposure-brightness comprehensive gain index corresponding to the entity regions is determined based on the brightness gain parameters and the exposure gain parameters. Specifically, the sharpness gain metric can be determined based on sharpness-related parameters such as effective detail, edge strength, and edge continuity to characterize the gains made by image processing on the detail representation and edge structure of the entity region. For the exposure-brightness comprehensive gain metric, image quality improvement is not determined solely by brightness parameters, but rather by a combination of brightness distribution and exposure status. For example, when the brightness of the entity region increases, it is also necessary to consider exposure-related parameters such as the proportion of near-black and near-white areas, overexposure or underexposure, and shadow visibility. When increased brightness improves shadow visibility without causing significant overexposure or highlight compression, it can be characterized as a positive gain in exposure-brightness. When increased brightness only results in an overall increase in brightness while causing a decrease in shadow detail, increased noise, or highlight compression, this brightness change is not directly considered an image quality gain. Furthermore, based on the sharpness gain metric and the exposure-brightness comprehensive gain metric, contrast gain metric and dynamic range gain metric can also be determined based on the region image data. Finally, the image quality gain index corresponding to the entity region is determined based on the sharpness gain index, exposure-brightness combined gain index and / or contrast, dynamic range, and other gain indexes. For example, the image quality gain index may include gain representations corresponding to multiple different image quality dimensions, and may further combine the contrast gain index and the dynamic range gain index to form a multi-dimensional image quality gain representation of the entity region, for subsequent image quality evaluation based on the weight parameters corresponding to the semantic category.

[0048] In an optional embodiment, generating a sharpness gain index corresponding to the entity region in the target image based on the sharpness gain parameter includes: determining the gradient magnitude corresponding to multiple pixels in the luminance channel of the target image; determining the local variance corresponding to the multiple pixels in the luminance channel of the target image; determining the temporal difference parameter corresponding to the multiple pixels between multiple frames in the target image; and determining the temporal stability parameter corresponding to the multiple pixels based on the temporal difference parameter; determining the target pixel as an effective detail pixel when the gradient magnitude of the target pixel is greater than a preset edge threshold, the local variance is greater than a preset variance threshold, and the temporal stability parameter is less than a preset flicker threshold; the target pixel is any one of the multiple pixels; determining the target effective detail intensity corresponding to the entity region based on the gradient magnitude corresponding to the effective detail pixel; and determining the target edge continuity corresponding to the entity region based on the distribution corresponding to the effective detail pixels; and determining the sharpness gain index based on the target effective detail intensity and the target edge continuity.

[0049] Among them, the target effective detail intensity describes the degree of enhancement of effective texture, contour, and edge details within the entity region; the target edge continuity describes the degree of continuity of the effective edge structure in the spatial distribution within the entity region. Effective detail pixels represent pixels within the entity region that can reflect the true image structure information. Gradient magnitude characterizes the degree of brightness change around a pixel; the more obvious the change, the more obvious the edge structure the region corresponding to that pixel may contain. The temporal stability parameter characterizes the stability of the image information corresponding to that pixel in the time dimension.

[0050] Specifically, firstly, the pixel range corresponding to the entity region to be evaluated in the target image is determined based on the semantic region mask, and the pixel values ​​of multiple pixels within the entity region in the luminance channel are obtained. Next, the gradient magnitude and local variance corresponding to multiple pixels are calculated. For any pixel p, the corresponding gradient magnitude can be calculated based on the luminance changes of that pixel in the horizontal and vertical directions. Simultaneously, the luminance variance within the neighborhood of the pixel is calculated to obtain the local variance parameter. The local variance describes the degree of texture change within the local region where the pixel is located. When there is stable texture or structural change in the local region, its corresponding local variance is usually high; while for flat areas such as the sky or noisy areas, its local variance may be low or unstable. Further, multiple consecutive frame images corresponding to the target image are obtained, and the temporal difference parameter is determined based on the luminance changes of pixels at the same position between multiple frame images. Then, the difference between the pixel values ​​corresponding to the same pixel position in adjacent frame images is calculated, and the corresponding temporal stability parameter is determined based on the trend of the difference changes across multiple consecutive frames. Further, for any target pixel in the entity region, it is determined whether the target pixel belongs to a valid detail pixel according to preset filtering conditions. When the gradient magnitude corresponding to a target pixel is greater than a preset edge threshold, it indicates that the pixel has significant structural changes; when the local variance is greater than a preset variance threshold, it indicates that the region where the pixel is located has effective texture information; when the time-domain stability parameter is less than a preset flicker threshold, it indicates that the information corresponding to the pixel remains stable in consecutive frames, then the target pixel is determined as an effective detail pixel. Subsequently, the target effective detail intensity corresponding to the entity region is determined based on the multiple effective detail pixels obtained through screening. In some embodiments, the gradient magnitudes corresponding to all effective detail pixels can be statistically analyzed, for example, by calculating the average, weighted sum, or normalized cumulative value of the gradient magnitudes of the effective detail pixels as the effective detail intensity corresponding to the entity region. Further, the target edge continuity is determined based on the spatial distribution of the effective detail pixels within the entity region. Finally, a sharpness gain index is determined based on the target effective detail intensity and the target edge continuity, thus comprehensively evaluating the degree of increase in effective detail and the ability to preserve edge structure after image enhancement. In one embodiment, the sharpness gain index can be calculated using a weighted fusion method: ,in, This represents the benefit metric for clarity. Indicates the strength of the effective details of the target. Indicates the continuity of the target edge. and These represent the weight coefficients of the corresponding evaluation parameters.

[0051] As can be seen, by combining spatial detail features and temporal stability features to filter effective detail pixels, the sharpness gain index can more accurately distinguish between real structural information and ineffective information such as noise and flicker. Compared with the method of evaluating image sharpness only using edge intensity, this embodiment can reduce the interference of noise enhancement and transient changes on the evaluation results, making the obtained sharpness gain index more consistent with the actual image quality changes in dynamic road environments in vehicular image processing systems.

[0052] In an optional embodiment, determining the side effect penalty index corresponding to the entity region based on the semantic region mask includes: determining strong edge pixels corresponding to the entity region in the target image; extracting brightness profile features along the gradient normal direction of the strong edge pixels; determining overshoot and undershoot values ​​corresponding to the strong edge pixels based on the brightness profile features; calculating the edge halo intensity corresponding to the entity region based on the overshoot value, the undershoot value, and the step amplitude value corresponding to the strong edge pixels; determining the oblique edges within the entity region; determining the aliasing intensity corresponding to the entity region based on the edge position of the oblique edges changing with the step of the scan line; and determining the side effect penalty index corresponding to the entity region based on the edge halo intensity and the aliasing intensity.

[0053] Among them, edge halo intensity is used to describe the brightness overshoot or undershoot phenomenon caused by excessive edge enhancement during image sharpening; jaggedness intensity is used to describe the degree of step-like edge distortion that appears after the oblique edge is processed by image enhancement.

[0054] Strong edge pixels represent edge locations within a solid area that exhibit significant brightness variations. They are determined based on whether the gradient magnitude of the corresponding pixel exceeds a preset edge threshold. Brightness profile features characterize continuous pixel brightness variation information acquired along the gradient direction of strong edge pixels. This is used to analyze brightness variation trends on both sides of the edge and to identify abnormal brightness fluctuations generated during edge enhancement.

[0055] Specifically, firstly, the pixel range corresponding to the target entity region is determined based on the semantic region mask, and edge feature analysis is performed on each pixel within the entity region. In some embodiments, the gradient magnitude of each pixel in the luminance channel can be calculated, and pixels with gradient magnitudes greater than a preset edge threshold are identified as strong edge pixels. Strong edge pixels typically correspond to locations with significant structural changes, such as vehicle outlines, road boundaries, and lane lines. Next, for each identified strong edge pixel, multiple pixels are sampled along the normal direction of the gradient direction corresponding to that pixel, and the luminance profile features corresponding to the strong edge pixel are extracted. Further, the overshoot and undershoot values ​​corresponding to the strong edge pixels are determined based on the luminance profile features. Firstly, the average background luminance of the regions far from the edge on both sides of the strong edge pixel is determined as the reference luminance for the regions before and after the edge. Subsequently, the overshoot value is determined based on the difference between the maximum luminance value in the luminance profile and the background luminance reference value, and the undershoot value is determined based on the difference between the minimum luminance value in the luminance profile and the background luminance reference value. For example, for any strong edge pixel, the average luminance of the background regions on both sides of the edge can be denoted as... and And determine the corresponding step amplitude. , This represents the step amplitude value corresponding to the edge. Further, it is based on the maximum brightness value in the brightness profile. and minimum brightness value Can be calculated ; ,in, Indicates overshoot value, This indicates an under-charge value. Further, based on the over-charge value... Underpayment and step amplitude The edge halo intensity corresponding to the entity region is calculated. In one implementation, the edge halo intensity can be determined by the ratio of overshoot and undershoot values ​​to the step amplitude. This allows for the quantification of the halo effect produced by sharpening processing on edge regions of different intensities. Then, the oblique edges within the entity region are identified, and the positional changes of these oblique edges along the image scanning direction are analyzed. Specifically, for an ideal continuous oblique edge, its position should change continuously with each scan line; however, after image enhancement or sampling processing, oblique edges may exhibit step-like jumps. Therefore, the amount of change in the oblique edge position between adjacent scan lines can be statistically analyzed, and the aliasing intensity can be determined based on this change. Finally, the processor determines the side effect penalty index corresponding to the entity region based on the edge halo intensity and aliasing intensity, for example, by using a weighted fusion method.

