Intelligent exposure adjustment system and method based on content recognition
By using a content recognition-based intelligent exposure adjustment system that combines object detection and deep learning models to dynamically adjust exposure parameters, the system solves the problems of adaptability and user experience of traditional exposure algorithms under complex lighting conditions, achieving accurate exposure and high-quality image effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI JUNZHENG TECH CO LTD
- Filing Date
- 2025-01-08
- Publication Date
- 2026-07-10
AI Technical Summary
Traditional automatic exposure algorithms struggle to accurately reflect the actual brightness requirements of different areas of an image under complex lighting conditions, resulting in overexposure or underexposure, inability to distinguish key objects from the background, poor adaptability, and a poor user experience.
An intelligent exposure adjustment system based on content recognition is adopted. It identifies key objects and their locations in the image through object detection algorithms, and combines deep learning models such as YOLOv5 to dynamically adjust exposure parameters to optimize the exposure effect, including weighted average brightness calculation of the region of interest and real-time exposure control.
It enables precise exposure adjustment under complex lighting conditions, improves image quality and dynamic range, enhances detail retention, meets users' personalized needs, and improves user experience.
Smart Images
Figure CN122372845A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image processing technology, and specifically relates to an intelligent exposure adjustment system and method based on content recognition. Background Technology
[0002] In current digital photography and video shooting technologies, exposure is one of the key factors affecting image quality. Exposure refers to controlling parameters such as camera exposure time and ISO sensitivity to ensure that the image sensor receives appropriate light intensity, thereby obtaining a clear, bright, and color-accurate image. Traditional exposure adjustment methods mainly rely on the camera's global or local brightness automatic exposure algorithms. These algorithms typically adjust exposure settings based on preset parameters and simple brightness statistics. However, this exposure adjustment method based on global or local brightness is ineffective when dealing with scenes under complex lighting conditions, easily leading to overexposure or underexposure.
[0003] In recent years, with the development of computer vision technology, content recognition technology has been gradually applied to the field of image processing. Content recognition can analyze different regions in an image, identify key objects and scene features, and thus provide a basis for more precise exposure adjustment. Content recognition technology mainly includes object detection, semantic segmentation, and scene classification. Among them, object detection technology can identify specific objects in an image and locate their position and size. For example, algorithms such as YOLO (You Only Look Once) and SSD (Single Shot MultiBox Detector) can efficiently detect multiple objects in an image in a real-time scene. Semantic segmentation technology can divide an image into multiple semantic regions and assign corresponding labels to each region. For example, algorithms such as DeepLab and Mask R-CNN can classify each pixel in an image, thereby achieving fine semantic segmentation. Scene classification technology can identify the overall scene type of an image, such as indoor, outdoor, urban, rural, etc. For example, deep learning models such as GoogLeNet and ResNet can perform scene classification on large-scale datasets, thereby providing contextual information for exposure adjustment.
[0004] Content recognition technology can significantly improve the performance of traditional automatic exposure algorithms, enabling more precise exposure adjustments. Specific applications include:
[0005] Key object recognition: This feature uses object detection technology to identify key objects in an image, such as faces and vehicles, and then focuses the exposure on them. For example, when shooting portraits, it ensures that the face is properly exposed, avoiding overexposure or underexposure of the background.
[0006] Scene feature analysis: Scene classification technology is used to identify the overall scene type of the image and exposure parameters are adjusted according to scene characteristics. For example, when shooting at night, the ISO is increased to improve image brightness; when shooting indoors, the exposure time is adjusted to avoid blur.
[0007] Regional exposure optimization: Image is divided into multiple semantic regions using semantic segmentation technology, and exposure parameters are adjusted based on the characteristics of each region. For example, when shooting landscapes, this ensures that the sky and ground are properly exposed, avoiding overexposure or underexposure problems caused by high dynamic range.
