AI-based news segmentation method

Abstract

This invention discloses an AI-based statistical method for news segmentation, comprising the following steps: performing face tracking on the news video, recording the positional information of each face, calculating the quality score of each tracked face, and recording the time point, facial features, and bounding box position information of the best face, as well as the duration of each tracked face; analyzing all tracked face results and extracting the presenter's segment; segmenting the video, sorting the presenter's time period list in descending order based on time point information to obtain a second time period list, using the first time point of the second time period list as the entry point for segmentation, and the second time point as the exit point for segmentation, thereby obtaining the time point information for all segments. This invention, by introducing face tracking, presenter segment extraction, and video segmentation, achieves automated segmentation of news videos.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a news segmentation method based on AI statistics. Background Technology

[0002] Existing news video processing technologies require video segmentation for different scenarios and themes to facilitate better content presentation and distribution. However, traditional segmentation methods often require manual intervention, resulting in low efficiency and inconsistent results. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention provides a news segmentation method based on AI statistics, which realizes the automated segmentation of news videos by introducing steps such as face tracking, host segment extraction and video segmentation.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0005] A news segmentation method based on AI statistics includes the following steps:

[0006] Face tracking is performed on news videos, the location information of each face is recorded, the quality score of each tracked face is calculated, and the time point, facial features and face bounding box location information of the best face are recorded, as well as the duration of each tracked face.

[0007] Analyze all tracked facial results and extract the host's segment;

[0008] The video is segmented. The host's time slot list is sorted in descending order based on the time point information to obtain a second time slot list. The first time point of the second time slot list is used as the entry point for segmentation, and the second time point of the second time slot list is used as the exit point for segmentation, thus obtaining the time point information of all segments.

[0009] One possible implementation involves analyzing all tracked facial results and extracting the presenter's segment, including:

[0010] Filter based on tracking duration, excluding segments with a duration less than the first threshold;

[0011] For segments that meet the duration requirement, calculate the area SF of the face bounding box, and calculate the occupancy percentage P by combining SF with the area SV of the video frame. The formula is P = SF / SV. Exclude segments where P is less than the second threshold.

[0012] Face clustering is used to cluster the best facial features in the final filtered segments, and the number of hosts is counted.

[0013] The bubble sort method is used to compare the time periods of the hosts, merge the overlapping time periods, and obtain the final list of host time periods.

[0014] In one possible implementation, the first threshold is greater than 8 seconds.

[0015] In one possible implementation, the second threshold is greater than 1.5%.

[0016] In one possible implementation, the number of hosts is less than or equal to 2.

[0017] In one possible implementation, the face clustering specifically includes:

[0018] Using a face recognition algorithm, extract the feature vector of each detected face;

[0019] For each pair of feature vectors, calculate the similarity between the feature vectors;

[0020] The feature vectors are grouped according to their similarity to form clusters.

[0021] In one possible implementation, the face recognition algorithm is based on a deep learning convolutional neural network.

[0022] In one possible implementation, the similarity between feature vectors is calculated using one of cosine similarity, Euclidean distance, or Manhattan distance as a similarity metric.

[0023] The present invention offers the following advantages: compared to traditional methods, it boasts high efficiency, a high degree of automation, and stable results. Through reasonable screening and clustering methods, presenter segments can be extracted quickly and accurately, providing a reliable foundation for subsequent news video processing and distribution. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the steps of the AI-based news segmentation method according to an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] See Figure 1 The diagram shows a flowchart of a news segmentation method based on AI statistics according to an embodiment of the present invention, including the following steps:

[0027] S10, perform face tracking on news videos, record the location information of each face, calculate the quality score of each tracked face, and record the time point, face features and face bounding box location information of the best face, and record the duration of each tracked face;

[0028] S20: Analyze all tracked facial results and extract the host segment;

[0029] S30, perform video segmentation, sort the host's time segment list in descending order according to the time point information to obtain a second time segment list, take the first time point of the second time segment list as the entry point of the segmentation, and the second time point of the second time segment list as the exit point of the segmentation, and obtain the time point information of all segments.