[0056] For easier understanding, please refer to Figure 4 , Figure 4This is a schematic flowchart illustrating a method for calculating an image side effect penalty index, as provided in an embodiment of this application. Figure 4 As shown, firstly, step S41 is executed to determine the target entity pixel region based on the semantic region mask. Specifically, the semantic region mask output by the semantic segmentation module is obtained, and the pixel range corresponding to the target entity is determined according to the mask. The target entity pixel region can be areas where image sharpness needs to be evaluated, such as vehicles, pedestrians, road edges, and buildings. Next, step S42 is executed to extract strong edge pixels within the target entity pixel region. Specifically, gradient calculation is performed on the target entity pixel region to obtain the gradient magnitude corresponding to each pixel within each region; when the gradient magnitude corresponding to a pixel is greater than a preset edge threshold, the pixel is determined as a strong edge pixel. Then, step S43 is executed to extract brightness profile features along the normal direction of the strong edge gradient, solving for the step amplitude A, overshoot O, and undershoot U. Specifically, multiple pixels are sampled along the normal direction of the corresponding gradient direction for the strong edge pixels, and the brightness information corresponding to each sampled point is obtained to form a brightness profile curve passing through the strong edge location. Since image sharpening may cause abnormal brightness enhancement or reduction on both sides of the edges, the corresponding step amplitude A, overshoot value O, and undershoot value U are determined based on the brightness changes in the brightness profile curve to characterize the degree of abnormal brightness changes caused by edge enhancement. Further, step S44 is executed to calculate the halo intensity of the entity region based on the step amplitude, overshoot value, and undershoot value. Specifically, the step amplitude A is determined based on the brightness change amplitude corresponding to the strong edge region, and the edge halo intensity is calculated by combining the overshoot value O and undershoot value U. The halo intensity is the ratio of the sum of the overshoot value O and the undershoot value U to the step amplitude A. The edge halo intensity indicates the degree to which bright or dark edges appear in the edge region after sharpening; the higher the edge halo intensity, the more pronounced the halo side effect of the current image enhancement process. Further, step S45 is executed to identify oblique edges in the region and statistically analyze the step change amount of the scan line edges to calculate the jaggedness intensity. Specifically, oblique edges are detected within the target entity region, and the edge position changes corresponding to different scan lines are obtained according to the image scanning direction. For an ideal, continuous oblique edge, its position should change smoothly with the scan line; however, after image sharpening or sampling, the oblique edge may exhibit a step-like abrupt change. Finally, step S46 is executed to weightedly fuse halo intensity and aliasing intensity, outputting a side effect penalty index for the entity region. Specifically, fusion weights are set according to the halo intensity and aliasing intensity corresponding to the target entity region, and the side effect penalty index is calculated using a weighted fusion method. For example, halo intensity and aliasing intensity can be weighted according to preset weights to obtain an evaluation result that comprehensively reflects the degree of image enhancement side effects on the target entity region.

[0057] It can be seen that by detecting the edge halo phenomenon and oblique edge jaggedness generated during the sharpening process, edge anomalies and structural distortions caused by excessive sharpening intensity can be identified. These negative effects are introduced into the image quality evaluation process, so that the subsequent ISP parameter optimization process can effectively suppress the side effects of image enhancement while improving the image detail performance, thereby improving the naturalness and visual stability of the output image of the vehicle image processing system.

[0058] In an optional embodiment, determining the overshoot and undershoot values ​​corresponding to the strong edge pixels based on the brightness profile features includes: determining a first average brightness value and a second average brightness value corresponding to positions far from the edge on both sides of the strong edge pixel; the first average brightness value represents the average brightness of multiple pixels in a first background region located on one side of the strong edge pixel and far from the strong edge pixel; the second average brightness value represents the average brightness of multiple pixels in a second background region located on the other side of the strong edge pixel and far from the strong edge pixel; determining the absolute value of the difference between the first average brightness value and the second average brightness value as the step amplitude value corresponding to the strong edge pixel; determining the difference between the maximum brightness value in the brightness profile features and the larger of the first average brightness value and the second average brightness value as the overshoot value; and determining the difference between the smaller of the first average brightness value and the second average brightness value and the minimum brightness value in the brightness profile as the undershoot value.

[0059] The first average brightness value corresponds to the brightness statistics of the first background region on one side of the strong edge pixel, and the second average brightness value corresponds to the brightness statistics of the second background region on the other side of the strong edge pixel. Both the first and second background regions are background regions distributed along the gradient normal direction corresponding to the strong edge pixel, with the strong edge pixel as a reference, and are located on either side of the strong edge pixel. The step amplitude represents the brightness change amplitude between the background regions on both sides of the strong edge pixel, reflecting the degree of brightness change corresponding to the edge itself. The overshoot value represents the degree to which the brightness peak near the edge exceeds the normal edge brightness change range after the image signal processor performs sharpening processing; the undershoot value represents the degree to which the brightness valley near the edge is below the normal brightness change range.

[0060] Specifically, firstly, based on the strong edge pixels and their corresponding brightness profile features, the background brightness information of the regions far from the edge on both sides of the strong edge pixels is determined. Since the region near the strong edge pixels may already be affected by sharpening enhancement, brightness statistics are performed on a first background region and a second background region at a certain distance from the strong edge pixels to avoid interference from the enhanced region in the baseline brightness calculation. Specifically, the average brightness of multiple pixels in the first background region is calculated to obtain the first average brightness value, and the average brightness of multiple pixels in the second background region is calculated to obtain the second average brightness value. The first and second average brightness values ​​represent the brightness levels of the real background regions on both sides of the strong edge pixels, respectively. Next, the step amplitude value corresponding to the strong edge pixels is determined based on the first and second average brightness values. This eliminates the influence of edge direction and brightness change direction on the evaluation results, allowing the step amplitude value to accurately reflect the degree of brightness change of the edge itself. Then, the overshoot value is determined based on the brightness profile features. Specifically, the maximum brightness value in the brightness profile extracted along the gradient normal direction is obtained, and the larger of the first and second brightness averages is determined as the brightness reference on both sides of the edge. The overshoot value is obtained by calculating the difference between the maximum brightness value and the larger value. The overshoot value can be defined as follows: ,in, for Overshoot value of the point; Represents the point relative to the strong edge in the brightness profile. Location; Indicates points along strong edges Brightness profile; This represents the average brightness of the area furthest from the edge on one side of the edge. This represents the average brightness of the region on the other side of the edge, away from the edge, namely the first average and the second average. Similar to determining the overshoot value, the undershoot value is determined based on the brightness profile characteristics. First, the minimum brightness value in the brightness profile is obtained, and the smaller of the first and second average brightness values ​​is determined as the background brightness reference for the dark side. The undershoot value is obtained by calculating the difference between this reference value and the minimum brightness value. The undershoot value can be defined as follows: ,in, for The overshoot value at the point is determined. Finally, the determined overshoot and undershoot values ​​are used as the basic parameters for calculating the edge halo intensity, and normalized calculation is performed in conjunction with the aforementioned step amplitude values.

[0061] For easier understanding, please refer to Figure 5 , Figure 5 This is a schematic diagram illustrating how edge halo parameters are determined based on brightness profile features, as provided in an embodiment of this application. Figure 5As shown, firstly, based on the strong edge pixels determined within the target entity region, pixel sampling is performed along the brightness profile direction corresponding to the strong edge pixels to obtain the brightness change curve passing through the strong edge pixels. The brightness profile direction can be the sampling direction along the image edge change direction. By selecting a certain range of pixels on both sides of the strong edge pixels, brightness change information of the edge region (including the strong edge pixels) during the change from the first background region (far from the edge) to the second background region (far from the edge) can be obtained. Further, based on the positional relationship of the strong edge pixels in the brightness profile direction, the first background region, the edge region, and the second background region are determined. The first and second background regions are located on both sides of the edge region and are used to represent stable brightness regions far from the edge; the edge region includes the strong edge pixels, representing the location where the image brightness changes significantly. Since sharpening processing may change the brightness distribution within a certain range near the edge, selecting the first and second background regions far from the edge as brightness reference regions can reduce the impact of enhancement processing on the background brightness statistics. Specifically, in the brightness change curve corresponding to brightness value and position, the horizontal axis represents the position (along the cross-sectional direction), and the vertical and horizontal axes represent the brightness values. First, the brightness values ​​corresponding to multiple pixels in the first background area are counted to obtain the first average brightness value. The brightness values ​​of multiple pixels within the second background region are counted to obtain the average second brightness value. Furthermore, based on the first average brightness value Second brightness average Determine the step amplitude value A corresponding to the edge region. The step amplitude value A represents the degree of change in background brightness between the two sides of the edge. In this embodiment, it can be based on the first average brightness value. Compared with the second average brightness The difference between them determines the step amplitude A, that is By using the step amplitude A, the influence of edges with different brightness directions on the evaluation results can be eliminated, allowing subsequent halo evaluation to be normalized based on the edge's own brightness variation. Furthermore, the overshoot value O corresponding to strong edge pixels is determined based on the brightness profile characteristics. In the brightness profile curve, sharpening enhancement may cause peak regions exceeding the target brightness variation range near the edges; therefore, the maximum brightness value in the brightness profile is obtained and combined with the first average brightness value. Second brightness average A luminance reference is determined on both sides of the edge. Based on the difference between the maximum luminance value and the corresponding luminance reference, an overshoot value O is obtained. The overshoot value O characterizes the degree of luminance surge caused by edge enhancement. Similar to determining the overshoot value O, an undershoot value U is further determined based on luminance profile characteristics. Specifically, the processor obtains the minimum luminance value in the luminance profile and, based on the first luminance average... Second brightness average A reference value for the dark background brightness is determined. The undershoot value U is obtained by calculating the difference between this reference value and the minimum brightness value. The undershoot value U characterizes the degree of brightness reduction after edge enhancement. Finally, the edge halo intensity is calculated based on the determined overshoot value O, undershoot value U, and step amplitude A. By normalizing the overshoot value O and undershoot value U with the step amplitude A, parameters for evaluating the degree of halo in the edge region can be obtained. When the overshoot value O and undershoot value U are large relative to the step amplitude A, it indicates that the edge halo phenomenon caused by the current sharpening process is more pronounced.

