Automobile data recorder with license plate identification function and license plate identification method

Abstract

The invention discloses an automobile data recorder with a license plate identification function and a license plate identification method, and the automobile data recorder comprises a multi-lens camera module which can shoot a dynamic image covering a plurality of visual angles, and then obtains the position of one or more license plates in the image according to the characteristics of the license plates. The method comprises the following steps: judging one or more regions of interest about a license plate in each frame of the dynamic image, setting individual exposure compensation weights of a plurality of regions according to the size and position characteristics of the license plate, and then performing exposure compensation of different degrees on each frame of the dynamic image according to the individual exposure compensation weights of the plurality of regions, and carrying out license plate identification on the license plate image subjected to exposure compensation.

Description

Technical Field

[0001] The specification discloses a driving recorder, specifically a driving recorder with a license plate recognition function capable of shooting multiple areas and a license plate recognition method running therein. Background Art

[0002] A dashcam that uses Automatic License Plate Recognition (ALPR) integrates multiple technical components. For example, it has a camera module that, in addition to recording the driving, also needs to obtain high-resolution and detailed images to perform license plate recognition regardless of the ambient light; it also has optical character recognition (OCR) to recognize the character string in the license plate.

[0003] Conventional dashcams usually shoot images with a wide viewing angle and need to shoot clear images in low light environments. If license plate recognition is to be performed, it is difficult to obtain high-resolution license plate images that can successfully identify the content in wide-view images or low-light images. Furthermore, even if high-resolution photosensitive components, such as 4K photosensitive components, are used, it is still challenging to correctly identify the license plate in a relatively small area of ​​the image.

[0004] Furthermore, conventional dashcams are not equipped with high-performance image processors. Therefore, when the computing power is insufficient, it is difficult to perform real-time license plate recognition operations. Therefore, it is common to store image files and then use additional computer equipment to perform subsequent license plate recognition.

[0005] On the other hand, dashcams often capture images in various environments, so they need to identify license plates that change due to various light sources. They often encounter situations where the license plate is underexposed, overexposed, or the shutter speed is not fast enough, resulting in blurred images and being unable to be identified. All of the above situations will lead to the problem that the accuracy of license plate recognition using conventional dashcams is not high or the license plate recognition cannot be performed in real time. Summary of the invention

[0006] In response to the need for a driving recorder to identify license plates that are not easily identified due to insufficient light exposure, overexposure or insufficient shutter speed, the present invention proposes a license plate recognition method and a driving recorder with a license plate recognition function, wherein the conventional problem is solved by a multi-zone exposure compensation technology; on the other hand, when faced with the problem of insufficient resolution making it difficult to perform license plate recognition, the method proposed in the present invention also provides a solution through a multi-lens camera module.

[0007] According to the implementation scheme, the main components of a driving recorder with a license plate recognition function include a processing circuit and a camera module. In the license plate recognition method executed by the processing circuit, a dynamic image covering multiple areas is captured by the camera module, and one or more areas of interest in each frame of the dynamic image are determined according to the characteristics of the license plate, so as to set individual exposure compensation weights for the multiple areas, and then different degrees of exposure compensation are performed on each frame of the dynamic image according to the individual exposure compensation weights of the multiple areas. After obtaining the image of the license plate in each frame that has been exposed compensated and has the original resolution, license plate recognition is performed.

[0008] Furthermore, when the dash cam obtains dynamic images, the size of each frame of the image can be changed to a reduced image, and then the license plate position can be identified in the reduced image. Since the appearance characteristics of the license plate are larger than the text of the license plate content, after reducing the resolution, the license plate position in the reduced image can be identified using only relatively small resources in a small embedded system.

[0009] Furthermore, the features of the license plate, such as the size and position of the license plate of the vehicle in the foreground captured by the dashcam, can be identified by an object recognition model derived from a machine learning program to reduce the position of the license plate in the image.

