Method and system for increasing recognition distance of artificial intelligence image recognition system

Abstract

The invention provides a method and system for increasing the recognition distance of an artificial intelligence image recognition system, and belongs to the technical field of target recognition, and the method comprises the steps: S1, carrying out the region division of each frame of image in captured continuous frames of images, and enabling each frame of image to correspond to a preset number of image block regions; s2, respectively selecting a part of image block region from each frame of image as a region to be recognized, and inputting the region to be recognized into an artificial intelligence image recognition system to obtain a recognition result; and S3, when the target object is detected, determining a target position of the to-be-identified area in the corresponding image frame, and dynamically updating the to-be-identified area in a subsequent frame according to the target position. The method has the beneficial effects that the recognition distance of an AI recognition system is remarkably increased and the recognition efficiency and accuracy of the system are enhanced by performing region division on the continuous frame images, selectively recognizing partial image block regions and dynamically updating the to-be-recognized region after the target object is detected on the premise of not increasing the hardware cost.

Description

Technical Field

[0001] The present invention relates to the technical field of target recognition, and in particular, to a method and system for improving the recognition distance of an artificial intelligence image recognition system. Background Art

[0002] Artificial Intelligence (AI) image recognition systems have been widely applied to various intelligent devices, bringing great convenience to people's lives and work. Such systems capture consecutive frame images based on a camera, and then send the captured images into an AI image recognition model to identify target objects in the images using advanced algorithms and technologies.

[0003] Currently, for an AI image recognition system based on a camera, when dealing with long-distance image recognition tasks, its recognition effect will significantly decline. The root cause of this problem is that as the image shooting distance increases, the number of pixels occupied by the target object in the image will correspondingly decrease, resulting in the loss of image detail information, which in turn affects the recognition accuracy of the AI algorithm.

[0004] In the prior art, high-resolution cameras or hardware upgrades are usually relied on to enhance the recognition distance. However, both of these methods for enhancing the recognition distance will increase the hardware cost and power consumption, and it is difficult to achieve efficient long-distance image recognition on devices with limited computing power (such as Android TVs). In addition, when the prior art improves the recognition distance, it often sacrifices the recognition range and the integrity of functions, reducing the practicality and user experience of the system. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a method for improving the recognition distance of an artificial intelligence image recognition system; on the other hand, it also provides a system for improving the recognition distance of an artificial intelligence image recognition system.

[0006] The technical problems solved by the present invention can be realized by the following technical solutions:

[0007] The first aspect of the present invention is to provide a method for improving the recognition distance of an artificial intelligence image recognition system, including:

[0008] Step S1, performing region division on each frame image in the captured consecutive frame images to obtain a preset number of image block regions corresponding to each frame image;

[0009] Step S2, respectively selecting some image block regions from each frame image in the consecutive frame images as regions to be recognized, and inputting the selected regions to be recognized into the artificial intelligence image recognition system to obtain a recognition result;

[0010] Step S3, when the recognition result detects that the area to be recognized contains the target object, determine the target position of the area to be recognized in the corresponding image frame, and dynamically update the area to be recognized according to the target position in subsequent frames.

[0011] Preferably, the step S1 includes:

[0012] Step S11, obtain consecutive frame images captured by a camera, where the consecutive frame images include multiple frames of images;

[0013] Step S12, for each frame image in the consecutive frame images, perform equal-area division in the horizontal and vertical directions of the image to obtain a preset number of image block areas.

[0014] Preferably, the area of the image block area accounts for one-fourth of the frame image where it is located.

[0015] Preferably, the area to be recognized is one of the image block areas in each frame image, and before the target object is recognized, the positions of the image block areas selected in different frame images are different.

[0016] Preferably, in the step S2, select partial image block areas from each frame image in the consecutive frame images as the area to be recognized according to a preset rule, where the preset rule is the selection order of the positions of the areas to be recognized in consecutive preset number of frame images.

[0017] Preferably, in the step S2, all the image block areas selected from the consecutive preset number of frame images in the consecutive frame images cover all areas of the complete image.

