CN107301245A - A kind of power information video searching system - Google Patents

Abstract

A power information video search system, including an acquisition device, a processing device, a classification device, and a retrieval device. According to the scope of capture, the web page analysis module obtains links with video playback addresses in each video website; according to the current link, the video address extraction module extracts Get the real download address of the video, and then call the video downloader to download; the video standardization module will convert the downloaded videos into the same format and store them in the video library; the video feature extraction module will analyze the video in the video library and extract the structure Features and key frames constitute structured information and are stored in the video library.

Description

A video search system for electric power information

technical field

The invention relates to a search system. Especially a power information video search system.

Background technique

There are a large number of video resources in the power information system stored on the Internet, including a large number of surveillance videos and fault videos. When the staff of the power system needs to quickly find relevant resources based on a certain picture, they can only rely on text information. Like this often can't find correct video, and even if find related video, how to download effectively and fast also is the problem that will face.

Contents of the invention

In order to overcome the deficiencies of the existing technical solutions, the present invention provides a technical solution of a power information video search system.

In order to achieve the above object, the technical solution of the present invention is: a power information video search system, including an acquisition device, a processing device, a classification device and a retrieval device,

The acquisition device includes a web page analysis module, a video text library, a video address extraction module and a video downloader. The web page analysis module is used to obtain the link of the video playback address and parse out the text information of the video through HTML text. The video text library is used to store the video The text information, the video address extraction module is used to obtain the real download address of the video, and the video downloader downloads the video through the method of video resource segmentation;

The processing device includes a video standardization module, a video library and a video feature extraction module. The video standardization module is used to convert the video downloaded by the video downloader to form a standard video with a unified format. The video feature extraction module structures the standard video Process and obtain the structural features and key frames of the standard video, and the video library is used to store the standard video and its structural features and key frames;

The classification device includes a classification module and a classification video library. The classification module performs cluster analysis according to the information in the video text library and the video library. The classification video library stores the clustered analyzed videos and related information according to the cluster structure, thereby forming a retrieval database ;

The retrieval device includes a retrieval interface and a query module, the user inputs images and retrieval information through the retrieval interface and displays the retrieval results on the retrieval interface to return to the user,

It is characterized by:

According to the scope of crawling, the web page analysis module obtains links with video playback addresses in each video website, analyzes the HTML text corresponding to such links, parses and extracts text information related to the video, and performs Chinese word segmentation on these text information as video content. The title is stored in the video text library;

According to the current link, the video address extraction module extracts the real download address of the video, and then calls the video downloader to download;

The video standardization module converts the downloaded videos into the same format and stores them in the video library;

The video feature extraction module will analyze the videos in the video library, extract structural features and key frames to form structured information, and store them in the video library;

The classification module first performs pre-classification according to the video text library, and then classifies according to the structured information of the video text library and video, establishes a video database and stores it in the classified video library;

The user provides an image example for retrieval through the retrieval interface, and the query module returns the videos matching the retrieval conditions to the user.

Beneficial effect:

(1) Retrieve videos according to pictures, which effectively improves the retrieval efficiency of staff;

(2) Build a retrieval system with comprehensive database information;

(3) Use the video resource segmentation method of dichotomy to optimize the download speed of the video, and the staff can obtain the video in time.

Description of drawings

Fig. 1 is a system block diagram of the present invention.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, a power information video search system includes an acquisition device, a processing device, a classification device and a retrieval device,

The acquisition device includes a web page analysis module, a video text library, a video address extraction module and a video downloader. The web page analysis module is used to obtain the link of the video playback address and parse out the text information of the video through HTML text. The video text library is used to store the video The text information, the video address extraction module is used to obtain the real download address of the video, and the video downloader downloads the video through the method of video resource segmentation;

The processing device includes a video standardization module, a video library and a video feature extraction module. The video standardization module is used to convert the video downloaded by the video downloader to form a standard video with a unified format. The video feature extraction module structures the standard video Process and obtain the structural features and key frames of the standard video, and the video library is used to store the standard video and its structural features and key frames;