[0008] Automatic exposure mainly consists of three steps: brightness acquisition, brightness analysis, and exposure adjustment. For digital cameras, image brightness can be directly calculated from the Y component, which accounts for 93% of the image energy in the YUV color space. Brightness analysis requires setting a target brightness value as a reference to determine whether the image is overexposed or underexposed. Exposure adjustment involves adjusting the camera's exposure time and ISO (for digital cameras, this is achieved using electronic signal amplification and gain technology) based on the brightness analysis results, ensuring that the image brightness matches the preset target brightness. Traditional automatic exposure algorithms focus on brightness acquisition and brightness analysis, which are collectively referred to as image metering techniques. Traditional metering techniques mainly include the following categories:
[0009] (1) Global average metering: This technique sums and averages the brightness values of each pixel in the image. (2) Central average metering: This technique assigns a larger proportion of the brightness weight to the central part and a smaller proportion to the surrounding part for average brightness calculation.
[0010] (3) Central portion metering: This technique measures the average brightness of the central rectangular area of the predetermined viewfinder only.
[0011] (4) Zone weighted metering, the biggest difference between it and center average metering is that zone weighted metering divides the viewfinder into several metering zones. Each metering zone is metered independently and then the overall exposure is calculated by weighting the whole zone.
[0012] (5) Metering based on brightness histogram: This technique uses the geometric distribution information of the brightness histogram to adjust the exposure parameters so that the brightness histogram of the output image meets certain requirements.
[0013] However, the shortcomings of existing technologies are:
[0014] Traditional automatic exposure algorithms primarily rely on global or fixed-local brightness statistics, determining exposure parameters by measuring the average brightness of the entire image or a local area. Under complex lighting conditions, global or local average brightness cannot accurately reflect the actual brightness requirements of different areas in the image. For example, in high dynamic range scenes, the brightness difference between bright and dark areas is significant; global average brightness may lead to overexposure in bright areas and underexposure in dark areas. Traditional algorithms also cannot distinguish between key objects and the background in an image. For instance, when shooting portraits, the face is the key object and should be properly exposed, while the background can be adjusted as needed. Global or local brightness statistics cannot achieve this level of detail. Furthermore, traditional algorithms have poor adaptability to complex lighting conditions. In different shooting environments, such as indoors, outdoors, and at night, the effectiveness of global or local brightness statistics methods decreases, failing to provide consistent image quality. The goal of automatic exposure is to accurately expose the subject, and traditional automatic exposure techniques, relying on global or fixed-local brightness statistics, are ineffective when there are multiple subjects in the frame.
[0015] In addition, the terminology commonly used in this technology includes:
[0016] Object detection: The task of object detection is to find all objects of interest in an image and determine their category and location.
[0017] Metering technology: Metering technology refers to the camera's use of its built-in metering system to measure the light intensity in a scene and determine appropriate exposure parameters (such as exposure time and the sensitivity of electronic devices) based on the measured light information, thereby ensuring that the exposure effect of the photo is optimal. The accuracy of metering directly affects the brightness, contrast, and overall visual effect of the photo. Summary of the Invention
[0018] In order to solve the above problems, the purpose of this application is:
[0019] The aim is to provide a content recognition-based intelligent exposure adjustment system and method. By using a target detection algorithm to identify the type and location coordinates of the subject being photographed, the system can automatically and centrally adjust the exposure parameters of the digital camera according to the user's preferences by setting the exposure weight ratio of the area of interest, thereby achieving the best shooting effect and human-computer interaction experience.
[0020] Compared to traditional automatic exposure algorithms, the content recognition-based intelligent exposure adjustment system and method described in this application can solve the following types of problems:
[0021] (1) Insufficient scene adaptability: Traditional automatic exposure algorithms usually rely on the global average brightness or the brightness of a specific area to adjust exposure parameters, which may result in the inability to accurately reflect the true situation of the image under complex lighting conditions (such as high-contrast scenes). This invention can improve the overall image quality by identifying different objects or regions in the image and adjusting the exposure individually for each part.
[0022] (2) Dynamic Range Limitation: Traditional exposure algorithms often struggle to handle scenes with a large dynamic range, easily leading to overexposure or underexposure. This invention enables more precise control of exposure by analyzing image content to optimize exposure settings, reducing overexposure or underexposure and improving the dynamic range performance of the image.
[0023] (3) Loss of detail: In low-light or high-light environments, traditional exposure algorithms may cause loss of image detail. This method, through in-depth analysis of image content, can enhance the details in dark or bright areas while maintaining overall exposure balance, making the image clearer and more natural.