[0030] In one embodiment of the present invention, in S20, analyzing all tracked face results and extracting the host segment includes:

[0031] S10, filter according to the tracking duration and exclude segments with a duration less than the first threshold; in specific application examples, the first threshold is generally greater than 8 seconds.

[0032] S20. For segments that meet the duration requirement, calculate the area SF of the face bounding box, and calculate the occupancy percentage P by combining SF and the area SV of the video frame. The formula is P = SF / SV. Exclude segments where P is less than the second threshold. In specific application examples, P is generally greater than 1.5%.

[0033] S30: Use face clustering to cluster the best face features in the final filtered segments and count the number of hosts; considering the characteristics of news, in specific application examples, the number of hosts is generally less than or equal to 2.

[0034] S40: Use the bubble sort method to compare the time periods of the hosts, merge the intersecting time periods, and obtain the final list of host time periods.

[0035] In one embodiment of the present invention, in S30, face clustering specifically includes:

[0036] S301 uses a face recognition algorithm to extract the feature vector of each detected face; the face recognition algorithm can be based on a deep learning convolutional neural network.

[0037] S302, For each pair of feature vectors, calculate the similarity between the feature vectors; the similarity between feature vectors can be calculated using one of the following indicators: cosine similarity, Euclidean distance, or Manhattan distance. A smaller distance or a larger similarity indicates that the two facial features are more similar.

[0038] S303. Group the feature vectors according to their similarity to form clusters. Clustering algorithms can include hierarchical clustering, K-means clustering, density-based clustering of applications with noise (DBSCAN), etc., which can assign feature vectors to different clusters based on similarity.

[0039] The AI-based news segmentation method described above incorporates steps such as face tracking, presenter segment extraction, and video segmentation to automate the segmentation of news videos. Compared to traditional methods, it offers advantages such as high efficiency, high automation, and stable results. Through appropriate filtering and clustering methods, presenter segments can be extracted quickly and accurately, providing a reliable foundation for subsequent news video processing and distribution.

[0040] It should be understood that the exemplary embodiments described herein are illustrative and not restrictive. Although one or more embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art will understand that various changes in form and detail may be made without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. An AI statistics-based news disassembly method, characterized in that, The method comprises the following steps: Face tracking is performed on the news video, position information of each face is recorded, a quality score of each tracked face is calculated, and time point, face feature, and face frame position information of an optimal face are recorded, and a time length of each tracked face is recorded; All tracked face results are analyzed to extract a host segment; Video splitting is performed, a first time period list is sorted in descending order according to time point information, a second time period list is obtained, a first time point of the second time period list is used as an entry point of the splitting, a second time point of the second time period list is used as an exit point of the splitting, and time point information of all split segments is obtained; The host segment extraction includes: Segments with a time length less than a first threshold value are filtered out; For segments meeting the time length requirement, an area SF of the face frame is calculated, a percentage P of the area SF to the area SV of the video frame is calculated, and segments with P less than a second threshold value are filtered out; Optimal face features in the finally filtered segments are clustered using face clustering, and the number of hosts is counted; A bubble sort method is used to compare the time of the host time period, and intersected time periods are merged to obtain a final host time period list; The face clustering specifically includes: A face recognition algorithm is used to extract a feature vector of each detected face; For each pair of feature vectors, the similarity between the feature vectors is calculated; The feature vectors are grouped according to the similarity to form a cluster.

2. The AI statistics-based news article disassembling method of claim 1, wherein, The first threshold value is greater than 8 seconds.

3. The AI statistics-based news article disassembling method of claim 1, wherein, The second threshold value is greater than 1.5%. 4.The AI-statistics-based news abstracting method of claim 1, wherein, The number of hosts is less than or equal to 2.

5. The AI statistics-based news article disassembling method of claim 1, wherein, The face recognition algorithm is based on a convolutional neural network of deep learning.

6. The AI statistics-based news article disassembling method of claim 1, wherein, The similarity between the feature vectors is calculated by using one of cosine similarity, Euclidean distance, or Manhattan distance as a similarity index.

Images