The classification device includes a classification module and a classification video library. The classification module performs cluster analysis according to the information in the video text library and the video library. The classification video library stores the clustered analyzed videos and related information according to the cluster structure, thereby forming a retrieval database ;

The retrieval device includes a retrieval interface and a query module, the user inputs images and retrieval information through the retrieval interface and displays the retrieval results on the retrieval interface to return to the user,

It is characterized by:

According to the scope of crawling, the web page analysis module obtains links with video playback addresses in each video website, analyzes the HTML text corresponding to such links, parses and extracts text information related to the video, and performs Chinese word segmentation on these text information as video content. The title is stored in the video text library;

According to the current link, the video address extraction module extracts the real download address of the video, and then calls the video downloader to download;

The video standardization module converts the downloaded videos into the same format and stores them in the video library;

The video feature extraction module will analyze the videos in the video library, extract structural features and key frames to form structured information, and store them in the video library;

The classification module first performs pre-classification according to the video text database, and then classifies according to the structured information of the video text database and video, establishes a video database and stores it in the classified video database;

The user provides an image example for retrieval through the retrieval interface, and the query module returns the videos matching the retrieval conditions to the user.

Among them, the structured processing of the video feature extraction module is specifically:

Step a1, construct the video structure. The video data structure is divided into three levels: scene, shot and frame. A frame is an independent static image. A group of frames constitutes a shot, a group of shots constitutes a scene, and a scene constitutes a video. Video structure The construction process is divided into two steps,

Step a1.1, extracting shots from the video stream,

Perform lens cut detection on the video stream, and find the image frame when the camera switch occurs in the video, specifically:

Divide the image frame into sub-blocks of 8×8 pixels, calculate the average value of each sub-block, and calculate the absolute value of the difference between the average values of the sub-blocks at the same position in two consecutive frames in the video sequence as the inter-frame difference ;

Calculate the ratio of the new edge pixel increase and the edge pixel reduction ratio of two adjacent frame images, and take the maximum value as the ratio difference;

If the inter-frame difference is greater than the preset value, and the ratio difference is greater than the preset value, then it is determined that a lens switch has occurred, and the lens is extracted;

Wherein, the shot is a basic video data unit, and the cut of the shot is divided into two types: sudden change and gradual change. Sudden change means that one shot is directly cut to another shot without any editing techniques, while gradual change means that an editing effect is added between one shot and another to make the change appear more gradual. The process of finding out the image frame when the video lens is switched is called the lens cut detection. Edges are those gray discontinuous points in the gray space of an image.

Step a1.2, extract keyframes from the shot,

A key frame is an image that can describe the main content of a shot. Since the redundant information in the video is eliminated, using a key frame to represent a shot can greatly reduce the workload of building a video index. The method of extracting a key frame is as follows:

When there is a significant change between the current frame of the shot and the latest image judged to be a key frame, the current frame is used as a new reference key frame. First, take the first frame of the shot as a reference key frame, and then use Compare the frame image with the features of this key frame image, if the change is large, use the current frame as a new key frame, and then continue to compare with the subsequent frame image, and so on to obtain key frames one after another;

Step a2, feature extraction, convert the RBG color space of the key frame to the color space HSV mode that is more in line with people's subjective understanding of color, quantize the color space into several color bars, and then automatically divide the image into several color bars with color segmentation technology Regions, each region uses a color bar to form an index, and the description of the image is converted into a color index set.

Among them, the video resource segmentation method of the video downloader uses the dichotomy method to download the video by confirming the IP address and attribution of different fragment resources in the video resource, specifically:

Step b1, determine the playing time of the video to be downloaded, and obtain the starting point and end point The IP address and attribution, if the origin and end point The IP address and the attribution are the same, then confirm the video IP address, attribution, and go to step b4; if the starting point and end point If the IP address and the attribution are not the same, go to step b2;

Step b2, get the middle point of the video clip IP address and attribution, if step b2 is executed 10 times, go to step b4;

Step b3, if the intermediate point and starting point The IP address and the attribution are the same, then confirm the video clip segment IP address, attribution, and Compose a new video clip And go to step b2;

if the middle point and end point The IP address and the attribution are the same, then confirm the video clip segment IP address, attribution, and Compose a new video clip And go to step b2;

if the middle point with the starting point and end point IP address and attribution are not the same, then the Compose a new video clip and go to step b2, then set the Compose a new video clip And go to step b2;

Step b4, recording steps b2-b3 Determine the segmentation position of each video segment in the original video and the corresponding IP address and attribution, and determine the operator of each video segment, and download the video in blocks, thereby optimizing the download speed.