[0010] Among them, after the license plate position is recognized by reducing the image, it can be mapped to the original frames of the image to obtain the license plate with the original resolution to obtain the image of the license plate with the original resolution after exposure compensation for license plate recognition.

[0011] Furthermore, the license plate may be recognized using character recognition technology or a license plate recognition model derived from the above-mentioned machine learning program.

[0012] Furthermore, when shooting dynamic images, the processing circuit sets the maximum gain value, the minimum shutter speed and the exposure value of one or more photosensitive components in the camera module, and performs image processing on each frame of the generated image, including at least one of noise reduction, contrast adjustment and grayscale conversion. Regional tone mapping can also be used to enhance the visibility of each frame of the image.

[0013] According to one embodiment, the camera module is a camera system with multiple lenses, including at least a first camera lens for shooting wide-angle scene images and a second camera lens for shooting narrow-angle detail images, capturing the dynamic images covering multiple areas of wide and narrow angles.

[0014] According to another embodiment, the camera module may be a multi-view camera system, which includes a rotatable camera lens capable of changing the shooting angle. During operation, the shooting angle is controlled by a rotating shaft to obtain dynamic images of multiple areas covering different angles.

[0015] In another embodiment, the camera module is a camera system with multiple lenses, which includes at least a first camera lens for shooting a first viewing angle and a second camera lens for shooting a second viewing angle, so as to respectively shoot dynamic images of multiple areas at different viewing angles.

[0016] Furthermore, once the gun is determined to be pulled out of the holster according to the sensing result, the control unit sends a trigger signal to start the video recording device, and when it is determined that the gun is out of the holster for a time period exceeding a threshold, the sensing device enters a power saving mode. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram showing an embodiment of a driving recorder with a license plate recognition function;

[0018] Figure 2 Displays a diagram of the main circuit implementation in a driving recorder with a license plate recognition function;

[0019] Figure 3 A flow chart showing an embodiment of a license plate recognition method;

[0020] FIG. 4A to FIG. 4C An example diagram showing an example of obtaining a region of interest in a license plate recognition method;

[0021] Figure 5 A schematic diagram showing an embodiment of setting multi-region exposure compensation weights in a license plate recognition method;

[0022] Figure 6 A schematic diagram showing an embodiment of a shooting angle of view when a vehicle is driving;

[0023] FIG. 7A to FIG. 7C A first schematic diagram showing an embodiment of a multi-lens driving recorder shooting viewing angle when a vehicle is driving;

[0024] FIG. 8A to FIG. 8B A second schematic diagram showing an embodiment of a multi-lens driving recorder shooting angle when a vehicle is driving; and

[0025] FIG. 9A to FIG. 9B A third schematic diagram showing an embodiment of shooting angles of a multi-lens driving recorder when a vehicle is driving. DETAILED DESCRIPTION

[0026] The present invention provides a driving recorder with a license plate recognition function and a license plate recognition method executed therein. One of the embodiments of the present invention is to solve the problem that when recognizing a license plate, it is easy to be affected by insufficient light source, insufficient shutter speed, blurry image or overexposure of the license plate, which may cause recognition errors, or to solve the problem of high reflection caused by the design of reflective license plates. Another embodiment proposes a driving recorder with a multi-lens camera module, which can solve the problem of difficulty in license plate recognition caused by insufficient resolution of the license plate in the captured image.

[0027] For related implementation scenarios, please refer to Figure 1 The schematic diagram of the embodiment of the driving recorder with license plate recognition function is shown, wherein the driving recorder 100 installed in the vehicle is shown, and the vehicle foreground image is captured through the vehicle windshield, and this scenario schematically shows a front vehicle 101 and a front vehicle 102. It is mentioned here that the setting of the driving recorder 100 can also be used to capture images from the rear of the vehicle or other perspectives.