[0062] As can be seen, this embodiment establishes an adaptive brightness benchmark by utilizing the average background brightness on both sides of the edge, and calculates the overshoot and undershoot values ​​by combining the maximum and minimum values ​​in the brightness profile, thereby achieving a refined quantification of the abnormal brightness changes at the edges caused by the sharpening process.

[0063] In an optional embodiment, determining the image quality evaluation parameters corresponding to the entity region based on multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index includes: obtaining driving safety parameters characterizing the current driving state of the target vehicle; determining the position parameters of the entity region within the field of view of the on-board image processing system based on the semantic region mask; determining the semantic category confidence levels corresponding to the multiple semantic categories based on the semantic region mask; and determining the entity region area parameters corresponding to the entity region; determining multiple semantic category weight parameters corresponding to the entity region based on the semantic category confidence levels, the entity region area parameters, the position parameters, and the driving safety parameters; the multiple semantic category weights include vehicle weight parameters, pedestrian weight parameters, road surface weight parameters, sky weight parameters, vegetation weight parameters, and building weight parameters; and calculating the image quality index based on the multiple semantic category weight parameters to obtain the image quality evaluation parameters.

[0064] Among them, driving safety parameters characterize the safety requirements corresponding to the current driving state of the target vehicle, and can include parameters that reflect the degree of driving risk, such as vehicle speed, vehicle steering status, driver assistance status, and the proximity of surrounding targets. Positional parameters characterize the spatial relationship of entity regions within the display field of view of the in-vehicle image processing system, such as whether the entity region is located in the driver's primary observation area, the center of the field of view, or the edge area. Semantic category confidence characterizes the reliability of the semantic segmentation model's recognition results for different entity regions, and the entity region area parameter is used to characterize the proportion of the corresponding semantic region in the target image.

[0065] Specifically, firstly, driving safety parameters corresponding to the current driving state of the target vehicle are obtained, and the position parameters of different entity regions within the field of view of the onboard image processing system are determined based on the semantic region mask. Then, the semantic category confidence and entity region area parameters corresponding to each semantic category are determined based on the semantic region mask. The semantic category confidence can be determined by the output of the semantic segmentation model to reflect the reliability of the current entity region classification result. When the confidence corresponding to a certain semantic category is low, it indicates that there may be recognition uncertainty in that region, thus reducing the impact of that region on the final image quality evaluation result. The entity region area parameter reflects the proportion of different semantic regions in the current image. For example, when the road region occupies a large area, the importance of structural information such as road texture and lane lines for driving judgment increases accordingly. Further, the semantic category weight parameters corresponding to each entity region are determined based on the semantic category confidence, entity region area parameters, position parameters, and driving safety parameters. In one implementation, these parameters can be normalized, and the corresponding semantic category weights can be calculated using a weighted fusion method. For example, for vehicle areas, the weight of the vehicle area can be determined by considering the confidence level of the vehicle area, the area proportion of the vehicle area, the positional relationship of the vehicle area from the center of the field of view, and the current proximity status of the target. ,in, This represents the vehicle weight parameter. Indicates the semantic confidence level of the vehicle. This indicates the vehicle area parameter. Let S represent the vehicle position parameter and S represent the driving safety parameter. Similarly, road region weight parameters and sky region weight parameters can be calculated separately to form a dynamic set of weight parameters corresponding to multiple semantic categories. Finally, the image quality index of the corresponding entity region is calculated based on the multiple semantic category weight parameters to obtain the comprehensive image quality assessment parameters.

[0066] It can be seen that by introducing a dynamic weighting mechanism based on driving state, semantic recognition results, and regional spatial distribution, the image quality assessment process can adjust the importance of different entity regions according to changes in the current driving environment. Compared to methods using fixed region weights or uniform evaluation of the entire image, this embodiment can automatically increase the evaluation contribution of key regions when a vehicle approaches, driving risks change, or the area of ​​focus changes, making the image quality assessment results more aligned with driving safety requirements. Simultaneously, by combining semantic category confidence and region area parameters, the impact of semantic recognition errors and small, irrelevant regions on the evaluation results can be reduced, improving the stability and accuracy of image quality assessment parameters.

[0067] Step S304: Search and optimize the image quality optimization parameters in the image signal processor according to the image quality evaluation parameters to obtain the image quality adjustment parameters.

[0068] The image quality optimization parameters include at least sharpening parameters, noise reduction parameters, and exposure parameters. Sharpening parameters control the degree to which the image signal processor enhances high-frequency details such as edges and textures in the target image, while noise reduction parameters control the degree to which the image signal processor suppresses noise generated during image acquisition and enhancement. The search optimization process uses a search optimization algorithm to iteratively adjust the sharpening and noise reduction parameters based on the difference between the current image quality assessment parameters and the target image quality requirements, to determine the parameter combination that improves the image quality assessment results. The search optimization algorithm can include one or more of the following: coordinate descent, Bayesian optimization, CMA-ES, Nelder-Mead simplex search, multi-armed search, reinforcement learning strategies, lookup table plus local search, golden section search, and grid coarse search plus fine search.

[0069] Specifically, an initial parameter set corresponding to the image signal processor (ESP) is generated based on the current environmental conditions and written into the ESP, enabling the ESP to process subsequently acquired images based on the current candidate parameter set. Subsequently, a stable target image frame output by the ESP under the current candidate parameter set is acquired, and the image quality evaluation parameters corresponding to the target image frame are recalculated. Further, the parameter optimization direction is determined based on the image quality evaluation parameters corresponding to the current candidate parameter set, and the sharpening and noise reduction parameters in the candidate parameter set are adjusted. For example, if the sharpness gain parameter corresponding to the current candidate parameter set is low, and the side effect penalty parameter does not exceed a preset range, the sharpening parameter can be appropriately increased or the noise reduction parameter decreased to enhance effective detail information; if a high side effect penalty parameter is detected, such as obvious edge halo, jagged edges, or noise enhancement, the sharpening parameter can be decreased or the noise reduction parameter increased to suppress the negative impact of image enhancement. After completing one parameter adjustment, the updated candidate parameter set is rewritten into the ESP, and the processes of target image frame acquisition, image quality evaluation parameter calculation, and parameter optimization direction determination are repeated. Through multiple rounds of iterative search, the sharpening and noise reduction parameters of the image signal processor gradually approach the parameter combination that meets the target image quality requirements. After reaching the preset number of iterations or after the image quality evaluation parameters meet the target convergence condition, the image quality evaluation results corresponding to multiple iteration cycles can be smoothed, and the candidate parameter group corresponding to the highest image quality evaluation parameters during the optimization process is determined as the final image quality adjustment parameters.

[0070] It can be seen that by using image quality assessment parameters to perform closed-loop search optimization of image quality optimization parameters (such as sharpening and noise reduction parameters) in the image signal processor, the ISP parameter adjustment process is transformed from an offline configuration method that relies on human experience to an automatic optimization method based on actual image quality feedback. This can improve the recognizability of details in key areas such as vehicles and roads while avoiding image distortion caused by over-sharpening or noise reduction, thereby improving the image adaptability of the vehicle image processing system under different road environments and lighting conditions.