[0018] Preferably, the step S3 further includes:

[0019] When the recognition result detects that the area to be recognized does not contain the target object, return to the step S2, continue to select partial image block areas from the next frame image as the area to be recognized, and input them into the artificial intelligence image recognition system for target recognition until the target object is recognized.

[0020] Preferably, the step S3 includes:

[0021] Step S31, when the recognition result detects that the area to be recognized contains the target object, use the frame image where the area to be recognized with the recognized target object is located as the target image;

[0022] Step S32, determine the target position of the target object in the target image;

[0023] Step S33, determine the offset value of the target object in the target image according to the target position;

[0024] Step S34: Determine the position of the area to be recognized in the next frame of image according to the offset value.

[0025] Preferably, the preset quantity is nine.

[0026] The second aspect of the present invention is to provide a system for improving the recognition distance of an artificial intelligence image recognition system, which is used to implement the method for improving the recognition distance of an artificial intelligence image recognition system as described above, and includes:

[0027] An area division module, which is used to respectively divide each frame of the captured consecutive frame images into a preset number of image block areas corresponding to each frame of image;

[0028] An area selection module, connected to the area division module, which is used to respectively select some image block areas from each frame of the consecutive frame images as the areas to be recognized, and input the selected areas to be recognized into the artificial intelligence image recognition system to obtain a recognition result;

[0029] A target dynamic tracking module, connected to the area selection module, which is used to determine the target position of the area to be recognized in the corresponding image frame when the recognition result detects that the area to be recognized contains a target object, and dynamically update the area to be recognized according to the target position in subsequent frames.

[0030] The advantages or beneficial effects of the technical solution of the present invention are as follows:

[0031] By respectively dividing each frame of the consecutive frame images, selectively recognizing some image block areas, and dynamically updating the area to be recognized after detecting the target object, the present invention significantly improves the recognition distance of the AI recognition system, enhances the recognition efficiency and accuracy of the system without increasing the hardware cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic flowchart of a method for improving the recognition distance of an artificial intelligence image recognition system based on a camera in a preferred embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of the overall system composition, processing flow and effect in a preferred embodiment of the present invention;

[0034] Figure 3 It is a schematic flowchart of step S1 in a preferred embodiment of the present invention;

[0035] Figure 4 It is a schematic diagram of the selection of image block areas in a preferred embodiment of the present invention;

[0036] Figure 5In a preferred embodiment of the present invention, it is a schematic flowchart of step S3;

[0037] Figure 6 In a preferred embodiment of the present invention, it is a schematic flowchart of dynamically locking the area to be recognized;

[0038] Figure 7 In a preferred embodiment of the present invention, it is a structural block diagram of the system. Detailed implementation manners

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0041] Next, the present invention will be further described in conjunction with the accompanying drawings and specific embodiments, but it is not a limitation of the present invention.

[0042] Embodiment 1

[0043] In a preferred embodiment of the present invention, in view of the above problems existing in the prior art, a method for improving the recognition distance of an artificial intelligence image recognition system based on a camera is provided. As Figure 1 shown, it includes:

[0044] Step S1: Perform region division on each frame image in the captured continuous frame images to obtain a preset number of image block regions corresponding to each frame image;

[0045] Step S2: Select some image block regions from each frame image in the continuous frame images as the regions to be recognized, and input the selected regions to be recognized into the artificial intelligence image recognition system to obtain a recognition result;

[0046] Step S3: When the recognition result detects that the region to be recognized contains the target object, determine the target position of the region to be recognized in the corresponding image frame, and dynamically update the region to be recognized according to the target position in the subsequent frames.

[0047] Specifically, the embodiments of the present invention aim to overcome the problem of insufficient recognition distance in existing AI image recognition systems, and propose an enhanced method for AI image recognition based on a camera, especially applied to an AI image recognition system in a limited computing power device such as an Android TV, enabling the AI image recognition system to significantly improve the recognition distance without hardware upgrade and maintaining the integrity of the recognition range and functions.