Among them, the cluster analysis method of the classification module performs data mining on the key frames in the video library, automatically clusters the key frames, and uses the combination of video semantic information and visual features of the key frames, specifically:

Step c1, perform pre-classification according to the video text, classify videos with similar text information into one category, and ensure that the main content of the video belongs to one category;

Step c2, on the basis of pre-classification, clustering is carried out according to the color features of the key frames of the video library in each major category, and the key frames with similar color features are aggregated into a sub-category;

Because there is a semantic gap between the low-level features of videos and the high-level features of videos, that is, semantic features, the videos retrieved with a single video content feature are likely to be inconsistent with user expectations. Therefore, a classification method based on the combination of visual features and advanced semantic features of video is proposed, which overcomes the shortcomings of the classification method based solely on visual features. Such a classification method can realize that the image frames contained in a class are similar, and make the distance between classes as large as possible.

In step c3, the videos after the cluster analysis and the corresponding text information are stored to form a classified video library, thereby providing a convenient data classification system for retrieval.

Among them, the retrieval process is as follows:

In step d1, the user provides an image, and the retrieval module extracts the features of the image, and then performs matching in the classification video library;

Step d2, calculate the distance between the feature vector of the image to be retrieved and the cluster center vector of each class in the key frame feature library, and find out the three closest classes;

Step d3, then calculate the distance between the feature vector of each image frame in the three categories and the feature vector of the image to be retrieved;

Step d4, find out the 20 nearest image frames;

In step d5, count the top 5 videos most associated with the 20 image frames, and return a total of 15 results.

The above-mentioned embodiment is only an embodiment of the present invention, but should not be construed as limiting the scope of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention.

Claims (9)