[0024] (4) Poor user experience: Traditional exposure algorithms may not meet users' personalized needs, such as highlighting facial details when shooting portraits. This invention can identify and prioritize specific objects or areas based on user needs, providing a more personalized exposure adjustment scheme and improving the user experience.
[0025] In summary, the content recognition-based intelligent exposure adjustment method solves the problems of adaptability, dynamic range, detail preservation, and user experience of traditional automatic exposure algorithms under complex lighting conditions through more precise scene analysis and exposure control.
[0026] Specifically, the present invention provides a content-recognition-based intelligent exposure adjustment system, the system comprising four modules:
[0027] Image sensor: Captures images of the current scene in real time via a sensor camera; its image data is transmitted to the image processing unit (ISP) in RAW format via the MIPI data transmission protocol for preprocessing; Object detection module: Contains an object detection model, which uses a pre-trained object detection model to identify the main objects in the image and the location coordinates of their regions of interest; the type of object to be identified depends entirely on the object classification ability of the object detection model;
[0028] Image Processing Unit (ISP): Includes an exposure adjustment control algorithm that calculates the ambient brightness suitable for the current scene based on the coordinates of the region of interest identified by the target detection module.
[0029] The calculation process is completed in the exposure adjustment control algorithm embedded in the ISP, and the weighted average ambient brightness of each region of interest is calculated by adding the detection object weight ratio. That is, the calculated weighted average ambient brightness of the region of interest is applied to the exposure adjustment control algorithm in the ISP; when the weighted average ambient brightness is not equal to the target brightness, the exposure time and sensitivity of the sensor camera are appropriately adjusted through the I2C transmission control protocol.
[0030] Real-time image effect display module: The image exposure effect can be observed in real time through the frame marking module OSD.
[0031] The ISP preprocesses the image data received from the image sensor and outputs it in YUV format in two main and secondary bitstreams. The ISP performs preprocessing operations on the image data received from the image sensor, including black level correction, demosaic, automatic white balance, automatic exposure, distortion correction, noise reduction, color space conversion, and sharpening.
[0032] The main bitstream is output to the encoder or local display at a higher resolution of 1920x1080 so that the image effect can be observed in real time.
[0033] The secondary bitstream is output to the target detection algorithm at a lower resolution of 640x360 for content recognition, reducing the amount of computational data required by the target detection algorithm and enabling faster output of results.
[0034] The target detection model used in the target detection is the YOLOv5 target detection model.
[0035] In the ISP, the addition of a detection object weight ratio to calculate the weighted average ambient brightness of each region of interest includes: increasing the weight ratio of the object of interest according to personal preference, so that the object of interest is better exposed correctly, thereby improving the human-computer interaction experience;
[0036] The weight ratio of the detected object is an adjustable hyperparameter between 0 and 1, and its ambient brightness calculation formula is as follows:
[0037]
[0038] In the formula, Lum is the weighted average ambient brightness; i is the pixel position in the image; I(i) is the region of interest (ROI) bounded by the coordinates of the top-left corner P0 and the bottom-right corner P1 of the diagonal of each identified object from P0 to P1; N is the total number of pixels within the metering range; y i This represents the brightness value of each pixel, ranging from 0 to 255.
[0039] The image sensor is a CMOS image sensor.
[0040] This application also relates to a content-recognition-based intelligent exposure adjustment method, applicable to any of the aforementioned systems, comprising the following steps:
[0041] S1: Parameter settings; set the required exposure adjustment step size, target brightness, and weight ratio of the detected object w. i These parameters are transmitted to the automatic exposure control module in the ISP. They are experimental values and must be preset before the system goes into operation. They can be adjusted according to the real-time image display. The exposure adjustment step size controls the convergence speed of automatic exposure, and the target brightness is adjusted to the most suitable sensitivity for the human eye. The weight ratio is w. i This can highlight the exposure concentration of the bounding box of interest formed on the detected object;
[0042] S2: Image Acquisition; Use a camera to acquire images of the current scene, either a single frame or multiple frames; The acquired images can be in RAW format or other formats to retain more image information;
[0043] S3: Content Recognition; A pre-trained deep learning model, i.e., an object detection model, is used to perform content recognition on the image. For each object, the coordinates of its top-left corner P0 and bottom-right corner P1, as well as the object classification label, are extracted. A rectangle with P0 to P1 as the diagonal is used as the region of interest (ROI) for exposure adjustment. S4: Exposure Parameter Calculation; Based on the identified object and its ROI coordinates P0 and P1, the ambient brightness suitable for the current scene is calculated. This calculation process is completed in the exposure adjustment control algorithm embedded in the ISP module, and a detection object weight ratio is added to calculate the weighted average ambient brightness of each ROI. The detection object weight ratio is an adjustable hyperparameter between 0 and 1, and its ambient brightness calculation formula is:
[0044]
[0045] In the formula, Lum is the weighted average ambient brightness; i is the pixel position in the image; I(i) is the region of interest range corresponding to each identified object; N is the total number of pixels within the metering range; y i This represents the brightness value of each pixel, ranging from 0 to 255.
