Double-recording video similar partition intelligent detection method and system

Abstract

The invention discloses a double-recording video similar partition intelligent detection method and system, and relates to the technical field of double-recording video intelligent analysis. GOP structures of a video to be detected are analyzed, and I frames are extracted from each GOP to serve as key frames; the method comprises the following steps: performing intelligent partition detection on a to-be-detected video frame, performing adaptive straight line detection, dynamic rectangle generation and precise sub-graph segmentation to obtain a plurality of to-be-detected partitions, then classifying main partitions, and performing main and auxiliary partition matching by adopting different methods according to partition types. According to the method, methods such as image processing, a straight line detection technology, character erasing and key point matching are combined, and the similar main and auxiliary partitions in the double-recording video are efficiently and accurately detected.

Description

Technical Field

[0001] This invention relates to the field of intelligent analysis technology for dual-recorded videos, and in particular to an intelligent detection method and system for similar partitions in dual-recorded videos. Background Technology

[0002] In industries such as finance and insurance, "dual recording" (audio and video recording) has become a key means of ensuring transparency and compliance in business operations. However, during dual recording, the screen is usually divided into two independent display areas: a main area and a secondary area. Existing quality inspection and auditing systems struggle to efficiently and accurately identify similar main and secondary areas when processing massive amounts of audio and video data. This is especially true when the main area contains dynamic TTS subtitles and compliance prompt borders, which severely impact the accuracy of feature extraction, leading to a high false positive rate for traditional template matching algorithms.

[0003] Therefore, there is an urgent need to develop a technical solution that can effectively address these problems. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for intelligent detection of similar video regions in dual-recording videos. The technical solution adopted is as follows: A method for intelligent detection of similar regions in dual-recorded videos includes the following steps: Step 1: parse the GOP structure of the video to be detected and extract I-frames from each GOP as keyframes. Step 2: Perform intelligent partition detection on the video to be detected. Intelligent partition detection includes adaptive line detection, dynamic rectangle generation, and precise sub-graph segmentation to obtain multiple partitions to be detected. Step 3: Classify the video to be detected into primary partitions. Select the partition with the largest area or the central area as the primary partition, and classify the primary partition based on the three-level classification model. Step 4: Perform primary and secondary partition matching based on TTS broadcast processing, display processing, and normal processing, and output the primary and secondary partition matching results.

[0005] By adopting the above technical solutions and combining image processing, line detection technology, text erasure and key point matching, accurate detection of the main and secondary partitions in dual-recorded videos is achieved, which is suitable for diverse dual-recording scenarios of financial institutions (such as in-person / remote / mobile), significantly reducing the missed detection rate, and is especially suitable for scenarios that need to capture dynamic interactive actions (such as signing process, document presentation).

[0006] It can efficiently and accurately detect similar primary and secondary partitions in dual-recorded videos.

[0007] Optionally, step 1 includes the following sub-steps: Step 11: Parse the GOP structure of the video and extract I-frames as keyframes from each GOP; First, parse the GOP (Group of Pictures) structure of the video, prioritizing the extraction of I-frames as keyframes from each GOP. I-frames are independently encoded frames, independent of information from other frames. This process is achieved using FFmpeg or other video processing libraries.

[0008] Optionally, the adaptive line detection in step 2 includes the following sub-steps: Step 211: Convert the color image to grayscale and apply an adaptive Gaussian filter to smooth the image; Step 212: Enhance image edge information by combining the Canny edge detector and the Sobel operator. For the Sobel operator, calculate the gradient in the x and y directions respectively. Step 213: Detect all straight lines in the binary image based on the Hough transform; Step 214: Pair up all detected lines and find their intersection points; Step 215: Perform cluster analysis on all intersections of straight lines, and select the set parameters to identify the true partition boundaries.

[0009] By adopting the above technical solution, the influence of noise can be reduced, and straight lines in the image can be effectively detected, thereby extracting the partition edges in the image.

[0010] Optionally, the dynamic rectangle generation in step 2 includes the following sub-steps: Step 221: Generate quadrilateral candidates by combining the intersection points of straight lines; Step 222: Select the partitions that meet the conditions based on the set area threshold as the final candidate partitions.