[0028] In particular, the driving recorder 100 has a license plate recognition function, which can recognize the license plates of the front vehicle 101 and the front vehicle 2 102 from the captured dynamic image. The circuit design can refer to Figure 2 The diagram of the main circuit implementation in the driving recorder is shown.

[0029] Figure 2 The main circuit of the driving recorder 100 includes a processing circuit that can execute the license plate recognition program, such as a single-chip system 201, which (but not limited to) may include multiple processing units, such as an image signal processor (ISP) 201a, a central processing unit (CPU) 201b, a graphics processing unit (GPU) 201c, and a neural-network processing unit (NPU) 201d. The single-chip system 201 is electrically connected to other circuit components, such as a camera module 205 that can transmit image signals, a driving circuit 203 that drives the camera module 205 to operate, a storage medium 207, and a communication circuit 209. The camera module 205 can be a camera system with a multi-lens module. In the illustrated embodiment, it includes a first camera lens 251 and a second camera lens 252. The camera module 205 further includes a photosensitive component and a processing circuit (not shown) that generates images through the first camera lens 251 and the second camera lens 252.

[0030] Furthermore, the driving recorder 100 is provided with a storage medium 207 for storing the dynamic image signal captured by the camera module 205 and the result data after the license plate recognition process. The communication circuit 209 can then transmit the relevant images and data to an external system.

[0031] It is worth mentioning that the dash cam 100 is an edge device with low computing power. Therefore, in the license plate recognition method executed by the single-chip system 201, it is necessary to consider that the computing performance will be affected by the higher resolution and larger size of the image, which will cause problems in the operation of the dash cam 100, such as overheating, crashes, insufficient computing power, and frame drops. Therefore, the dash cam with license plate recognition function proposed by the present invention will consider computing resources such as computing power and memory and various image processing problems when calculating.

[0032] According to the embodiment of license plate recognition, refer to Figure 2 and Figure 3 The flowchart of the embodiment of the license plate recognition method shown in FIG. FIG. 4A to FIG. 4C An embodiment diagram of obtaining a region of interest in a license plate recognition method is shown.

[0033] According to the embodiment, considering that the driving recorder 100 will capture images in different environments, the images will have uneven exposure under the light sources of various environments, and the images may have problems such as too bright foreground, too dark foreground, overexposure and / or underexposure of the license plate. In addition, the camera module 205 in the driving recorder 100 needs to obtain high-quality wide-angle images and also needs to enhance the image quality in low-light environments. Therefore, the license plate recognition method proposed in the present invention will propose a solution to divide the image into multiple areas for exposure compensation in response to this problem. Furthermore, the vehicle usually accounts for a small proportion of the wide-angle image, which also makes it difficult to recognize the license plate. Therefore, one of the embodiments is to propose a solution of a multi-lens camera module.

[0034] exist Figure 3 In the displayed license plate recognition method embodiment process, the driving recorder 100 with a license plate recognition function will capture dynamic images covering multiple perspectives through the camera module 205 when in operation (step S301). During the shooting process, image pre-processing can be performed first, and the gain value (such as the highest gain value), shutter speed (such as the lowest shutter speed) and exposure value of one or more photosensitive components in the camera module are first set by the processing circuit in the driving recorder; after obtaining a dynamic image covering continuous frames, preliminary image processing can be performed on each frame of the generated image, including noise reduction, contrast adjustment and grayscale, etc., and one or more processing procedures can be selected according to needs.

[0035] It is worth mentioning that in order to obtain clear images, the parameters of the camera module 205 in the driving recorder 100 are mainly set with shutter priority, and other parameters are adjusted accordingly. For example, when the scene is low light, the camera module will adjust the gain of the photosensitive component to increase the sensitivity (ISO), but it will also increase the noise in the image. If de-nosing is performed at this time, the image will be blurred. Therefore, local tone mapping can be used in the image processing program to enhance the visibility of each frame of the image and avoid the problem of blurring.