[0071] In an optional embodiment, the step of searching and optimizing the image quality optimization parameters in the image signal processor according to the image quality evaluation parameters to obtain image quality adjustment parameters includes: generating a first parameter group corresponding to the image signal processor according to the current environmental conditions; writing the first parameter group into the image signal processor, and obtaining a target image frame in a stable state after the image signal processor processes the image according to the first parameter group; determining the target image quality evaluation parameters corresponding to the target image frame in the image quality evaluation parameters; the target image quality evaluation parameters include target image quality gain parameters and target side effect penalty parameters; determining the optimization direction corresponding to the first parameter group according to the target image quality gain parameters and the target side effect penalty parameters; searching and optimizing the first parameter group according to the optimization direction to obtain an optimized second parameter group; writing the second parameter group into the image signal processor, and when a preset search convergence condition is met, smoothing the image quality evaluation parameters corresponding to multiple iteration cycles; determining a target parameter group from the candidate parameter groups corresponding to the multiple iteration cycles according to the smoothed image quality evaluation parameters, and determining the target parameter group as the image quality adjustment parameters.

[0072] The first parameter group, the second parameter group, and the candidate parameter group all represent a set of image processing parameters for the image signal processor within a certain optimization cycle. These parameter groups include at least sharpening parameters, noise reduction parameters, contrast, brightness, exposure parameters, and dynamic range evaluation metrics. The target image quality assessment parameters characterize the image processing effect corresponding to the current candidate parameter group. The search convergence condition characterizes the stabilization of the target image quality assessment parameters across multiple consecutive optimization cycles, or the stopping condition corresponding to a preset iteration limit, a high-risk vehicle condition, or a rapidly changing scene.

[0073] Specifically, firstly, the first parameter set corresponding to the image signal processor is determined based on the current environmental conditions. These current environmental conditions can be determined based on environmental information corresponding to the target image, such as current lighting conditions, road scene type, and image noise level. An initial parameter set matching the current environment can be selected from a preset parameter set as the starting parameters for subsequent search and optimization. The first parameter set may include initial sharpening parameters, initial noise reduction parameters, initial brightness, initial exposure parameters, and initial dynamic range parameters, used to provide the initial search position during the search and optimization process. Next, the first parameter set is written to the image signal processor, enabling it to process subsequently acquired images based on this set. Since the image signal processor may contain cached, filtered, or temporal noise reduction history information, after the parameters are written, the processor does not immediately acquire the output image. Instead, it waits for the image processing state to stabilize before acquiring the corresponding stable target image frame to avoid the transitional state during parameter switching affecting the image quality evaluation results. Then, quality analysis is performed on the target image frame to determine the target image quality evaluation parameters corresponding to the current first parameter set. Furthermore, when optimizing based on the target image quality gain parameters, sharpening and noise reduction parameters (or other parameters, such as exposure, local contrast, global contrast, etc.) can be adjusted, and the corresponding optimization result can be determined based on changes in effective details, edge structure, and image side effects. When optimizing image brightness and contrast, brightness, contrast, or exposure parameters can be adjusted, and the corresponding optimization result can be determined based on the image's brightness distribution, brightness and darkness levels, and the truncation of high and low brightness areas. When optimizing dynamic range, image processing parameters related to dynamic range can be adjusted, and the corresponding optimization result can be determined based on the image's effective brightness range and the preservation of details in bright and dark areas. Subsequently, the sharpening and noise reduction parameters in the first parameter group are adjusted according to the determined optimization direction to obtain the updated second parameter group. For example, the sharpening gain parameters, edge enhancement parameters, spatial noise reduction parameters, or temporal noise reduction parameters can be updated according to a preset adjustment step size, making the second parameter group closer to the target image quality requirements than the first parameter group. Next, the second parameter set is rewritten into the image signal processor, and the stable target image frame processed based on the second parameter set is reacquired. The corresponding target image quality assessment parameters are then recalculated. By repeating the above process, the parameter search process gradually converges. In an optional implementation, the image quality assessment parameters in each iteration can be used as optimization feedback values, and the correspondence between the corresponding candidate parameter set and the evaluation result can be recorded. When the number of iterations reaches the preset maximum number of iterations, the vehicle enters a high-risk state, or the scene changes rapidly, the image quality assessment parameters corresponding to multiple iteration cycles are smoothed to reduce the impact of single-frame image fluctuations, random noise, and other factors on the optimization results.Subsequently, the parameter group with the highest corresponding image quality evaluation parameter is selected from multiple candidate parameter groups as the final image quality adjustment parameter. In this way, by adopting stable image frame evaluation, multi-round iterative search, and evaluation result smoothing, the instantaneous fluctuations in the parameter adjustment process can be reduced, the stability of the final image quality adjustment parameter can be improved, and the vehicle image processing system can obtain a more suitable ISP processing effect in complex road environments.

[0074] In an optional embodiment, the step of searching and optimizing the first parameter group according to the optimization direction to obtain an optimized second parameter group includes: determining the target image quality optimization parameter corresponding to the optimization direction in the first parameter group; determining the parameter adjustment method corresponding to the target image quality optimization parameter according to the optimization direction, and determining the parameter adjustment amount corresponding to the target image quality optimization parameter according to a preset search step size; adjusting the target image quality optimization parameter in the first parameter group according to the parameter adjustment method and the parameter adjustment amount to obtain a candidate parameter group; when each image quality optimization parameter in the candidate parameter group is within the corresponding safe parameter boundary, and the single parameter change amount corresponding to the target image quality optimization parameter does not exceed a preset change limit, the candidate parameter group is determined as the optimized second parameter group.

[0075] The image quality optimization parameters may include at least one of the following: sharpening parameters, noise reduction parameters, contrast parameters, brightness parameters, exposure parameters, and image processing parameters related to dynamic range adjustment. The parameter adjustment method indicates the direction and operation of the adjustment for the target image quality optimization parameters, such as increasing or decreasing sharpening intensity, increasing or decreasing noise reduction intensity, adjusting the sharpening gain of the effective frequency band, adjusting overshoot or undershoot suppression parameters, or adjusting brightness, contrast, and exposure-related parameters. Safety parameter boundaries limit the allowable value range of each image quality optimization parameter, and preset variation limits restrict the magnitude of change in the target image quality optimization parameters during a single optimization process, thereby avoiding excessive disturbance to image quality from a single parameter adjustment.

[0076] Specifically, based on the current image quality assessment parameters, the image quality target that needs improvement is determined, and the target image quality optimization parameters corresponding to this target are determined from the first parameter group. For example, when the sharpness gain of the target image is insufficient and the current side effect penalty parameter is low, the sharpening parameter can be used as the target image quality optimization parameter, and increasing the sharpening intensity of the effective frequency band is determined as the parameter adjustment method. Next, the parameter adjustment amount corresponding to the target image quality optimization parameter is determined according to the preset search step size, and the parameter is adjusted once according to the determined parameter adjustment method to obtain the candidate parameter group. The preset search step size is set according to the specific situation. When using the coordinate descent search strategy, each round of search can select one target image quality optimization parameter from the first parameter group for adjustment, while other image quality optimization parameters remain unchanged. Thus, the target image is re-acquired based on the adjusted candidate parameter group, and the corresponding image quality assessment parameters are calculated. By comparing the sharpness gain parameter and the side effect penalty parameter before and after adjustment, it can be determined whether the current parameter adjustment can improve the image quality, and the subsequent search direction and parameter adjustment amount can be determined accordingly. For example, when the side effect penalty parameter corresponding to the candidate parameter group exceeds the preset side effect threshold, it indicates that the current image enhancement level is too high, which may produce obvious side effects such as edge halo, jagged edges, or noise enhancement. In this case, reducing the sharpening parameter, increasing the noise reduction parameter, or increasing the corresponding side effect suppression parameter can be determined as the subsequent parameter adjustment method to reduce the side effects produced by image enhancement. When the sharpness gain parameter corresponding to the candidate parameter group is lower than the preset sharpness threshold, and the side effect penalty parameter does not exceed the preset side effect threshold, it indicates that the current image still has room for further enhancement of effective details. In this case, the sharpening parameter corresponding to the effective frequency band can be increased, or the excessively strong noise reduction parameter can be reduced to increase the intensity of effective details and improve edge continuity. Furthermore, after determining the target image quality optimization parameters and their adjustment amounts, a safety check is performed on the adjusted candidate parameter group. Specifically, it can be determined whether each image quality optimization parameter in the candidate parameter group is within the corresponding safe parameter boundary, and whether the single parameter change amount of the target image quality optimization parameter exceeds the preset change limit. When a candidate parameter set meets the aforementioned safety conditions, it can be used as the second parameter set obtained from the current search, and subsequent image processing and image quality evaluation can continue based on the second parameter set. When a candidate parameter set does not meet the safety conditions, the parameter adjustment amount can be limited according to the safety parameter boundary or preset change limit, so that the adjusted parameters meet the corresponding safety conditions again, and a new candidate parameter set that meets the requirements can be formed. Finally, after completing one search of the current target image quality optimization parameters, the next image quality optimization parameters to be optimized can be determined based on the updated image quality evaluation parameters, and the coordinate descent search strategy can be used to optimize each parameter one by one until the preset search convergence condition is met.In this way, image quality optimization parameters can be gradually adjusted based on the relationship between sharpness gains and side effect penalties, while controlling the magnitude of individual parameter changes, without having to make large adjustments to multiple parameters at once. This avoids abrupt changes in image quality caused by excessive parameter adjustments, suppresses image enhancement side effects while improving effective details and edge structure, and improves the stability of the automatic image quality parameter optimization process.