[0011] By adopting the above technical solution, which combines quadrilateral analysis with area parameter screening, the accuracy of rectangular partition detection is not only improved, but also the algorithm's adaptability to complex backgrounds is enhanced.

[0012] Optionally, in step 2, precise sub-image segmentation involves performing perspective correction and boundary expansion compensation on each candidate partition to obtain multiple partitions to be detected.

[0013] By adopting the above technical solution, candidate partitions can be accurately obtained and the integrity of the candidate partitions can be guaranteed.

[0014] Optionally, in step 3, the three-level classification model classifies the primary partition using the following sub-steps: Step 321: The L1 layer of the three-level classification model performs OCR text density analysis on the main partition and determines whether it is a TTS category based on whether the coverage of the recognized text exceeds the set text density threshold. Step 322: The L2 layer of the three-level classification model performs HSV color clustering on the main partition, executes the k-means clustering algorithm in the HSV color space, pays special attention to the saturation S channel, calculates the average saturation of each cluster, and determines whether it belongs to the display category based on the threshold. Step 323: The L3 layer of the three-level classification model uses a pre-trained ResNet-18 model to perform deep feature classification on the main partition. The ResNet-18 outputs the probability distribution of each class, and the class with the highest probability is selected as the final classification result of the main partition.

[0015] By adopting the above technical solutions, from text density and color features to deep learning models, efficient and accurate classification of each main region in an image is achieved.

[0016] Optionally, in step 321, the text coverage rate is the sum of the areas of all text regions divided by the area of ​​the main partition.

[0017] Optionally, in step 4, the TTS broadcast processing includes the following sub-steps: Step 411: Text region detection. Use OCR technology to detect and record all text regions in the image; Step 412: Text region erasure, using the DeepEraser model to repair the image region containing TTS subtitles; Step 413: Using SIFT feature matching, extract SIFT feature points from the repaired main partition image and all sub-partition images, and use Flann for fast nearest neighbor search; Display processing includes the following sub-steps: Step 421: CIELAB k-means color clustering, converting the image from RGB to CIELAB color space; Step 422: Execute the k-means clustering algorithm to separate the main color regions; Step 423: Morphological operations, apply erosion to remove small noise and separate the prompt box; Step 424: ORB pyramid cross-scale matching, construct ORB feature pyramids for the main partition image and all sub-partition images, and perform feature matching at different scales; The processing of ordinary classes includes the following sub-steps: Step 431: Enhance histogram matching by converting the image from the RGB color space to the HSV color space and performing histogram equalization on the luminance channel to enhance contrast. Step 432: Calculate the RGB 3D joint histogram; Step 433: Calculate the final similarity score based on the comprehensive similarity index; Step 434: Calculate the similarity between the primary partition image and all secondary partition images, and select the pair with the highest similarity as the matching primary and secondary partitions.

[0018] By adopting the above technical solutions, customized processing and matching strategies are applied to different types of partitions, thereby significantly improving the accuracy, robustness, and practicality of partition image analysis and matching.

[0019] A dual-recording video similarity partitioning intelligent detection system includes a memory and a processor. The memory stores the video to be detected and a detection program designed using a dual-recording video similarity partitioning intelligent detection method. The processor is communicatively connected to the memory, inputs the video to be detected into the detection program, runs the detection program, and outputs the detection results.

[0020] Optionally, a display is also included, which is communicatively connected to the processor, and the processor controls the display to show the detection results.

[0021] In summary, the present invention has at least the following beneficial technical effects: This invention provides an intelligent method and system for detecting similar partitions in dual-recorded videos. Combining image processing, line detection technology, text erasure, and key point matching, it efficiently and accurately detects similar primary and secondary partitions in dual-recorded videos. It achieves automatic partition recognition and similarity comparison of primary and secondary video frames in dual-recorded videos, reducing the workload of manual review and significantly improving the efficiency and accuracy of manual review. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the intelligent detection method for similar partitions in dual-recorded video according to the present invention. Figure 2 This is a frame example of the effect of a TTS broadcast class in a specific embodiment of the present invention; Figure 3 These are example frame diagrams illustrating specific embodiments of the present invention. Figure 4 This is a frame example effect diagram of a general class in a specific embodiment of the present invention; Figure 5 These are effect diagrams of multiple partitions to be detected according to specific embodiments of the present invention; Figure 6 This is a specific embodiment of the invention demonstrating whether a device is classified as a TTS (Text-to-Speech) device. Figure 7 This is a specific embodiment of the invention used to determine whether a display category effect image is included; Figure 8 This is a diagram showing the final classification result of the main partition in a specific embodiment of the present invention; Figure 9 This is a diagram showing the effect of SIFT feature matching in a specific embodiment of the present invention; Figure 10 This is a diagram illustrating the cross-scale matching effect of the ORB pyramid in a specific embodiment of the present invention. Figure 11 This is a diagram showing the matching effect of primary and secondary partitions in a specific embodiment of the present invention. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings.