[0036] In particular, since the images are usually wide-angle images, during the license plate recognition process, the vehicle position in the image (i.e., the license plate position) and the surrounding environment will be identified and segmented to obtain a dynamic image covering multiple areas.

[0037] For reference Figure 4A One of the frames 400 of the foreground image captured by the dash cam is shown, which schematically shows a vehicle 401 in front. The license plate of the vehicle 401 is then identified to determine one or more regions of interest (ROI) based on the size and position characteristics of the license plate (step S303). The single-chip system 201 in the dash cam 100 sets individual exposure compensation weights for multiple regions based on the regions of interest (step S305) to perform exposure compensation for different regions (step S307). In the automatic exposure process, the individual exposure compensation weights for the multiple regions are as follows: Figure 5 As shown, the example shows that the exposure compensation weight of each area is 5%, 45%, 30% and 20%, so that different degrees of exposure compensation can be performed on each frame of the dynamic image.

[0038] Furthermore, when considering that the computational resources required for image processing frame by frame are too large, in one embodiment, when obtaining the dynamic image, the size of each frame image can be changed to a reduced image, or the resolution can be reduced (step S309). Figure 4B The image with reduced resolution is shown schematically. After the image processing procedure, the license plate position is identified according to the features of the license plate in the reduced image (or the image with reduced resolution) (step S311). Figure 4B The image produced by reducing the size or resolution of the high-quality image taken by the driving recorder is displayed, including the vehicle 401 ′ after image processing, based on which the first area of ​​interest 403 is determined, and the license plate position is further searched to form the second area of ​​interest 405.

[0039] It is worth mentioning that in the above-mentioned license plate recognition method, it can be referred to that in step S303, after determining the area of ​​interest and considering that each object in each frame has a different exposure, multiple areas can be dynamically set for each frame image, and the sizes of these areas can be different. Among them, as in steps S305 and S307, different degrees of exposure compensation are performed in different areas, mainly to solve the problem that the image taken by the driving recorder may be unevenly bright and dark due to ambient light, or it may be caused by the objects in front being located at different distances when the headlights are irradiated in front. The problem of uneven brightness. Because the objects at each position in front of the vehicle will have different brightness reflections, the problem of uneven exposure in different areas will be generated in the image, especially for each license plate in the image. The exposure has a different amount, so the dynamic exposure compensation mechanism is used to correct the overexposed and underexposed areas to obtain a uniformly exposed license plate image.

[0040] For example, see Figure 5 The displayed image is a frame 500 of a vehicle captured by a dashcam. For example, a vehicle is located in the middle of the frame 500. One or more regions of interest in each frame of the dynamic image can be determined based on the characteristics of the license plate. Afterwards, the entire frame 500 can be divided into multiple regions according to the distance, exposure, and the location of the region of interest, so as to set the exposure compensation weights of the multiple regions. In actual implementation, the number of regions and the exposure compensation weight values ​​set are not limited to this example.

[0041] according to Figure 5 In the schematic diagram of the embodiment shown, the frame 500 is divided into multiple areas, including a first area 501, where the exposure compensation weight of this area is set to 5%; a second area 502, where a higher (the highest in this case) exposure compensation weight of 45% is set because this area covers the vehicle 510; a third area 503, where this area is within the determined area of ​​interest, and the exposure compensation weight can be set to 30%; and a fourth area 504, where this area covers a larger range and is close to the driving recorder, and the exposure compensation weight can be set to 20%.

[0042] After different degrees of exposure compensation are completed for each frame according to the above steps, the entire frame image will present a uniformly exposed picture, so that the license plate recognition method can successfully obtain the image of the license plate with original resolution after exposure compensation in each frame, such as Figure 4C The license plate 405 ′ obtained through image acquisition is displayed for license plate recognition.