[0048] As Figure 2 shows the overall system composition, processing flow and effect schematic diagram of the present invention. Specifically, the camera captures and generates continuous image frames. The present invention separately performs block processing on each frame of the continuous frame images captured by the camera to divide each frame of image into multiple image block regions; then, for each frame of image, some of the image block regions are respectively selected as the regions to be recognized, and the regions to be recognized selected from each frame of image are sequentially sent into the artificial intelligence image recognition system. The artificial intelligence image recognition system separately performs target recognition on each region to be recognized and outputs the recognition result; when it is detected that a certain region contains a target object, that region is locked, the position of that region is recorded, and the region is dynamically adjusted and preferentially detected in subsequent frames to maintain the real-time and accuracy of recognition.

[0049] Figure 2 In it, the farthest recognition distance of the existing AI image recognition system is L1, and the farthest recognition distance of the AI image recognition system processed by the method shown in the embodiments of the present invention is L1', and L1' is much greater than L1.

[0050] The method of the embodiments of the present invention significantly improves the recognition distance of the AI image recognition system without changing the hardware and the AI model by separately performing target recognition on some of the image block regions after partitioning and combining dynamic locking of the target region, without increasing the hardware cost.

[0051] As a preferred embodiment, wherein, as Figure 3 shown, step S1 includes:

[0052] Step S11, obtaining continuous frame images captured by the camera, and the continuous frame images include multiple frames of images;

[0053] Step S12, for each frame of the continuous frame images, performing equal-region partitioning in the horizontal and vertical directions of the image to obtain a preset number of image block regions.

[0054] In this embodiment, the continuous frame images are composed of multiple frames of images, and each frame of image represents the scene picture captured by the camera at a certain moment.

[0055] Specifically, the image is divided both horizontally and vertically, and each frame of the image is segmented into a preset number of image block regions to facilitate subsequent independent processing of each region.

[0056] As a preferred embodiment, the preset number is 9.

[0057] Specifically, in this embodiment, after each frame of the image is block-processed, each frame of the image is divided into 9 image block regions, ensuring that each image block region can cover a part of the scene of the complete image, making the target recognition process relatively simple.

[0058] In the embodiment of the present invention, taking each frame of the image being divided into 9 image block regions as an example, but not limited thereto, the division method and the division number can be selected according to needs.

[0059] In this embodiment, the divided image block regions may not overlap with each other, or may partially overlap. Non-overlapping means that each region is independent and has no part shared with other regions.

[0060] Furthermore, each image block region partially overlaps with one or several other image block regions, increasing the accuracy of target object detection.

[0061] As a preferred embodiment, the area of the image block region accounts for one-fourth of the area of the frame image where it is located.

[0062] Specifically, in this embodiment, the area of each image block region accounts for 1 / 4 of the area of the frame image where it is located, ensuring that each region has sufficient information for recognition, while avoiding an overly heavy processing burden caused by an overly large region, and further optimizing the processing efficiency.

[0063] As a preferred embodiment, the region to be recognized is one of the image block regions in each frame of the image, and before the target object is recognized, the positions of the image block regions selected in different frames of the image are different.

[0064] Specifically, in this embodiment, during the process of recognizing the target object, the system does not recognize all regions of each frame of the image, but selects one of the image block regions as the region to be recognized. And before the target object is recognized, the positions of the image block regions selected in different frames of the image are different. This selection method can further reduce the processing amount and improve the recognition efficiency.

[0065] As a preferred embodiment, in step S2, a part of the image block regions are respectively selected from each frame of the continuous frame images according to a preset rule, where the preset rule is the selection order of the positions of the regions to be recognized in a continuous preset number of frames of images.

[0066] As a preferred embodiment, in step S2, all the image block regions selected from a continuous preset number of frames in the continuous frame images cover all regions of the complete image.

[0067] Specifically, in this embodiment, different image block regions are selected in a specific order from different frame images.

[0068] By way of example and not limitation, as Figure 4 shows a schematic diagram of the selection of image block regions, demonstrating the fixed selection regions and sampling order for each frame in the continuous frame images generated by the camera.