[0024] This invention discloses a method and system for intelligent detection of similar partitions in dual-recorded videos.

[0025] Reference Figures 1-11 Example 1: A method for intelligent detection of similar regions in dual-recorded videos, comprising the following steps: Step 1: parse the GOP structure of the video to be detected and extract I-frames from each GOP as keyframes. Step 2: Perform intelligent partition detection on the video to be detected. Intelligent partition detection includes adaptive line detection, dynamic rectangle generation, and precise sub-image segmentation to obtain multiple partitions to be detected. Step 3: Classify the video to be detected into primary partitions. Select the partition with the largest area or the central area as the primary partition, and classify the primary partition based on the three-level classification model. Step 4: Perform primary and secondary partition matching based on TTS broadcast processing, display processing, and normal processing, and output the primary and secondary partition matching results.

[0026] By combining image processing, line detection technology, text erasure, and key point matching, this method efficiently and accurately detects similar primary and secondary video segments in dual-recorded videos. It achieves automatic segmentation recognition and similarity comparison of primary and secondary video frames in dual-recorded videos, reducing the workload of manual review and significantly improving the efficiency and accuracy of manual review.

[0027] Example 2, step 1 includes the following sub-steps: Step 11: Parse the GOP structure of the video and extract I-frames from each GOP as keyframes; I-frame extraction prioritizes the extraction of keyframes. First, the video's Group of Pictures (GOP) structure is parsed, and I-frames are extracted from each GOP as keyframes. I-frames are independently encoded frames, independent of information from other frames. This process is achieved using FFmpeg or other video processing libraries.

[0028] Example 3, step 2 of adaptive line detection includes the following sub-steps: Step 211: Convert the color image to grayscale and apply an adaptive Gaussian filter to smooth the image; Step 212: Enhance image edge information by combining the Canny edge detector and the Sobel operator. For the Sobel operator, calculate the gradient in the x and y directions respectively. Step 213: Detect all straight lines in the binary image based on the Hough transform; Step 214: Pair up all detected lines and find their intersection points; Step 215: Perform cluster analysis on all intersections of straight lines, and select the set parameters to identify the true partition boundaries.

[0029] It can reduce the impact of noise, effectively detect straight lines in images, and thus extract the partition edges in images.

[0030] Example 4, step 2 of dynamic rectangle generation includes the following sub-steps: Step 221: Generate quadrilateral candidates by combining the intersection points of straight lines; Step 222: Select the partitions that meet the conditions based on the set area threshold as the final candidate partitions.

[0031] By combining quadrilateral analysis with area parameter selection, the accuracy of rectangular partition detection is improved, and the algorithm's adaptability to complex backgrounds is enhanced.

[0032] In Example 5, the precise sub-image segmentation in step 2 involves performing perspective correction and boundary expansion compensation on each candidate partition to obtain multiple partitions to be detected.

[0033] It can accurately obtain candidate partitions and ensure the integrity of the candidate partitions.