[0043] According to the license plate recognition method embodiment, the license plate in each frame can be identified using an object recognition model obtained through a machine learning program. When training a specific object recognition model, a large amount of image data of an object (such as a license plate) in various light and shadows and various angles is first generated and input into a machine learning algorithm. Through image processing technology, the image features of the object are learned and classification is performed, thereby training an object recognition model that can recognize objects and can recognize objects in images. Taking a license plate as an example, the license plate feature is a square appearance, the background color is uniform white or other colors, and the license plate string is generally a combination of dark English and numbers. In this way, a license plate recognition model can be trained to recognize the position of the license plate in an image that has been reduced in size or resolution.

[0044] through Figure 3After performing image recognition of the license plate position in step S311, the license plate position can be mapped to each frame of the original image to obtain the original resolution of the license plate, so as to obtain an image of the license plate with original resolution after exposure compensation (step S313), and then perform license plate recognition. The license plate recognition method can use a character recognition technology to recognize the license plate, such as an optical character recognition technology, or a license plate recognition model obtained by a machine learning program to recognize the license plate (step S315).

[0045] In the license plate recognition method proposed by the present invention, firstly, different exposure compensation weights can be set for different areas in the image according to the region of interest, and different degrees of exposure compensation are performed on each area of ​​the image before performing license plate recognition. In addition to this license plate recognition stage, a multi-lens camera system can also be used to continue to improve the method of performing license plate recognition only with wide-angle images.

[0046] Reference first Figure 6 The schematic diagram of the embodiment of the shooting angle when the vehicle is driving is shown. Figure 6 Displaying a road with multiple vehicles traveling simultaneously, the main vehicle 60 uses a single-lens camera module to obtain Figure 6 The image of the shooting angle 600 is displayed in the figure, and the shooting angle 600 refers to the range of light that can be sensed by the photosensitive component of the camera module in the driving recorder in front of the main vehicle 60 or in a specific direction.

[0047] FIG. 7A to FIG. 7C A first schematic diagram showing an embodiment of using a multi-lens driving recorder to shoot viewing angles while a vehicle is traveling. This example shows that the camera module in the driving recorder is a camera system with multiple lenses, which includes at least a first camera lens for shooting wide-angle situational images, and a second camera lens for shooting relatively narrow-angle detail images. The camera lens can simultaneously shoot dynamic images covering multiple areas of wide and narrow viewing angles.

[0048] Fig. 7A The driving recorder on the display host vehicle 70 includes a camera system with at least two lenses, which simultaneously captures two images, a first perspective 701 with a wide viewing angle and a second perspective 702 with a narrow viewing angle. The first perspective 701 is a wide-view image in front of the host vehicle 70, and the second perspective 702 uses a longer focal length lens, so it can capture the image of the vehicle in front more clearly.

[0049] Then use license plate recognition technology, such as Figure 7B As shown, the image of different viewing angles is used to determine the region of interest 703 formed by the vehicle and the license plate in the image. And further as Figure 7C As shown, the content of the license plate 705 in the region of interest 703' within a smaller viewing angle is identified.

[0050] FIG. 8A to FIG. 8B Next, a second schematic diagram of an embodiment of using a multi-lens driving recorder to shoot viewing angles while a vehicle is driving is shown. This example shows that the camera module of the driving recorder adopts a multi-viewing angle camera system, including a rotatable camera lens that can change the shooting viewing angle. The shooting viewing angle can be controlled by a rotating shaft during operation to obtain dynamic images covering multiple areas of different viewing angles.

[0051] Fig. 8A There are multiple vehicles on the lane, wherein the multi-lens camera system in the driving recorder in the host vehicle 80 has a rotatable camera lens that can change the shooting angle, wherein the main lens shoots a wide-angle image of the first angle of view 801, wherein the processing circuit is driven by the driving circuit (such as Figure 2 The display driving circuit 203 controls the rotatable camera lens to take images with narrower viewing angles such as the second viewing angle 802 and the third viewing angle 803 in sequence, and can clearly take pictures of the first car 811 and the second car 812 in front of the main car 80. In actual implementation, the rotatable camera lens can also obtain images of other viewing angles.