[0077] For easier understanding, please refer to Figure 6 , Figure 6 This is a flowchart illustrating an iterative adjustment process for ISP parameters based on feedback optimization, provided in an embodiment of this application. Figure 6As shown, firstly, the current ISP parameter set is obtained, and the in-vehicle image is input. Then, image processing is performed on the in-vehicle image according to the current ISP parameter set to obtain the in-vehicle image corresponding to the current parameter conditions. The current ISP parameter set may include image signal processing parameters such as sharpening parameters and noise reduction parameters used to control image enhancement effects. Further, semantic segmentation and region masking are performed on the in-vehicle image, and a hardware statistical unit statistically analyzes image quality indicators and side effect indicators. Specifically, on one hand, different semantic category regions in the in-vehicle image are determined through semantic segmentation and region masking, and the corresponding evaluation range is determined based on the location of different semantic regions; on the other hand, the hardware statistical unit statistically analyzes image quality indicators and side effect indicators to obtain the sharpness evaluation indicator and image side effect evaluation indicator corresponding to the current in-vehicle image. Further, indicator weights are assigned according to semantic categories. Specifically, since different semantic category regions have different focuses on image quality—for example, regions such as vehicles, pedestrians, and road edges focus more on detail enhancement, while flat regions such as the sky focus more on noise suppression—different weights can be assigned to different regions according to the importance of the semantic category, so that the subsequent evaluation results can more accurately reflect the actual display effect of the current in-vehicle image. Further, a comprehensive evaluation value, Score, is determined by fusing the sharpness evaluation results and side effect evaluation results corresponding to each semantic region to obtain the comprehensive evaluation value Score corresponding to the current ISP parameter set. The comprehensive evaluation value Score is used to determine whether the current ISP parameters meet the preset optimization target. Then, it is determined whether the comprehensive evaluation value Score meets the target interval. When the determination result is yes, it means that the current ISP parameter set can meet the preset image quality requirements. At this time, the parameters are maintained and the optimal record is updated, that is, the current ISP parameter set is taken as the optimal parameters for the current stage. Further, scene / illuminance / semantic distribution change monitoring is performed to detect changes in the current environmental conditions. When an environmental change is detected, the next round of parameter search optimization is triggered to iteratively adjust the ISP parameters. When the determination result is no, it means that the comprehensive evaluation value Score corresponding to the current ISP parameter set has not reached the target interval. At this time, the next set of candidate parameters is searched. That is, the next set of candidate ISP parameters is determined based on the current ISP parameter set and the search strategy. In the process of generating candidate parameters, safety boundaries and single-step change limits are combined to constrain the parameter adjustment range and avoid excessive changes in ISP parameters that lead to unstable image quality. Finally, after determining the next set of candidate parameters, the candidate parameters are used as the new current ISP parameter set, and the vehicle-mounted image is re-inputted for image quality evaluation, thus forming a cyclical process of parameter search, image evaluation, and feedback adjustment.

[0078] As can be seen, the embodiments of this application can dynamically optimize ISP parameters based on semantic region information, hardware sharpness indicators, and side effect statistics of vehicle images. When the current ISP parameters meet the target range, the optimal parameters can be maintained and recorded; when they do not meet the target range, the next set of candidate parameters can be searched and optimized by combining safety boundaries and single-step change constraints. This enables adaptive image quality adjustment for different scenes, illumination, and semantic distribution changes, improving the image sharpness and stability of the vehicle image processing system in complex environments.

[0079] For easier understanding, please refer to Figure 7 , Figure 7This is a flowchart of an automatic adjustment process for image quality parameters provided in this application embodiment. First, after the system begins parameter tuning, it reads a basic parameter table, which includes illuminance, gain, exposure, temperature, and scene category. By obtaining the basic parameters corresponding to the current vehicle operating environment, an initial adjustment basis can be provided for subsequent ISP parameter search. Further, a search space and safety boundary are set according to the environmental parameters in the basic parameter table, including sharpening, noise reduction, halo suppression, and edge protection. The search space is used to limit the adjustable range of ISP parameters, and the safety boundary is used to limit the effective range during ISP parameter adjustment to avoid parameter adjustments exceeding the image processing capability or causing abnormal image quality. Next, a candidate parameter group is generated based on the set search space and safety boundary, and the candidate parameter group is written to the image signal processor and waits for N frames to stabilize, allowing the ISP to perform image processing based on the current candidate parameters. Subsequently, the output image of the image signal processor in the stabilized frame is obtained as the image quality evaluation object corresponding to the current candidate parameters. Further, low-resolution semantic segmentation is performed on the ISP output image to obtain a semantic region mask. Specifically, different target regions in the image are determined through semantic segmentation processing, and the target regions are located based on the semantic region mask. Subsequently, image features corresponding to different semantic regions are statistically analyzed according to semantic region mapping statistical blocks. Simultaneously, hardware operators are used to statistically analyze image metrics on the ISP output image to obtain relevant metrics for evaluating the image enhancement effect. Specifically, on one hand, a sharpness gain metric is calculated based on the statistical results corresponding to the semantic regions, including metrics such as effective detail, edge strength, and continuity, to reflect the effect of the current ISP parameters on image detail enhancement. On the other hand, a side effect penalty metric is calculated, including metrics such as noise, halo, jagged edges, false color, moiré patterns, zipper patterns, and compression blocks, to reflect the potential image quality loss caused by the current ISP parameters. Further, a semantically weighted comprehensive evaluation value is calculated based on the sharpness gain metric and the side effect penalty metric. Specifically, the semantically weighted comprehensive evaluation value is calculated based on the importance of different semantic regions and their corresponding metric weights, considering the impact of sharpness gain and side effects. Then, it is determined whether the side effects exceed a threshold. When the judgment result is yes, it means that the side effects such as noise, halo, and aliasing caused by the current ISP parameters exceed the preset limit. At this time, the parameters are tightened, including reducing sharpness or increasing noise reduction / suppression terms, to reduce the abnormal effects produced by image enhancement. When the judgment result is no, it is further judged whether the sharpness is insufficient. When the judgment result is yes, that is, the sharpness is insufficient, it means that the image detail enhancement effect corresponding to the current ISP parameters has not met the requirements. At this time, the enhancement parameters are executed, including increasing the effective frequency band sharpness or reducing excessive noise reduction, to improve the image texture and edge performance.If the judgment result is negative, the next step is to determine whether the comprehensive evaluation value has converged. If the judgment result is negative, i.e., the score has not converged, the next candidate parameter is selected through the search algorithm. If the judgment result is positive, i.e., the comprehensive evaluation value has converged, the current optimal parameter is saved as the final adjusted ISP parameter. Furthermore, during parameter optimization, it is necessary to determine whether a stopping condition has been triggered. Specifically, it is determined whether the iteration stopping condition is met, such as reaching the maximum number of iterations, meeting the convergence condition, or the parameter adjustment reaching a safe and stable state. If the stopping condition is met, i.e., the judgment result is positive, the process either reverts to or maintains the current optimal parameter and ends the parameter tuning process. If the stopping condition is not met, i.e., the judgment result is negative, single-step limiting, hysteresis, and parameter boundary checks are performed, and the "write to image signal processor and wait for N frames to stabilize" step is started again to perform the next round of parameter search and optimization.

[0080] Step S305: Write the image quality adjustment parameters into the image signal processor to adjust the image signal processor's processing of subsequently acquired images.

[0081] The image quality adjustment parameters are a combination of target parameters determined by search and optimization based on the image quality evaluation parameters, and are used to update and control the image processing process in the image signal processor.

[0082] Specifically, after determining the image quality adjustment parameters, these parameters are sent to the corresponding parameter configuration interface of the image signal processor to update the internal parameter configuration used to control image enhancement processing. When the vehicle continues to travel and acquires new road environment images, the image signal processor processes the newly acquired images based on the updated image quality adjustment parameters to output a target image that meets the current environmental quality requirements. Furthermore, before writing the image quality adjustment parameters into the image signal processor, the change in the image quality adjustment parameters relative to historical image quality parameters can be assessed. When the parameter change is detected to exceed the range, the change can be limited to restrict the magnitude of a single parameter adjustment, preventing problems such as noticeable flickering, brightness jumps, or visual discontinuities in the output image due to sudden changes in sharpening or noise reduction parameters. Simultaneously, during vehicle travel, when the vehicle status, target proximity status, or system diagnostic status meets preset safety lock conditions, the parameter update operation can be paused, and the current parameter state of the image signal processor can be maintained or the process can revert to a pre-verified parameter set to ensure image display stability during driving.

[0083] In an optional embodiment, after writing the image quality adjustment parameters to the image signal processor, the method further includes: determining the parameter change of the image quality adjustment parameters relative to the historical image quality parameters of the image signal processor; performing amplitude limiting processing on the parameter change, and updating the image quality adjustment parameters according to the limited parameter change to obtain target image quality adjustment parameters; if the vehicle driving state, target approach state, or system diagnostic state meets preset vehicle safety locking conditions, then the step of writing the image quality adjustment parameters to the image signal processor is stopped, and the current image quality parameters of the image signal processor are maintained, or the image signal processor is rolled back to a preset verified parameter set; if the vehicle safety locking conditions are not met, then the target image quality adjustment parameters are written to the image signal processor.