[0034] In Example 6, step 3, the three-level classification model classifies the primary partition using the following sub-steps: Step 321: The L1 layer of the three-level classification model performs OCR text density analysis on the main partition and determines whether it is a TTS category based on whether the coverage of the recognized text exceeds the set text density threshold. Step 322: The L2 layer of the three-level classification model performs HSV color clustering on the main partition. The k-means clustering algorithm is executed in the HSV color space, paying particular attention to the saturation S channel. The average saturation of each cluster is calculated, and a threshold is used to determine whether it belongs to the display category. Code example is as follows: hsv_image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) hsv_pixels = hsv_image.reshape((-1, 3)) kmeans = KMeans(n_clusters=3).fit(hsv_pixels) labels = kmeans.labels_.reshape((image.shape[0], image.shape[1])) for cluster in range(3): saturation_mean = np.mean(hsv_pixels[labels.flatten() == cluster, 1]) if saturation_mean > threshold: category = 'Display Category'; Step 323: The L3 layer of the three-level classification model uses a pre-trained ResNet-18 model for deep feature classification of the principal partition. The ResNet-18 model outputs the probability distribution of each class, and the class with the highest probability is selected as the final classification result for the principal partition. Code example is as follows: import torch from torchvision.models import resnet18 model = resnet18(pretrained=True) model.fc = torch.nn.Linear(model.fc.in_features, 3) # Modify the fully connected layer to adapt to the new classification task model.eval() with torch.no_grad(): outputs = model(input_tensor) probabilities = torch.softmax(outputs, dim=1) predicted_class = torch.argmax(probabilities, dim=1).item() if predicted_class == 0: category = 'General Class' elif predicted_class == 1: category = 'Display Category' elif predicted_class == 2: category = 'TTS broadcasting'; In Example 7, step 321, the text coverage rate is the sum of the areas of all text regions divided by the area of ​​the main partition.

[0035] From text density and color features to deep learning models, efficient and accurate classification of each main region in an image has been achieved.

[0036] Example 8, step 4, the TTS broadcast processing includes the following sub-steps: Step 411: Text region detection. Use OCR technology to detect and record all text regions in the image; Step 412: Text region erasure, using the DeepEraser model to repair the image region containing TTS subtitles; Step 413: Using SIFT feature matching, extract SIFT feature points from the repaired main partition image and all sub-partition images, and use Flann for fast nearest neighbor search; The code example is as follows: sift = cv2.SIFT_create() kp1, des1 = sift.detectAndCompute(main_image, None) kp2, des2 = sift.detectAndCompute(reference_image, None) flann=cv2.FlannBasedMatcher(dict(algorithm=1, trees=5), dict(checks=50)) matches = flann.knnMatch(des1, des2, k=2); Display processing includes the following sub-steps: Step 421: CIELAB k-means color clustering, converting the image from RGB to CIELAB color space; Step 422: Execute the k-means clustering algorithm to separate the main color regions; code example is as follows: lab_pixels = lab_image.reshape((-1, 3)) kmeans = KMeans(n_clusters=3).fit(lab_pixels) labels=kmeans.labels_.reshape((image.shape[0], image.shape[1])); Step 423: Morphological operations, apply erosion to remove small noise and separate the prompt box; The code example is as follows: orb = cv2.ORB_create() kp1, des1 = orb.detectAndCompute(eroded_mask, None) kp2, des2 = orb.detectAndCompute(reference_image, None) bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True) matches = bf.match(des1, des2) matches = sorted(matches, key=lambda x: x.distance); Step 424: ORB pyramid cross-scale matching, construct ORB feature pyramids for the main partition image and all sub-partition images, and perform feature matching at different scales; The processing of ordinary classes includes the following sub-steps: Step 431: Enhance histogram matching by converting the image from the RGB color space to the HSV color space and performing histogram equalization on the luminance channel to enhance contrast. Step 432: Calculate the RGB 3D joint histogram; Step 433: Calculate the final similarity score based on the comprehensive similarity index; Step 434: Calculate the similarity between the primary partition image and all secondary partition images, and select the pair with the highest similarity as the matching primary and secondary partitions.

[0037] Customized processing and matching strategies are applied to different categories of partitions, thereby significantly improving the accuracy, robustness, and practicality of partition image analysis and matching.

[0038] Example 9: A dual-recording video similarity partitioning intelligent detection system includes a memory and a processor. The memory stores the video to be detected and a detection program designed using a dual-recording video similarity partitioning intelligent detection method. The processor is communicatively connected to the memory, inputs the video to be detected into the detection program, and runs the detection program to output the detection results.

[0039] Example 10 also includes a display, which is communicatively connected to the processor, and the processor controls the display to show the detection results.