[0046] S5: Exposure Adjustment; Based on the target brightness analysis, it determines whether the calculated ambient brightness is overexposed or underexposed; Based on the principle of prioritizing exposure time adjustment, when the ambient brightness is greater than the target brightness (i.e., overexposed), the ISP will control the reduction of the camera's exposure time or sensitivity by a certain step size; when the ambient brightness is less than the target brightness (i.e., underexposed), the ISP will control the increase of the camera's exposure time or sensitivity by a certain step size; The adjustment step size is an experimental value that needs to be adjusted according to the real-time image display effect. It is given in units of sensor exposure lines. Initially, an arbitrary initial value can be set. If the real-time image exposure convergence speed is slow, the step size will be increased, and vice versa.
[0047] S6: Cyclic adjustment; if the ambient brightness is not equal to the target brightness, it will jump to step S2 and repeat each step in sequence until the ambient brightness equals the target brightness.
[0048] Step S1 further includes:
[0049] The exposure adjustment step size affects the speed of exposure convergence and needs to be adjusted according to the actual situation on site.
[0050] Assuming the object being photographed has a fixed reflectivity of 18%, and 18% is exactly the median gray value in the visual effect produced by the human eye, which is the median gray value for an 8-bit pixel with a bit depth of 128. Therefore, the method described above sets the target brightness to a gray value of 128.
[0051] The weight ratio of the detected object is w i The value ranges from 0 to 1.
[0052] In step S2, image sensor data is acquired in real time, and the acquired image data is transmitted to the image processing unit ISP in RAW format via the MIPI data transmission protocol.
[0053] In step S3, the pre-trained deep learning model, i.e. the object detection model, adopts YOLOv5; the main objects to be identified include faces, landscapes, animals, and buildings.
[0054] The method can be applied to Linux embedded system electronic devices in the fields of photography and videography.
[0055] Therefore, the advantage of this application is:
[0056] To address the aforementioned problems in existing technologies, this invention proposes an intelligent exposure adjustment method and system based on content recognition. This technical solution uses a deep learning model to perform content recognition on images, identifying key objects and scene features, and adjusts exposure parameters based on the recognition results to achieve more precise exposure adjustment. The key points are:
[0057] Multi-level content recognition: Combining multiple content recognition technologies such as object detection, semantic segmentation, and scene classification, it comprehensively analyzes key objects and scene features in the image to ensure the accuracy of exposure adjustment. Dynamic exposure parameter adjustment: Based on the recognition results, it dynamically adjusts exposure parameters such as shutter speed, aperture value, and ISO to ensure that all areas in the image receive appropriate exposure.
[0058] Real-time processing capability: Optimize the computational efficiency of deep learning models to enable real-time processing on embedded devices, meeting the needs of practical applications.
[0059] Strong generalization ability: By training deep learning models with large-scale datasets, the generalization ability under complex lighting conditions is improved, ensuring good exposure results in various scenarios.
[0060] In summary, by combining content recognition technology, this invention achieves more precise and flexible exposure adjustment, significantly improving image quality and user experience. Attached Figure Description
[0061] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.
[0062] Figure 1 This is a schematic diagram of the system architecture involved in this application.
[0063] Figure 2 This is a flowchart illustrating the method involved in this application. Detailed Implementation
[0064] To better understand the technical content and advantages of the present invention, the present invention will now be described in further detail with reference to the accompanying drawings.