[0084] Among them, historical image quality parameters represent the image processing parameters currently being used by the image signal processor. These parameters may include current sharpening parameters, current noise reduction parameters, and other ISP configuration parameters. Parameter change represents the degree of difference between the image quality adjustment parameters obtained after search optimization and the current historical image quality parameters. Amplitude limiting is used to limit the maximum range of change during a single image parameter adjustment process. That is, when the parameter change exceeds a threshold, the portion exceeding the threshold is constrained, ensuring that the updated parameters change smoothly within a preset range. Vehicle safety lock conditions characterize situations where the current driving scenario or system operating state is unsuitable for dynamic adjustment of image parameters. Vehicle safety lock conditions can be determined based on one or more of the following states: vehicle driving state, target approach state, and system diagnostic state.

[0085] Specifically, firstly, after determining the image quality adjustment parameters based on the image quality assessment parameters, the historical image quality parameters currently used by the image signal processor are obtained, and the parameter change between the image quality adjustment parameters and the historical image quality parameters is calculated. For sharpening and noise reduction parameters, the difference in their corresponding parameter dimensions can be calculated separately. Next, it is determined whether the parameter change exceeds a preset adjustment range. When the parameter change is less than or equal to a preset threshold, the current change can be used directly for parameter updates; when the parameter change is greater than the preset threshold, a limiting process is performed on the parameter change. In an optional implementation, the parameter change can be limited in the following way: ,in, This indicates the change in parameters after limiting. This indicates the change in parameters before limiting. This represents the preset maximum allowable change. This ensures that the change amplitude during a single parameter update remains within a preset range. Then, the original image quality adjustment parameters are updated based on the limited parameter change to obtain the target image quality adjustment parameters. Further, before writing the target image quality adjustment parameters into the image signal processor, the current vehicle operating status, target proximity status, and system diagnostic status are acquired, and it is determined whether these statuses meet preset vehicle safety locking conditions. Specifically, when the vehicle is detected to be in a driving state requiring stable image display, such as high-speed driving, complex road environments, or a situation where a close-range target is rapidly approaching, the vehicle safety locking conditions are determined to be met. The image quality adjustment parameter writing operation is then paused, and the image quality parameters currently used by the image signal processor are maintained. Alternatively, if there is a potential risk of abnormality in the current image quality parameters, the image signal processor is reverted to a pre-verified parameter set to ensure the stability of the image output. When no vehicle safety locking conditions are detected, the parameter update operation is determined to be permitted. The target image quality adjustment parameters are then written into the image signal processor, allowing the image signal processor to process subsequent acquired images based on the updated sharpening and noise reduction parameters. In this way, by dynamically limiting the timing of parameter updates based on vehicle and system status, uncertain image parameter adjustments can be avoided in high-risk driving scenarios or under abnormal system conditions, thereby improving the reliability and safety of the vehicle image processing system in complex driving environments.

[0086] In some embodiments, the image quality adjustment method provided in this application can also be used in scenarios and fields such as surround view systems (AVM), forward-facing ADAS cameras, reversing cameras, and dashcams. For a clearer understanding of the image quality adjustment method, please continue reading. Figure 8 , Figure 8This is a block diagram of an image quality parameter adjustment method provided in this application embodiment. As can be seen, firstly, the CMS camera module acquires images of the vehicle's surrounding environment and sends the acquired image data to the ISP processing pipeline. The ISP processing pipeline performs image signal processing on the image data, generating an ISP output image (YUV / RGB), and sends the YUV / RGB image to the CMS display screen for the driver to view information about the vehicle's surrounding environment. Simultaneously, the YUV / RGB image is also sent to the subsequent image quality analysis process for image quality evaluation. Further, the YUV / RGB image is downsampled, frame-sliced, and cropped to reduce the data processing volume in the subsequent semantic analysis process. The preprocessed image is input to the edge-side NPU semantic segmentation module, which performs semantic recognition on the target objects in the image. Specifically, the edge-side NPU semantic segmentation module performs semantic segmentation processing on the input image to obtain corresponding semantic region masks. These semantic region masks include masks for the sky, road surface, vehicles, plants, buildings, and pedestrians. Subsequently, the semantic region mask is input to the mask mapping module, which maps it to full resolution or a statistical grid to determine the image evaluation region corresponding to different semantic regions. Further, the hardware indicator statistics module performs statistical analysis on image quality-related indicators based on the YUV / RGB of the ISP output image and the semantic region information. Specifically, the hardware indicator statistics module calculates image quality gain indicators and side effect indicators. Image quality gain indicators include sharpness, brightness, exposure, contrast, and dynamic range parameters, used to evaluate the effect of the current ISP parameters on image detail enhancement; side effect indicators include noise, halo, jagged edges, false color, moiré patterns, zipper patterns, and compression blocks, used to evaluate image anomalies generated during the current ISP parameter processing. Further, based on the importance of different semantic regions and the weight of the corresponding evaluation indicators, the sharpness and side effect indicators corresponding to different regions are weighted and fused to obtain the target evaluation result used to evaluate the effect of the current ISP parameters. The parameter search and optimization module optimizes ISP parameters based on the semantically weighted objective function and environmental parameters such as vehicle status, illumination, gain, exposure, and temperature. It then generates a new set of candidate ISP parameters based on the current image quality evaluation results and iteratively adjusts the ISP processing parameters to ensure the comprehensive evaluation results meet the preset optimization objectives. To ensure the stability of the ISP parameter adjustment process, the parameter search and optimization module interacts with a safety protection module. The safety protection module performs protection processing including amplitude limiting, hysteresis, stable frame confirmation, and rollback. Amplitude limiting restricts the range of ISP parameter adjustments to prevent excessive parameter changes; hysteresis restores historically valid parameters when parameter adjustments cause image quality degradation; stable frame confirmation determines whether the current parameters are stable; and rollback ensures reliable execution of the parameter optimization process.Finally, the optimized ISP parameters are fed back to the ISP processing pipeline through the ISP parameter control interface, enabling the ISP processing pipeline to reprocess the input image according to the new ISP parameters and continuously perform image evaluation and parameter optimization processes, thereby forming a closed-loop adjustment.

[0087] As can be seen, this embodiment can leverage the synergistic effects of the CMS camera module, ISP processing pipeline, edge-side NPU semantic segmentation module, hardware indicator statistics module, and parameter search and optimization module to achieve image quality evaluation and automatic ISP parameter optimization based on semantic region awareness. This architecture can dynamically adjust ISP parameters according to vehicle status, illumination, gain, exposure, temperature, and scene changes. While improving effective details, edge strength, and edge continuity, it reduces side effects such as noise, halo, jagged edges, false color, moiré patterns, zipper patterns, and compression blocks, thereby improving the image display quality and stability of the in-vehicle image processing system under different driving environments.

[0088] As can be seen, by implementing the image quality adjustment method provided in the above embodiments, this method distinguishes different entity regions in the target image through semantic region masks, and performs fusion evaluation of the image quality indicators of each entity region based on the weights corresponding to different semantic categories. This enables the image quality evaluation process to be associated with the visual needs of different regions, avoiding the problem of insufficient sharpness enhancement or over-processing in some regions due to a unified evaluation based on the entire frame image. At the same time, the image quality optimization parameters are searched and optimized according to the fused image quality evaluation parameters, so that the sharpening enhancement process and the noise reduction process can be coordinated and adjusted based on the image quality evaluation results, thereby improving the effective image detail performance while reducing the generation of image side effects. Furthermore, by writing the optimized image quality adjustment parameters into the image signal processor, the image signal processor can adjust the subsequent image processing process according to the optimization results, realizing the automatic adjustment of vehicle image quality parameters and improving the quality and stability of images around vehicles in complex road environments.

[0089] This application embodiment can divide the vehicle chip into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0090] When dividing each function into modules according to its corresponding function. Figure 9This is a functional module block diagram of an image quality parameter adjustment device provided in an embodiment of this application. The adjustment device is applied to the main processor of an in-vehicle image processing system. The image quality parameter adjustment device 900 includes: The image signal processor 910 is used to process the acquired image to obtain the target image; The neural network processor 920 is used to perform semantic segmentation on the target image to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions in the road environment of the target image; the entity regions include vehicles, pedestrians, road surface, sky, plants, and buildings; The indicator statistics module 930 is used to determine the image quality indicator corresponding to the entity region based on the semantic region mask. The main processor 940 is configured to determine the image quality evaluation parameters corresponding to the entity region based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index; search and optimize the image quality optimization parameters in the image signal processor based on the image quality evaluation parameters to obtain image quality adjustment parameters; and write the image quality adjustment parameters to the image signal processor to adjust the image signal processor's processing of subsequently acquired images.