[0040] The following specific embodiments illustrate the implementation principle of the intelligent detection method and system for similar partitions in dual-recording video according to the present invention: 1. Extract video frames using ffmpeg: ffmpeg -i input_video.mp4 -vf "select=eq(pict_type\,I)" -vsync vfrextracted_i_frame_%04d.jpg See the example frame for TTS broadcast. Figure 2 See the example frame for the display class. Figure 3 See the example of a regular frame. Figure 4 ; 2. Convert the color image to grayscale and apply an adaptive Gaussian filter to smooth the image; Step 212: Enhance image edge information by combining the Canny edge detector and the Sobel operator. For the Sobel operator, calculate the gradient in the x and y directions respectively. Step 213: In the probabilistic Hough transform, voting is performed based on the gradient direction of the edge points; Step 214: All detected straight lines are converted to standard form and normalized, and intersections less than 5 pixels from the image boundary are removed; Step 215: Perform cluster analysis on all intersections of straight lines, and select the set parameters to identify the true partition boundaries.

[0041] It can reduce the impact of noise, effectively detect straight lines in images, and thus extract the partition edges in images.

[0042] Example 4, step 2 of dynamic rectangle generation includes the following sub-steps: Step 221: Generate quadrilateral candidates by combining the intersection points of straight lines; Step 222: Select the partitions that meet the conditions based on the set area threshold as the final candidate partitions.

[0043] By combining quadrilateral analysis with area parameter selection, the accuracy of rectangular partition detection is improved, and the algorithm's adaptability to complex backgrounds is enhanced.

[0044] In Example 5, the precise sub-image segmentation in step 2 involves performing perspective correction and boundary expansion compensation on each candidate partition to obtain multiple partitions to be detected. For example... Figure 5 As shown; 3. The L1 layer of the three-level classification model performs OCR text density analysis on the main partition, and determines whether it belongs to the TTS category based on whether the recognized text coverage exceeds the set text density threshold; for example... Figure 6 As shown; 4. The L2 layer of the three-level classification model performs HSV color clustering on the main partition, executes the k-means clustering algorithm in the HSV color space, pays special attention to the saturation S channel, calculates the average saturation of each cluster, and determines whether it belongs to the display category based on a threshold; for example... Figure 7 As shown; 5. The L3 layer of the three-level classification model uses a pre-trained ResNet-18 model for deep feature classification of the principal partition. ResNet-18 outputs the probability distribution for each class, and the class with the highest probability is selected as the final classification result for the principal partition; for example... Figure 8 As shown; Example 8, step 4, the TTS broadcast processing includes the following sub-steps: Step 411: Text region detection. Use OCR technology to detect and record all text regions in the image; Step 412: Text region erasure, using the DeepEraser model to repair the image region containing TTS subtitles; Step 413: Using SIFT feature matching, extract SIFT feature points from the repaired main partition image and all sub-partition images, and use Flann for fast nearest neighbor search; such as Figure 9 As shown; Display processing includes the following sub-steps: Step 421: CIELAB k-means color clustering, converting the image from RGB to CIELAB color space; Step 422: Execute the k-means clustering algorithm to separate the main color regions; Step 423: Morphological operations, apply erosion to remove small noise and separate the prompt box; Step 424: ORB pyramid cross-scale matching, constructing ORB feature pyramids for the main partition image and all sub-partition images, and performing feature matching at different scales; such as Figure 10 As shown; The processing of ordinary classes includes the following sub-steps: Step 431: Enhance histogram matching by converting the image from the RGB color space to the HSV color space and performing histogram equalization on the luminance channel to enhance contrast. Step 432: Calculate the RGB 3D joint histogram; Step 433: Calculate the final similarity score based on the comprehensive similarity index; Step 434: Calculate the similarity between the primary partition image and all secondary partition images, and select the pair with the highest similarity as the matching primary and secondary partitions; for example... Figure 11 As shown.

[0045] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent detection of similar regions in dual-recorded videos, characterized in that, Includes the following steps: Step 1: parse the GOP structure of the video to be detected and extract I-frames from each GOP as keyframes. Step 2: Perform intelligent partition detection on the video to be detected. Intelligent partition detection includes adaptive line detection, dynamic rectangle generation, and precise sub-graph segmentation to obtain multiple partitions to be detected. Step 3: Classify the video to be detected into primary partitions. Select the partition with the largest area or the central area as the primary partition, and classify the primary partition based on the three-level classification model. Step 4: Perform primary and secondary partition matching based on TTS broadcast processing, display processing, and normal processing, and output the primary and secondary partition matching results.