[0069] In consecutive frames, in the Figure 4 shown order, one region is sequentially selected and sent to the AI image recognition system for recognition, while the remaining regions are not recognized. For example, the central region is selected in the first frame, the middle-left region in the second frame, the middle-right region in the third frame, the middle-lower region in the fourth frame, the lower-left region in the fifth frame, the lower-right region in the sixth frame, the upper-middle region in the seventh frame, the upper-left region in the eighth frame, and the upper-right region in the ninth frame, ensuring that all the image block regions selected from the continuous preset number (i.e., 9 frames) of images can cover all regions of the complete image, thereby increasing the probability of accurate recognition of the target object.

[0070] The subsequent continuous frame block sampling method follows by analogy. This block sampling method significantly reduces the processing amount of each frame image under limited computing power conditions, ensuring the efficiency of real-time detection.

[0071] Due to the fixed sampling order, the maximum detection delay occurs at the position where the image frame where the target object to be detected is located is detected with a maximum interval of 9 frames from the current frame where the detected region is located. In this embodiment, assuming that the interval time between each frame image is 33.33 milliseconds (i.e., the frame rate is approximately 30 frames per second), the maximum detection delay is only 8 * 33.33 ≈ 266.67 milliseconds. This delay level meets the requirements of most real-time detection applications.

[0072] It should be noted that the AI image recognition system is a general AI image recognition model.

[0073] As a preferred embodiment, step S3 further includes:

[0074] When the recognition result detects that the region to be recognized does not contain the target object, return to step S2, continue to select some image block regions from the next frame image as the region to be recognized, and input them into the artificial intelligence image recognition system for target recognition until the target object is recognized.

[0075] Specifically, before the AI image recognition system recognizes the target object, in consecutive frame images, every 9 consecutive frames form a cycle. An image block region is sequentially selected from each complete frame image according to a preset and specific order and sent into the model for detection. For example, the image block region at the center selected from the first frame is sent into the model for detection. If the target object is detected, the position of this region is locked, and this region is dynamically adjusted and preferentially detected in subsequent frames to improve the recognition efficiency and accuracy.

[0076] If the target object is not detected in the central region of the first frame image, the second frame image is continued to be processed. At this time, the image block region in the upper left is selected and also sent into the model for detection. And so on until the target object is detected.

[0077] As a preferred embodiment, as Figure 5 shown, step S3 includes:

[0078] Step S31, when the recognition result detects that the region to be recognized contains the target object, the frame image where the region to be recognized containing the target object is located is used as the target image;

[0079] Step S32, determine the target position of the target object in the target image;

[0080] Step S33, determine the offset value of the target object in the target image according to the target position;

[0081] Step S34, determine the position of the region to be recognized in the next frame image according to the offset value.

[0082] Specifically, as Figure 6 shown, during the process of the AI image recognition system analyzing consecutive frame images frame by frame, once the recognition result shows that a certain region to be recognized contains the target object, the system immediately marks the image frame corresponding to this region to be recognized as the target image.

[0083] Then, determine the specific position of the target object in the target image. Usually, it is achieved by parsing the coordinate information in the recognition result, and the coordinate information indicates the position of the target object in the target image.

[0084] Next, the offset value of the target object in the target image can be determined according to the coordinates. The offset value is the offset amount of the target object relative to the center of the target image or a preset point. This offset value reflects the relative position change of the target object in the image. By calculating the offset value, the possible position of the target object in the next frame image can be predicted more accurately, so as to dynamically adjust the position of the region to be recognized in the next frame image.

[0085] When the target object reaches the edge of the current area to be recognized, the area to be recognized is moved accordingly according to the offset value to ensure that the target object can continue to appear in the new area to be recognized, and the recognition and locking of the target are achieved based on real-time image analysis.

[0086] In addition, after the target object is recognized, the new area to be recognized supports selecting different area division methods according to actual needs, such as Figure 4 the division method shown in

[0087] or other division methods (such as directly dividing into 4 equal areas, etc.) to meet the recognition needs in different scenarios.