[0052] In addition to obtaining images covering environmental information from the first viewing angle 801, the driving recorder can also capture images of the first car 811 from the second viewing angle 802 and the second car 812 from the third viewing angle 803 through a rotatable camera lens, so that the license plate recognition process can be performed. Figure 8B The display screen covers the first car 811 ′ and the second car 812 ′ captured from different perspectives, so the first license plate 813 on the first car 811 ′ and the second license plate 814 on the second car 812 ′ are clearly obtained, and the license plates of the vehicles can be accurately identified.

[0053] FIG. 9A to FIG. 9B A third schematic diagram showing an embodiment of using a multi-lens driving recorder to shoot viewing angles while a vehicle is driving. The camera module of this driving recorder is a camera system with multiple lenses, and in particular, includes at least a first camera lens for shooting a first viewing angle and a second camera lens for shooting a second viewing angle, wherein the first viewing angle and the second viewing angle may be the same angle but different directions, or the two angles may be different and the directions may also be different, so as to be able to simultaneously shoot dynamic images of multiple areas of different viewing angles.

[0054] Fig. 9A The main vehicle 90 in the lane is shown to simultaneously shoot images of the first perspective 901 and the second perspective 902 with a driving recorder. In this example, the first perspective 901 and the second perspective 902 partially overlap. The shooting results can be referred to Fig. 9B The image shown includes a first viewing angle image 911 covering a first viewing angle 901 and a second viewing angle image 912 covering a second viewing angle 902 .

[0055] This example shows that in the picture captured by the multi-lens camera module, the first perspective image 911 includes vehicles 921 and 922 in front of the main vehicle 90, and the second perspective image 912 includes vehicles 922, 923 and 924, wherein both perspectives include vehicle 922. In this way, the driving recorder can obtain a clearer license plate of the vehicle in front, which is convenient for license plate recognition.

[0056] In summary, according to the above-mentioned embodiments of the driving recorder with license plate recognition function and the license plate recognition method running therein, in order to meet the demand of the driving recorder to perform license plate recognition under various light sources, or to avoid the problem of high reflection caused by the design of the license plate, a dynamic exposure compensation mechanism is used in the license plate recognition method to obtain an image with uniform exposure; on the other hand, a multi-lens camera module is proposed to provide images with different viewing angles, which can solve the problem that wide-viewing angle images are difficult to perform license plate recognition due to insufficient resolution.

[0057] The contents disclosed above are only preferred feasible embodiments of the present invention, and are not intended to limit the scope of the present invention. Therefore, all equivalent technical changes made using the contents of the present invention's specification and drawings are included in the scope of the present invention.

Claims

1. A driving recorder with license plate recognition function, characterized in that: The driving recorder comprises: a processing circuit; and A camera module for capturing a dynamic image covering multiple viewing angles; The processing circuit executes a license plate recognition method, including: Shooting the dynamic image covering multiple areas by the camera module; determining one or more regions of interest in each frame of the dynamic image according to a feature of a license plate to set exposure compensation weights for the multiple regions individually; Performing exposure compensation of different degrees on each frame of the dynamic image according to the individual exposure compensation weights of the plurality of regions; and The image of the license plate after the exposure compensation in each frame is obtained to perform license plate recognition.

2. The driving recorder with license plate recognition function according to claim 1, characterized in that: After the exposure compensation is completed, the size of each frame image of the driving recorder is changed to a reduced image, and then the position of the license plate is identified in the reduced image.

3. The driving recorder with license plate recognition function according to claim 2, characterized in that: The characteristics of the license plate are the size and position of the license plate of a vehicle in the foreground captured by the driving recorder.

4. The driving recorder with license plate recognition function according to claim 3, characterized in that: The location of the license plate in the reduced image is identified using an object recognition model derived through a machine learning process.