2. The intelligent detection method for similar partitions in dual-recording video according to claim 1, characterized in that: Step 2, adaptive line detection, includes the following sub-steps: Step 211: Convert the color image to grayscale and apply an adaptive Gaussian filter to smooth the image; Step 212: Enhance image edge information by combining the Canny edge detector and the Sobel operator. For the Sobel operator, calculate the gradient in the x and y directions respectively. Step 213: Detect all straight lines in the binary image based on the Hough transform; Step 214: Pair up all detected lines and find their intersection points; Step 215: Perform cluster analysis on all intersections of straight lines, and select the set parameters to identify the true partition boundaries.

3. The intelligent detection method for similar partitions in dual-recording videos according to claim 2, characterized in that: Step 2, dynamic rectangle generation, includes the following sub-steps: Step 221: Generate quadrilateral candidates by combining the intersection points of straight lines; Step 222: Select the partitions that meet the conditions based on the set area threshold as the final candidate partitions.

4. The intelligent detection method for similar partitions in dual-recording video according to claim 3, characterized in that: In step 2, precise sub-image segmentation involves performing perspective correction and boundary expansion compensation on each candidate partition to obtain multiple partitions to be detected.

5. The intelligent detection method for similar partitions in dual-recording videos according to claim 1, characterized in that: In step 3, the three-level classification model classifies the primary partition using the following sub-steps: Step 321: The L1 layer of the three-level classification model performs OCR text density analysis on the main partition and determines whether it is a TTS category based on whether the coverage of the recognized text exceeds the set text density threshold. Step 322: The L2 layer of the three-level classification model performs HSV color clustering on the main partition, executes the k-means clustering algorithm in the HSV color space, pays special attention to the saturation S channel, calculates the average saturation of each cluster, and determines whether it belongs to the display category based on the threshold. Step 323: The L3 layer of the three-level classification model uses a pre-trained ResNet-18 model to perform deep feature classification on the main partition. The ResNet-18 outputs the probability distribution of each class, and the class with the highest probability is selected as the final classification result of the main partition.

6. The intelligent detection method for similar partitions in dual-recording video according to claim 5, characterized in that: In step 321, the text coverage rate is the sum of the areas of all text regions divided by the area of ​​the main partition.

7. The intelligent detection method for similar partitions in dual-recording video according to claim 1, characterized in that: Step 4, the TTS broadcast processing includes the following sub-steps: Step 411: Text region detection. Use OCR technology to detect and record all text regions in the image; Step 412: Text region erasure, using the DeepEraser model to repair the image region containing TTS subtitles; Step 413: Using SIFT feature matching, extract SIFT feature points from the repaired main partition image and all sub-partition images, and use Flann for fast nearest neighbor search; Display processing includes the following sub-steps: Step 421: CIELAB k-means color clustering, converting the image from RGB to CIELAB color space; Step 422: Execute the k-means clustering algorithm to separate the main color regions; Step 423: Morphological operations, apply erosion to remove small noise and separate the prompt box; Step 424: ORB pyramid cross-scale matching, construct ORB feature pyramids for the main partition image and all sub-partition images, and perform feature matching at different scales; The processing of ordinary classes includes the following sub-steps: Step 431: Enhance histogram matching by converting the image from the RGB color space to the HSV color space and performing histogram equalization on the luminance channel to enhance contrast. Step 432: Calculate the RGB 3D joint histogram; Step 433: Calculate the final similarity score based on the comprehensive similarity index; Step 434: Calculate the similarity between the primary partition image and all secondary partition images, and select the pair with the highest similarity as the matching primary and secondary partitions.

8. A dual-recording video similarity partitioning intelligent detection system, characterized in that: It includes a memory and a processor. The memory stores the video to be detected and a detection program designed using the dual-recording video similarity partitioning intelligent detection method according to any one of claims 1-7. The processor is communicatively connected to the memory, inputs the video to be detected into the detection program, and runs the detection program to output the detection results.

9. The intelligent detection system for similar partitions in dual-recording video as described in claim 8, characterized in that: It also includes a display, which is communicatively connected to the processor, and the processor controls the display to show the detection results.

Images