[0088] Figure 2 The box where the hand is located in front of the AI image recognition system in

[0089] Embodiment 2

[0090] The present invention also provides a system for improving the recognition distance of an artificial intelligence image recognition system, which is used to implement the method for improving the recognition distance of an artificial intelligence image recognition system as described above, as Figure 7 shown, including:

[0091] An area division module 1, which is used to respectively divide each frame of the captured continuous frame images to obtain a preset number of image block areas corresponding to each frame of image;

[0092] An area selection module 2, connected to the area division module 1, which is used to respectively select some image block areas from each frame of the continuous frame images as the areas to be recognized, and input the selected areas to be recognized into the artificial intelligence image recognition system to obtain a recognition result;

[0093] A target dynamic tracking module 3, connected to the area selection module 2, which is used to determine the target position of the area to be recognized in the corresponding image frame when the recognition result detects that the area to be recognized contains a target object, and dynamically update the area to be recognized according to the target position in subsequent frames.

[0094] Specifically, in this embodiment, the present invention combines the sampling method of only taking a part of each frame of the continuous frame images in a fixed order with the dynamic locking of the target area innovatively, and without increasing the hardware cost, significantly improves the recognition distance of the AI recognition system, thus achieving the effect of significantly improving the image recognition distance on devices with limited computing power.

[0095] The present invention sequentially selects image block regions in consecutive frame images, aiming to identify the target object at the fastest speed. Taking the recognition of the hand gesture of a person in front of a TV as an example, when usually watching TV, the person's position is most likely to be directly in front of the TV. Therefore, in the first frame, the position of the middle region of the image is sampled first. Next, according to the probability of the target object appearing, the left and right positions are sampled in the second and third frames. This sequential sampling method ensures that the target position can be covered as early as possible, improving the recognition speed.

[0096] If the target cannot be detected at the first three positions, the system will sequentially sample the remaining regions. The sampling order can be adjusted according to actual needs, and even all four blocks can be selected for detection in some application scenarios. Through this sequential sampling method, the target object can be recognized in the shortest time.

[0097] Since the frame images of the camera are continuously generated, taking the Sony TV M6L chip as an example, the time for each frame image of the AI image recognition system is about 20 - 50 milliseconds, which is close to the frame rate of the hardware camera generating images (30 frames per second). If each frame of the image is divided into 4 blocks for processing and an independent thread is started for each block (for example, one working thread for each block), then the CPU / GPU usage rate of the device will increase to 4 times, bringing a huge computational pressure. On devices with limited computing power (such as smart TVs), this will significantly affect the original functions of the device (such as high-definition video playback) and increase power consumption.

[0098] In this case, since the frame generation speed of the image is synchronized with the recognition time, the delay is controlled within a reasonable range.

[0099] If each frame of the image is divided into multiple regions and each region is sent to the AI image recognition model for recognition simultaneously, it will cause a significant recognition delay without causing an increase in computational pressure and power consumption. For example, if each frame of the image is divided into 4 block regions and each block region requires about 20 - 50 milliseconds (already a very fast recognition speed on devices with limited computing power, such as smart TVs) for recognition time, then each frame of the image will generate an additional 3 times of recognition, that is, about 100 milliseconds of additional delay. Assuming that the camera generates images at a speed of 30 frames per second, then for processing 1 second of images, that is, 30 frames, there will be 100 * 30 = 3000 milliseconds, that is, 3 seconds of additional delay, and this delay will accumulate. When processing the second frame of the image, there will be an additional delay of about 3 seconds, and the delay will continue to accumulate. This is unacceptable for real-time recognition tasks (such as gesture recognition).

[0100] On the contrary, using the method of the present invention, consecutive image blocks are processed frame by frame in sequence, and each frame of the image is only processed in a low-power working thread, ensuring that the delay will not accumulate. In this way, the entire image recognition process can be carried out efficiently, ensuring real-time performance and response speed.

[0101] Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium as described above (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions for causing a terminal device to execute the methods described in various embodiments of the present invention.

[0102] Exemplarily, the terminal may include, but is not limited to, electronic devices such as smart phones, desktop computers, tablet computers, laptop computers, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, smart wearable devices, etc. Optionally, the operating systems running on the electronic devices may include, but are not limited to, Android system, IOS system, Linux system, Windows system, Unix system, etc.