[0091] In an optional embodiment, the index statistics module 930, in determining the image quality index corresponding to the entity region based on the semantic region mask, is specifically configured to: determine the image quality gain index corresponding to the entity region based on the semantic region mask; the image quality gain index characterizes the degree of image quality improvement of the entity region after image processing; determine the side effect penalty index corresponding to the entity region based on the semantic region mask; the side effect penalty index characterizes the degree of image side effects generated by image enhancement processing in the entity region; and perform image quality calculation on the entity region based on the image quality gain index and the side effect penalty index to obtain the image quality index corresponding to the entity region.

[0092] In an optional embodiment, the index statistics module 930, in determining the image quality gain index corresponding to the entity region based on the semantic region mask, is specifically configured to: extract region image data corresponding to the entity region from the target image based on the semantic region mask; calculate the image quality gain parameter corresponding to the entity region based on the region image data; the image quality gain parameter includes a sharpness gain parameter, a brightness gain parameter, and an exposure gain parameter; generate a sharpness gain index corresponding to the entity region in the target image based on the sharpness gain parameter; and determine an exposure-brightness comprehensive gain index corresponding to the entity region based on the brightness gain parameter and the exposure gain parameter; and determine the image quality gain index corresponding to the entity region based on the sharpness gain index and the exposure-brightness comprehensive gain index.

[0093] In an optional embodiment, the index statistics module 930, in generating the sharpness gain index corresponding to the entity region in the target image based on the sharpness gain parameter, is specifically configured to: determine the gradient magnitude corresponding to multiple pixels in the luminance channel of the target image; and determine the local variance corresponding to the multiple pixels in the luminance channel of the target image; determine the temporal difference parameter corresponding to the multiple pixels between multiple frames in the target image; and determine the temporal stability parameter corresponding to the multiple pixels based on the temporal difference parameter; when the gradient magnitude of the target pixel is greater than a preset edge threshold, the local variance is greater than a preset variance threshold, and the temporal stability parameter is less than a preset flicker threshold, the target pixel is determined as an effective detail pixel; the target pixel is any one of the multiple pixels; determine the target effective detail intensity corresponding to the entity region based on the gradient magnitude corresponding to the effective detail pixel; and determine the target edge continuity corresponding to the entity region based on the distribution corresponding to the effective detail pixels; and determine the sharpness gain index based on the target effective detail intensity and the target edge continuity.

[0094] In an optional embodiment, the index statistics module 930, in determining the side effect penalty index corresponding to the entity region based on the semantic region mask, is specifically configured to: determine strong edge pixels corresponding to the entity region in the target image; extract brightness profile features along the gradient normal direction of the strong edge pixels; determine overshoot and undershoot values ​​corresponding to the strong edge pixels based on the brightness profile features; calculate the edge halo intensity corresponding to the entity region based on the overshoot value, the undershoot value, and the step amplitude value corresponding to the strong edge pixels; determine the oblique edges within the entity region; determine the aliasing intensity corresponding to the entity region based on the edge position of the oblique edges changing with the step of the scan line; and determine the side effect penalty index corresponding to the entity region based on the edge halo intensity and the aliasing intensity.

[0095] In an optional embodiment, the index statistics module 930, in determining the overshoot and undershoot values ​​corresponding to the strong edge pixel based on the brightness profile features, is specifically configured to: determine a first average brightness value and a second average brightness value corresponding to the positions far from the edge on both sides of the strong edge pixel; the first average brightness value represents the average brightness of multiple pixels in a first background region located on one side of the strong edge pixel and far from the strong edge pixel; the second average brightness value represents the average brightness of multiple pixels in a second background region located on the other side of the strong edge pixel and far from the strong edge pixel; determine the absolute value of the difference between the first average brightness value and the second average brightness value as the step amplitude value corresponding to the strong edge pixel; determine the difference between the maximum brightness value in the brightness profile features and the larger value between the first average brightness value and the second average brightness value as the overshoot value; and determine the difference between the smaller value between the first average brightness value and the second average brightness value and the minimum brightness value in the brightness profile as the undershoot value.

[0096] In an optional embodiment, the main processor 940, in determining the image quality evaluation parameters corresponding to the entity region based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index, specifically performs the following: acquiring driving safety parameters characterizing the current driving state of the target vehicle; determining the position parameters of the entity region within the corresponding field of view of the in-vehicle image processing system based on the semantic region mask; determining the semantic category confidence levels corresponding to the multiple semantic categories based on the semantic region mask; and determining the entity region area parameters corresponding to the entity region; determining the multiple semantic category weight parameters corresponding to the entity region based on the semantic category confidence levels, the entity region area parameters, the position parameters, and the driving safety parameters; the multiple semantic category weights include vehicle weight parameters, pedestrian weight parameters, road surface weight parameters, sky weight parameters, plant weight parameters, and building weight parameters; and calculating the image quality index based on the multiple semantic category weight parameters to obtain the image quality evaluation parameters.

[0097] In an optional embodiment, the main processor 940, in the step of searching and optimizing the image quality optimization parameters in the image signal processor according to the image quality evaluation parameters to obtain image quality adjustment parameters, specifically performs the following steps: generating a first parameter set corresponding to the image signal processor based on the current environmental conditions; writing the first parameter set into the image signal processor, and after the image signal processor processes the image according to the first parameter set, obtaining a target image frame in a stable state; determining the target image quality evaluation parameters corresponding to the target image frame in the image quality evaluation parameters; the target image quality evaluation parameters include target image quality gain parameters and target side effect penalty parameters; determining the optimization direction corresponding to the first parameter set based on the target image quality gain parameters and the target side effect penalty parameters; searching and optimizing the first parameter set according to the optimization direction to obtain an optimized second parameter set; writing the second parameter set into the image signal processor, and iteratively executing the step of determining the target image quality evaluation parameters corresponding to the target image frame in the image quality evaluation parameters; When the preset search convergence condition is met, the image quality evaluation parameters corresponding to multiple iteration cycles are smoothed; based on the smoothed image quality evaluation parameters, a target parameter group is determined from the candidate parameter group corresponding to the multiple iteration cycles, and the target parameter group is determined as the image quality adjustment parameter.

[0098] In an optional embodiment, the main processor 940, in performing the search optimization on the first parameter group according to the optimization direction to obtain the optimized second parameter group, is specifically configured to: determine the target image quality optimization parameter corresponding to the optimization direction in the first parameter group; determine the parameter adjustment method corresponding to the target image quality optimization parameter according to the optimization direction, and determine the parameter adjustment amount corresponding to the target image quality optimization parameter according to a preset search step size; adjust the target image quality optimization parameter in the first parameter group according to the parameter adjustment method and the parameter adjustment amount to obtain a candidate parameter group; when each image quality optimization parameter in the candidate parameter group is within the corresponding safe parameter boundary, and the single parameter change amount corresponding to the target image quality optimization parameter does not exceed a preset change limit, determine the candidate parameter group as the optimized second parameter group.

[0099] In an optional embodiment, after writing the image quality adjustment parameters to the image signal processor, the main processor 940 is further configured to: determine the parameter change of the image quality adjustment parameters relative to the historical image quality parameters of the image signal processor; perform amplitude limiting processing on the parameter change, and update the image quality adjustment parameters according to the limited parameter change to obtain target image quality adjustment parameters; if the vehicle driving state, target approach state, or system diagnostic state meets preset vehicle safety locking conditions, then stop executing the step of writing the image quality adjustment parameters to the image signal processor, and maintain the current image quality parameters of the image signal processor, or roll back the image signal processor to a preset verified parameter set; if the vehicle safety locking conditions are not met, then write the target image quality adjustment parameters to the image signal processor.

[0100] As can be seen, the image quality parameter adjustment device provided in this application embodiment first acquires a target image processed by an image signal processor; secondly, it performs semantic segmentation on the target image to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions within the road environment in the target image; the entity regions include vehicle regions, road regions, and sky regions; the device determines the image quality index corresponding to the entity regions based on the semantic region mask; the device determines the image quality evaluation parameters corresponding to the entity regions based on the weights of multiple semantic categories corresponding to the semantic region mask and the image quality index; then, it searches and optimizes the image quality optimization parameters (such as sharpening and noise reduction parameters) in the image signal processor based on the image quality evaluation parameters to obtain image quality adjustment parameters; finally, the image quality adjustment parameters are written into the image signal processor to adjust the image signal processor's processing of subsequently acquired images. This improves the quality and stability of images around vehicles in complex road environments.

[0101] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the methods described in the above method embodiments, wherein the computer includes an in-vehicle chip SoC.

[0102] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. The computer program product can be a software installation package, and the computer includes an automotive-grade chip (SoC).

[0103] It should be noted that, for the sake of simplicity, the above embodiments are all described as a series of actions. Those skilled in the art should understand that this application is not limited to the described order of actions, as some steps in the embodiments of this application can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions, steps, modules, or units involved are not necessarily essential to the embodiments of this application. The descriptions of each embodiment in this application have different focuses; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments.