5. The driving recorder with license plate recognition function according to claim 2, characterized in that: After the position of the license plate is identified through the reduced image, it is mapped to each frame of the original image to obtain an image of the license plate with the original resolution after the exposure compensation, and then the license plate is recognized.

6. The driving recorder with license plate recognition function according to claim 5, characterized in that: The license plate is identified by using a character recognition technology or a license plate recognition model derived from a machine learning program.

7. The driving recorder with license plate recognition function according to claim 1, characterized in that: According to the determined one or more regions of interest and taking into account the exposure of different objects in each frame, the multiple regions of different sizes are dynamically set for each frame, so as to perform different degrees of exposure compensation in different regions.

8. The driving recorder with license plate recognition function according to claim 1, characterized in that: When shooting the dynamic image, the processing circuit sets a maximum gain value, a minimum shutter speed and an exposure value of one or more photosensitive components in the camera module, and performs image processing on each frame of the generated image, including at least one of noise reduction, contrast adjustment and grayscale.

9. The driving recorder with license plate recognition function according to claim 8, characterized in that: Apply a regional tone mapping to enhance the visibility of each frame.

10. The driving recorder with license plate recognition function according to any one of claims 1 to 9, characterized in that: The camera module is a camera system with multiple lenses, including at least a first camera lens for shooting wide-angle scene images and a second camera lens for shooting narrow-angle detail images, and shoots the dynamic images covering wide and narrow angles.

11. The driving recorder with license plate recognition function according to any one of claims 1 to 9, characterized in that: The camera module is a multi-angle camera system, including a rotatable camera lens capable of changing the shooting angle. During operation, the shooting angle is controlled by a rotating shaft to obtain the dynamic images covering different angles.

12. The driving recorder with license plate recognition function according to any one of claims 1 to 9, characterized in that: The camera module is a camera system with multiple lenses, which at least includes a first camera lens for shooting a first viewing angle and a second camera lens for shooting a second viewing angle, so as to respectively shoot the dynamic images at different viewing angles.

13. A license plate recognition method, executed in a driving recorder, characterized in that: The method comprises: Shooting a dynamic image covering multiple viewing angles through a camera module; determining one or more regions of interest in each frame of the dynamic image according to the characteristics of a license plate to set individual exposure compensation weights for the multiple regions; Performing exposure compensation of different degrees on each frame of the dynamic image according to the exposure compensation weights of the multiple regions; and The image of the license plate after the exposure compensation in each frame is obtained to perform license plate recognition.

14. The license plate recognition method according to claim 13, characterized in that: After the exposure compensation is completed, the size of each frame image of the driving recorder is changed to a reduced image, and then the position of the license plate is identified in the reduced image.

15. The license plate recognition method according to claim 14, characterized in that: The location of the license plate in the reduced image is identified using an object recognition model derived through a machine learning process.

16. The license plate recognition method according to claim 15, characterized in that: After the position of the license plate is identified through the reduced image, it is mapped to each frame of the original image to obtain an image of the license plate with the original resolution after the exposure compensation, and then the license plate is recognized.

17. The license plate recognition method according to claim 16, characterized in that: The license plate is identified by using a character recognition technology or a license plate recognition model derived from a machine learning program.

18. The license plate recognition method according to claim 13, characterized in that: When shooting the dynamic image, a processing circuit in the driving recorder sets a maximum gain value, a minimum shutter speed and an exposure value of one or more photosensitive components in the camera module, and performs image processing on each frame of the generated image, including at least one of noise reduction, contrast adjustment and grayscale.

19. The license plate recognition method according to claim 18, characterized in that: Apply a regional tone mapping to enhance the visibility of each frame.

20. The license plate recognition method according to any one of claims 13 to 19, characterized in that: According to the determined one or more regions of interest and taking into account the exposure of different objects in each frame, the multiple regions of different sizes are dynamically set for each frame, so as to perform different degrees of exposure compensation in different regions.

Images