[0065] This application proposes a content recognition-based intelligent exposure adjustment system and method, the system architecture of which is as follows: Figure 1 As shown, the composition of the entire system and the relationships between its modules are clearly illustrated. The entire system of this invention consists of the following four modules:
[0066] (1) Image Sensor Module: This module acquires images of the current scene in real time via a CMOS sensor camera. The image data is transmitted to the Image Processing Unit (ISP) in RAW format via the MIPI data transmission protocol for preprocessing to ensure the image conforms to human visual perception. After preprocessing, the image is output in YUV format as a primary and secondary stream. The primary stream is output at a higher resolution (1920x1080) to the encoder or local display for real-time image observation. The secondary stream is output at a lower resolution (640x360) to the object detection algorithm for content recognition, reducing the computational data required for faster result output.
[0067] (2) Object Detection Module: This module includes an object detection model that uses a pre-trained YOLOv5 object detection model to identify the main objects in an image and the location coordinates of their regions of interest. The type of object to be identified depends entirely on the object classification ability of the object detection model. The main objects to be identified can include, but are not limited to, faces, landscapes, animals, and buildings.
[0068] (3) Image Processing Unit: Based on the coordinates of the identified region of interest (ROI), the unit calculates the ambient brightness suitable for the current scene. This calculation is performed within the exposure adjustment control algorithm embedded in the ISP, and a weighted average ambient brightness for each ROI is calculated by incorporating the weight of the detected object. For example, the weight of the ROI can be increased according to personal preference to ensure proper exposure and improve the human-computer interaction experience. The calculated weighted average ambient brightness of the ROI is then applied to the exposure adjustment control algorithm in the ISP. If the weighted average ambient brightness is not equal to the target brightness, the exposure time and sensitivity of the CMOS sensor camera are adjusted appropriately via the I2C transmission control protocol.
[0069] (4) Real-time image effect display module: The image exposure effect can be observed in real time through the frame marking module OSD.
[0070] The process of the method is as follows: Figure 2 The document describes in detail the specific steps of the content recognition-based intelligent exposure adjustment method. The specific implementation steps of this method are as follows:
[0071] Step S1: Parameter Adjustment. Set the required exposure adjustment step size, target brightness, and weight ratio w of the detected object. i These parameters are transmitted to the automatic exposure control module in the ISP. They are experimental values and must be preset before the system goes into operation. They can be adjusted according to the real-time image display. The exposure adjustment step size controls the convergence speed of automatic exposure, and the target brightness is adjusted to the most suitable sensitivity for the human eye. The weight ratio is w. i This highlights the exposure concentration of the bounding box of interest formed by the detected object. For example, setting the weight ratio of a human figure in the image to 1 and the weight ratio of other detected objects to 0 will concentrate the exposure entirely on the human figure. The exposure adjustment step size affects the speed of exposure convergence and needs to be adjusted according to the actual situation. Assuming the subject has a fixed reflectance of 18%, and 18% is exactly the median gray value in the visual perception of the human eye, which is the median gray value for an 8-bit pixel with a bit depth of 128, this method sets the target brightness to a gray value of 128. The weight ratio w of the detected object... i The value ranges from 0 to 1.
[0072] Step S2: Image Acquisition. Use a camera to acquire images of the current scene, which can be single-frame or multi-frame images. The acquired images can be in RAW format or other formats to retain more image information.
[0073] Step S3: Content Recognition. A pre-trained deep learning model (such as YOLOv5) is used to perform content recognition on the image. The main objects to be recognized can include, but are not limited to, faces, landscapes, animals, and buildings. For each object, the coordinates of its upper-left corner P0 and lower-right corner P1, as well as the object classification label, are extracted; a rectangle with P0 to P1 as the diagonal is used as the region of interest for exposure adjustment. Step S4: Exposure Parameter Calculation. Based on the recognized objects and their region of interest coordinates P0 and P1, the ambient brightness suitable for the current scene is calculated. This calculation process is completed in the exposure adjustment control algorithm embedded in the ISP module, and a detection object weight ratio is added to calculate the weighted average ambient brightness of each region of interest. The detection object weight ratio is an adjustable hyperparameter between 0 and 1, and its ambient brightness calculation formula is:
[0074]
[0075] In the formula, Lum is the weighted average ambient brightness; i is the pixel position in the image; I(i) is the rectangular region of interest bounded by the coordinates of each identified object vertex P0 to P1, i.e., the coordinates of the upper left corner P0 and the lower right corner P1 of the diagonal of the rectangle; N is the total number of pixels within the metering range; y i This represents the brightness value of each pixel, ranging from 0 to 255.