[0104] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above descriptions are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A method for adjusting image quality parameters, characterized in that, The method, applied to a main processor in an in-vehicle image processing system, includes: Acquire the target image after it has been processed by the image signal processor; The target image is semantically segmented to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions within the road environment in the target image; the entity regions include vehicles, pedestrians, road surface, sky, plants, and buildings; The image quality index corresponding to the entity region is determined based on the semantic region mask; the image quality evaluation parameters corresponding to the entity region are determined based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index. Based on the image quality assessment parameters, the image quality optimization parameters in the image signal processor are searched and optimized to obtain the image quality adjustment parameters; The image quality adjustment parameters are written to the image signal processor to adjust the image signal processor's processing of subsequently acquired images.

2. The method as described in claim 1, characterized in that, Determining the image quality index corresponding to the entity region based on the semantic region mask includes: The image quality gain index corresponding to the entity region is determined based on the semantic region mask; the image quality gain index characterizes the degree of image quality improvement of the entity region after image processing. The side effect penalty index corresponding to the entity region is determined based on the semantic region mask; the side effect penalty index characterizes the degree of image side effects generated by image enhancement processing within the entity region. The image quality of the entity region is calculated based on the image quality gain index and the side effect penalty index to obtain the image quality index corresponding to the entity region.

3. The method as described in claim 2, characterized in that, The step of determining the image quality gain index corresponding to the entity region based on the semantic region mask includes: Extract the region image data corresponding to the entity region from the target image based on the semantic region mask; The image quality gain parameters corresponding to the entity region are calculated based on the image data of the region; the image quality gain parameters include sharpness gain parameters, brightness gain parameters, and exposure gain parameters. Generate a sharpness gain index corresponding to the entity region in the target image based on the sharpness gain parameter; and determine an exposure-brightness combined gain index corresponding to the entity region based on the brightness gain parameter and the exposure gain parameter. The image quality gain index corresponding to the entity region is determined based on the sharpness gain index and the exposure-brightness combined gain index.

4. The method as described in claim 3, characterized in that, The step of generating a sharpness gain index corresponding to the entity region in the target image based on the sharpness gain parameter includes: Determine the gradient magnitude corresponding to multiple pixels in the brightness channel of the target image; and determine the local variance corresponding to the multiple pixels in the brightness channel of the target image. Determine the temporal difference parameters corresponding to the multiple pixels among multiple frames in the target image; and determine the temporal stability parameters corresponding to the multiple pixels based on the temporal difference parameters; The target pixel is determined as a valid detail pixel when the gradient magnitude of the target pixel is greater than a preset edge threshold, the local variance is greater than a preset variance threshold, and the temporal stability parameter is less than a preset flicker threshold; the target pixel is any one of the plurality of pixels. The target effective detail intensity corresponding to the entity region is determined based on the gradient magnitude corresponding to the effective detail pixels; and the target edge continuity corresponding to the entity region is determined based on the distribution corresponding to the effective detail pixels. The sharpness gain metric is determined based on the effective detail intensity of the target and the continuity of the target edges.

5. The method as described in claim 2, characterized in that, The step of determining the side effect penalty index corresponding to the entity region based on the semantic region mask includes: Identify the strong edge pixels corresponding to the entity region in the target image; and extract brightness profile features along the gradient normal direction of the strong edge pixels. The overshoot and undershoot values ​​corresponding to the strong edge pixels are determined based on the brightness profile features. The edge halo intensity corresponding to the entity region is calculated based on the overshoot value, the undershoot value, and the step amplitude value corresponding to the strong edge pixel. Determine the oblique edges within the solid region; and determine the serration intensity corresponding to the solid region based on the change in edge position of the oblique edges with the step change of the scan line; The side effect penalty index corresponding to the entity region is determined based on the edge halo intensity and the serration intensity.

6. The method as described in claim 5, characterized in that, The step of determining the overshoot and undershoot values ​​corresponding to the strong edge pixels based on the brightness profile features includes: Determine the first average brightness and the second average brightness corresponding to the positions far from the edge on both sides of the strong edge pixel; the first average brightness represents the average brightness of multiple pixels in a first background region located on one side of the strong edge pixel and far from the strong edge pixel; the second average brightness represents the average brightness of multiple pixels in a second background region located on the other side of the strong edge pixel and far from the strong edge pixel. The absolute value of the difference between the first average brightness value and the second average brightness value is determined as the step amplitude value corresponding to the strong edge pixel. The difference between the maximum brightness value in the brightness profile feature and the larger of the first average brightness value and the second average brightness value is determined as the overshoot value; The difference between the smaller of the first average brightness value and the second average brightness value and the minimum brightness value in the brightness profile is determined as the undershoot value.

7. The method according to any one of claims 1-6, characterized in that, The step of determining the image quality evaluation parameters corresponding to the entity region based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index includes: Obtain driving safety parameters that characterize the current driving state of the target vehicle; The position parameters of the entity region within the corresponding field of view of the in-vehicle image processing system are determined based on the semantic region mask. Based on the semantic region mask, determine the semantic category confidence level corresponding to the plurality of semantic categories; and determine the entity region area parameter corresponding to the entity region; The semantic category confidence score, the entity region area parameter, the location parameter, and the driving safety parameter are used to determine multiple semantic category weight parameters corresponding to the entity region; the multiple semantic category weights include vehicle weight parameters, pedestrian weight parameters, road surface weight parameters, sky weight parameters, plant weight parameters, and building weight parameters; The image quality index is calculated based on the multiple semantic category weight parameters to obtain the image quality evaluation parameters.

8. The method according to any one of claims 1-6, characterized in that, The step of searching and optimizing the image quality optimization parameters in the image signal processor based on the image quality assessment parameters to obtain image quality adjustment parameters includes: Generate the first parameter group corresponding to the image signal processor based on the current environmental conditions; The first parameter set is written into the image signal processor, and after the image signal processor processes the image according to the first parameter set, a target image frame in a stable state is obtained. Determine the target image quality assessment parameters corresponding to the target image frame in the image quality assessment parameters; the target image quality assessment parameters include target image quality gain parameters and target side effect penalty parameters; The optimization direction corresponding to the first parameter group is determined based on the target image quality gain parameter and the target side effect penalty parameter; The first parameter set is searched and optimized according to the optimization direction to obtain the optimized second parameter set; The second parameter group is written into the image signal processor, and the step of determining the target image quality assessment parameters corresponding to the target image frame in the image quality assessment parameters is executed iteratively. When the preset search convergence condition is met, the image quality evaluation parameters corresponding to multiple iteration cycles are smoothed; based on the smoothed image quality evaluation parameters, a target parameter group is determined from the candidate parameter group corresponding to the multiple iteration cycles, and the target parameter group is determined as the image quality adjustment parameter.

9. The method as described in claim 8, characterized in that, The step of searching and optimizing the first parameter set according to the optimization direction to obtain the optimized second parameter set includes: Determine the target image quality optimization parameters corresponding to the optimization direction in the first parameter group; The parameter adjustment method corresponding to the target image quality optimization parameter is determined according to the optimization direction, and the parameter adjustment amount corresponding to the target image quality optimization parameter is determined according to the preset search step size; The target image quality optimization parameters in the first parameter group are adjusted according to the parameter adjustment method and the parameter adjustment amount to obtain a candidate parameter group; When each image quality optimization parameter in the candidate parameter group is within its corresponding safety parameter boundary, and the single parameter change corresponding to the target image quality optimization parameter does not exceed a preset change limit, the candidate parameter group is determined as the optimized second parameter group.

10. The method according to any one of claims 1-6, characterized in that, After writing the image quality adjustment parameters to the image signal processor, the method further includes: Determine the amount of change in the image quality adjustment parameter relative to the historical image quality parameter in the image signal processor; The parameter change is limited, and the image quality adjustment parameters are updated based on the limited parameter change to obtain the target image quality adjustment parameters. If the vehicle driving status, target approach status, or system diagnostic status meets the preset vehicle safety lock conditions, then the step of writing the image quality adjustment parameters to the image signal processor is stopped, and the current image quality parameters of the image signal processor are maintained, or the image signal processor is rolled back to the preset verified parameter set. If the vehicle safety locking conditions are not met, the target image quality adjustment parameters are written into the image signal processor.

11. A device for adjusting image quality parameters, characterized in that, The device includes: An image signal processor is used to process acquired images to obtain the target image; A neural network processor is used to perform semantic segmentation on the target image to obtain a semantic region mask; the semantic region mask corresponds to the mask of entity regions within the road environment in the target image; the entity regions include vehicles, pedestrians, road surface, sky, plants, and buildings; The indicator statistics module is used to determine the image quality indicators corresponding to the entity region based on the semantic region mask. The main processor is configured to determine the image quality evaluation parameters corresponding to the entity region based on the multiple semantic category weight parameters corresponding to the semantic region mask and the image quality index; search and optimize the image quality optimization parameters in the image signal processor based on the image quality evaluation parameters to obtain image quality adjustment parameters; and write the image quality adjustment parameters to the image signal processor to adjust the image signal processor's processing of subsequently acquired images.

12. A vehicle-mounted chip SoC, characterized in that, include: The processor, memory, communication interface, and program stored in the memory; when the program is executed by the processor, the vehicle-mounted chip SoC performs the method as described in any one of claims 1-10.

13. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the running steps corresponding to the image quality parameter adjustment method as described in any one of claims 1-10.

Images