[0103] The computer instructions can be executed entirely on the user's local computer, partially on the user's local computer, as a separate software package, partially on the user's local computer and partially on a remote computer, or entirely on a remote computer or server. It should also be noted that in some alternative embodiments, the functions noted in each step in the flowchart or each block in the block diagram may not occur in the order noted in the figure. For example, depending on the functions involved, two consecutive steps or two blocks shown may actually be executed substantially simultaneously, or these blocks may sometimes be executed in the reverse order.

[0104] The advantages or beneficial effects of adopting the above technical solution are as follows: By respectively dividing regions for each frame image in a sequence of frame images, selectively identifying some image block regions, and dynamically updating the regions to be identified after detecting the target object, the present invention significantly improves the recognition distance of the AI recognition system, enhances the recognition efficiency and accuracy of the system without increasing the hardware cost.

[0105] The above are only preferred embodiments of the present invention, and do not limit the implementation and protection scope of the present invention accordingly. For those skilled in the art, it should be able to realize that all equivalent replacements and obvious changes made by using the content of this specification and the drawings should be included in the protection scope of the present invention.

Claims

1. A method for improving the recognition distance of an artificial intelligence image recognition system, characterized in that: include: Step S1, performing region division on each frame of the captured continuous frame images to obtain a preset number of image block regions corresponding to each frame of the image; Step S2, selecting a portion of the image block area from each of the continuous frame images as the area to be recognized, and inputting the selected area to be recognized into the artificial intelligence image recognition system to obtain a recognition result; Step S3, when the recognition result detects that the area to be recognized contains the target object, determines the target position of the area to be recognized in the corresponding image frame, and dynamically updates the area to be recognized according to the target position in subsequent frames.

2. The method according to claim 1, characterized in that: The step S1 comprises: Step S11, obtaining continuous frame images captured by a camera, wherein the continuous frame images include multiple frame images; Step S12: for each frame image in the continuous frame images, equally divide the image into regions in the horizontal and vertical directions to obtain a preset number of image block regions.

3. The method according to claim 1, characterized in that The area of ​​the image block region occupies one quarter of the frame image.

4. The method according to claim 1, characterized in that The area to be identified is an image block area in each frame of image, and before the target object is identified, the positions of the image block areas selected in different frames of image are different.

5. The method according to claim 1, characterized in that: In the step S2, partial image block areas are selected from each frame of the continuous frame images as the area to be identified according to a preset rule, wherein the preset rule is the position selection order of the area to be identified in a preset number of consecutive frame images.

6. The method according to claim 1, characterized in that In the step S2, all the image block areas selected from a preset number of consecutive frame images in the consecutive frame images cover all areas of the complete image.

7. The method according to claim 1, characterized in that The step S3 further comprises: When the recognition result detects that the area to be recognized does not contain the target object, return to step S2, continue to select part of the image block area from the next frame image as the area to be recognized, and input it into the artificial intelligence image recognition system for target recognition until the target object is recognized.

8. The method according to claim 1, characterized in that The step S3 comprises: Step S31, when the recognition result detects that the to-be-recognized area contains the target object, the frame image of the to-be-recognized area where the target object is recognized is used as the target image; Step S32, determining the target position of the target object in the target image; Step S33, determining an offset value of the target object in the target image according to the target position; Step S34, determining the position of the area to be identified in the next frame of image according to the offset value.

9. The method according to claim 1, characterized in that: The preset number is 9.

10. A system for improving the recognition distance of an artificial intelligence image recognition system, characterized in that: The method for improving the recognition distance of an artificial intelligence image recognition system according to any one of claims 1 to 9 comprises: A region division module is used to divide each frame of the captured continuous frame images into regions to obtain a preset number of image block regions corresponding to each frame of the image; A region selection module, connected to the region division module, for selecting a portion of the image block area from each of the continuous frame images as the region to be identified, and inputting the selected region to be identified into the artificial intelligence image recognition system to obtain a recognition result; The target dynamic tracking module is connected to the area selection module and is used to determine the target position of the area to be identified in the corresponding image frame when the recognition result detects that the area to be identified contains the target object, and dynamically update the area to be identified according to the target position in subsequent frames.

Images