[0076] Step S5: Exposure Adjustment; Based on the target brightness, the calculated ambient brightness is determined to be either overexposed or underexposed; Based on the principle of prioritizing exposure time adjustment, when the ambient brightness is greater than the target brightness (i.e., overexposed), the ISP will control the reduction of the camera's exposure time or sensitivity in a certain step. The adjustment step is an experimental value that needs to be adjusted according to the real-time image display effect. It is given in units of sensor exposure lines. Initially, an arbitrary initial value can be set. If the real-time image exposure convergence speed is slow, the step will be increased, and vice versa. When the ambient brightness is less than the target brightness (i.e., underexposed), the ISP will control the increase of the camera's exposure time or sensitivity.
[0077] Step S6: Cyclic Adjustment. If the ambient brightness is not equal to the target brightness, the process will jump to step S2 and repeat each step in a cyclical manner until the ambient brightness equals the target brightness.
[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A content-recognition-based intelligent exposure adjustment system, characterized in that, The system comprises four modules: Image sensor: Captures images of the current scene in real time via a sensor camera; Its image data is transmitted to the image processing unit (ISP) in RAW format via the MIPI data transmission protocol for preprocessing; the object detection module contains an object detection model, which uses a pre-trained object detection model to identify objects in the image and the coordinates of the diagonal vertices of the rectangle of interest; the type of object to be identified depends entirely on the object classification ability of the object detection model; Image Processing Unit (ISP): Includes an exposure adjustment control algorithm. Based on the coordinates of the region of interest (ROI) identified by the target detection module, it calculates the ambient brightness suitable for the current scene for the identified object. The calculation process is completed within the exposure adjustment control algorithm embedded in the ISP, and the weighted average ambient brightness of each ROI is calculated by incorporating the weight ratio of the detected object. That is, the calculated weighted average ambient brightness of the ROI is applied to the exposure adjustment control algorithm in the ISP. When the weighted average ambient brightness is not equal to the target brightness, the exposure time and sensitivity of the sensor camera are appropriately adjusted through the I2C transmission control protocol. Real-time image effect display module: The image exposure effect can be observed in real time through the frame marking module OSD.
2. The intelligent exposure adjustment system based on content recognition according to claim 1, characterized in that, The ISP performs preprocessing operations on the image data received from the image sensor. The preprocessing includes black level correction, depigmentation, automatic white balance, automatic exposure, distortion correction, noise reduction, color space conversion, and sharpening. After that, it outputs the image data in YUV format in two main and secondary bitstreams. The main bitstream is output to the encoder or local display at a higher resolution of 1920x1080 so that the image effect can be observed in real time. The secondary bitstream is output to the target detection algorithm at a lower resolution of 640x360 for content recognition, reducing the amount of computational data required by the target detection algorithm and enabling faster output of results.
3. The intelligent exposure adjustment system based on content recognition according to claim 1, characterized in that, The target detection model used in the target detection is the YOLOv5 target detection model.
4. The intelligent exposure adjustment system based on content recognition according to claim 1, characterized in that, In the ISP, the addition of a detection object weight ratio to calculate the weighted average ambient brightness of each region of interest includes: increasing the weight ratio of the object of interest according to personal preference, so that the object of interest is better exposed correctly, thereby improving the human-computer interaction experience; The weight ratio of the detected object is an adjustable hyperparameter between 0 and 1, and its ambient brightness calculation formula is as follows: In the formula, Lum is the weighted average ambient brightness; i is the pixel position in the image; I(i) is the region of interest (ROI) bounded by the coordinates of the top-left corner P0 and the bottom-right corner P1 of the diagonal of each identified object from P0 to P1; N is the total number of pixels within the metering range; y i This represents the brightness value of each pixel, ranging from 0 to 255.
5. The intelligent exposure adjustment system based on content recognition according to claim 1, characterized in that, The image sensor is a CMOS image sensor.
6. A content-recognition-based intelligent exposure adjustment method, characterized in that, The method is applicable to any of the systems described in claims 1-5 above, and includes the following steps: S1: Set parameter tuning; Set the required exposure adjustment step size, target brightness, and weight ratio w of the detected object. i These parameters are transmitted to the automatic exposure control module in the ISP. They are experimental values and must be preset before the system goes into operation. They can be adjusted according to the real-time image display. The exposure adjustment step size controls the convergence speed of automatic exposure, and the target brightness is adjusted to the most suitable sensitivity for the human eye. The weight ratio is w. i This can highlight the exposure concentration of the bounding box of interest formed on the detected object; S2: Image Acquisition; Use a camera to acquire images of the current scene, either a single frame or multiple frames; The acquired images can be in RAW format or other formats to retain more image information; S3: Content Recognition; Use a pre-trained deep learning model, i.e., an object detection model, to perform content recognition on the image; For each object, extract the coordinates of its upper left corner P0 and lower right corner P1, as well as the object classification label; The rectangle from P0 to P1 is used as the region of interest for exposure adjustment; S4: Exposure parameter calculation; Based on the identified object and its region of interest coordinates P0 and P1, the ambient brightness suitable for the current scene is calculated. This calculation process is completed in the exposure adjustment control algorithm embedded in the ISP module, and a detection object weight ratio is added to calculate the weighted average ambient brightness of each region of interest. The detection object weight ratio is an adjustable hyperparameter between 0 and 1, and its ambient brightness calculation formula is as follows: In the formula, Lum is the weighted average ambient brightness; i is the pixel position in the image; I(i) is the rectangular region of interest bounded by the coordinates of each identified object vertex P0 to P1, i.e., the coordinates of the upper left corner P0 and the lower right corner P1 of the diagonal of the rectangle; N is the total number of pixels within the metering range; y i This represents the brightness value of each pixel, ranging from 0 to 255. S5: Exposure Adjustment; Based on the target brightness analysis, it determines whether the calculated ambient brightness is overexposed or underexposed; Based on the principle of prioritizing exposure time adjustment, when the ambient brightness is greater than the target brightness (i.e., overexposed), the ISP will control the reduction of the camera's exposure time or sensitivity by a certain step size; when the ambient brightness is less than the target brightness (i.e., underexposed), the ISP will control the increase of the camera's exposure time or sensitivity by a certain step size; The adjustment step size is an experimental value that needs to be adjusted according to the real-time image display effect. It is given in units of sensor exposure lines. Initially, an arbitrary initial value can be set. If the real-time image exposure convergence speed is slow, the step size will be increased, and vice versa. S6: Cyclic adjustment; if the ambient brightness is not equal to the target brightness, it will jump to step S2 and repeat each step in sequence until the ambient brightness equals the target brightness.
7. The intelligent exposure adjustment method based on content recognition according to claim 6, characterized in that, Step S1 further includes: The exposure adjustment step size affects the speed of exposure convergence and needs to be adjusted according to the actual situation on site. Assuming the object being photographed has a fixed reflectivity of 18%, and 18% is exactly the median gray value in the visual effect produced by the human eye, which is the median gray value for an 8-bit pixel with a bit depth of 128. Therefore, the method described above sets the target brightness to a gray value of 128. The weight ratio of the detected object is w i The value ranges from 0 to 1.
8. The intelligent exposure adjustment method based on content recognition according to claim 6, characterized in that, In step S2, image sensor data is acquired in real time, and the acquired image data is transmitted to the image processing unit ISP in RAW format via the MIPI data transmission protocol.
9. The intelligent exposure adjustment method based on content recognition according to claim 6, characterized in that, In step S3, the pre-trained deep learning model, i.e. the object detection model, adopts YOLOv5; the objects to be identified include human figures, vehicles, animals, and buildings.
10. The intelligent exposure adjustment method based on content recognition according to claim 6, characterized in that, The method can be applied to Linux embedded system electronic devices in the fields of photography and videography.