1. A power information video search system, comprising an acquisition device, a processing device, a classification device and a retrieval device, The acquisition device includes a web page analysis module, a video text library, a video address extraction module and a video downloader. The web page analysis module is used to obtain the link of the video playback address and parse out the text information of the video through HTML text. The video text library is used to store the video The text information, the video address extraction module is used to obtain the real download address of the video, and the video downloader downloads the video through the method of video resource segmentation; The processing device includes a video standardization module, a video library and a video feature extraction module. The video standardization module is used to convert the video downloaded by the video downloader to form a standard video with a unified format. The video feature extraction module structures the standard video Process and obtain the structural features and key frames of the standard video, and the video library is used to store the standard video and its structural features and key frames; The classification device includes a classification module and a classification video library. The classification module performs cluster analysis according to the information in the video text library and the video library. The classification video library stores the clustered analyzed videos and related information according to the cluster structure, thereby forming a retrieval database ; The retrieval device includes a retrieval interface and a query module, the user inputs images and retrieval information through the retrieval interface and displays the retrieval results on the retrieval interface to return to the user, It is characterized by: According to the scope of crawling, the web page analysis module obtains links with video playback addresses in each video website, analyzes the HTML text corresponding to such links, parses and extracts text information related to the video, and performs Chinese word segmentation on these text information as video content. The title is stored in the video text library; According to the current link, the video address extraction module extracts the real download address of the video, and then calls the video downloader to download; The video standardization module converts the downloaded videos into the same format and stores them in the video library; The video feature extraction module will analyze the videos in the video library, extract structural features and key frames to form structured information, and store them in the video library; The classification module first performs pre-classification according to the video text database, and then classifies according to the structured information of the video text database and video, establishes a video database and stores it in the classified video database; The user provides an image example for retrieval through the retrieval interface, and the query module returns the videos matching the retrieval conditions to the user. 2. A kind of electric power information video search system according to claim 1, is characterized in that the structured processing of video feature extraction module is specifically: Step a1, construct the video structure. The video data structure is divided into three levels: scene, shot and frame. A frame is an independent static image. A group of frames constitutes a shot, a group of shots constitutes a scene, and a scene constitutes a video. Video structure The construction process is divided into two steps, Step a1.1, extracting shots from the video stream, Perform lens cut detection on the video stream, and find the image frame when the camera switch occurs in the video, specifically: Divide the image frame into sub-blocks of 8×8 pixels, calculate the average value of each sub-block, and calculate the absolute value of the difference between the average values of the sub-blocks at the same position in two consecutive frames in the video sequence as the inter-frame difference ; Calculate the ratio of the new edge pixel increase and the edge pixel reduction ratio of two adjacent frame images, and take the maximum value as the ratio difference; If the inter-frame difference is greater than the preset value, and the ratio difference is greater than the preset value, then it is determined that a lens switch has occurred, and the lens is extracted; Step a1.2, extract keyframes from the shot, A key frame is an image that can describe the main content of a shot. Since the redundant information in the video is eliminated, using a key frame to represent a shot can greatly reduce the workload of building a video index. The method of extracting a key frame is as follows: When there is a significant change between the current frame of the shot and the latest image judged to be a key frame, the current frame is used as a new reference key frame. First, take the first frame of the shot as a reference key frame, and then use Compare the frame image with the features of this key frame image, if the change is large, use the current frame as a new key frame, and then continue to compare with the subsequent frame image, and so on to obtain key frames one after another; Step a2, feature extraction, convert the RBG color space of the key frame to the color space HSV mode that is more in line with people's subjective understanding of color, quantize the color space into several color bars, and then automatically divide the image into several color bars with color segmentation technology Regions, each region uses a color bar to form an index, and the description of the image is converted into a color index set. 3. A power information video search system according to claim 2, characterized in that: the shot is a basic video data unit, and the cut of the shot is divided into sudden change and gradual change. 4. Sudden change means that one shot is directly cut to another shot without any editing techniques, while gradual change means that an editing effect is added between one shot and another shot to make the change appear smoother. 5. The process of finding out the image frame when the video lens is switched is called the lens shear detection. 6. Edges are those discontinuous gray points in the gray space of an image. 7. A kind of electric power information video search system according to claim 1, is characterized in that, the method for the video resource segmentation of video downloader uses dichotomy to pass through to the IP address of different piece resource in video resource, the place of attribution Confirm to download the video, specifically: Step b1, determine the playing time of the video to be downloaded, and obtain the description of the starting point: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image002.jpg And end point description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image004.jpg IP address and attribution, if the starting point Description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image002a.jpg And end point description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image004a.jpg If the IP address and attribution are the same, confirm the video description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image006.jpg IP address, attribution, and enter step b4; if the starting point description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image002aa.jpg And end point description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image004aa.jpg If the IP address and the attribution are not the same, go to step b2; Step b2, obtain the description of the middle point of the video clip: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image008.jpg IP address and attribution, if step b2 is executed 10 times, go to step b4; Step b3, if the intermediate point description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image008a.jpg and starting point Description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image002aaa.jpg If the IP address and attribution are the same, then confirm the description of the video segment: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image010.jpg IP address, attribution of the segment, and description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image012.jpg Description of new video clips: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image006a.jpg And go to step b2; If the intermediate point description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image008aa.jpg and end point description: description: description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image004aaa.jpg If the IP address and attribution are the same, then confirm the description of the video segment: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image012a.jpg IP address, attribution of the segment, and description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image010a.jpg Description of new video clips: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image006aa.jpg And go to step b2; If the intermediate point Description: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image008aaa.jpg Description with starting point: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image002aaaa.jpg and end point description: description: description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image004aaaa.jpg If the IP address and attribution location are different, it will explain: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image010aa.jpg Description of new video clips: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image006aaa.jpg And enter step b2, and then specify: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image012aa.jpg Description of new video clips: Description: Description: E:\CPC client\cases\inventions\ab262ccc-d459-4128-a5d6-c287b7c5a38e\new\100001\dest_path_image006aaaa.jpg And go to step b2; Step b4, recording steps b2-b3 Determine the segmentation position of each video segment in the original video and the corresponding IP address and attribution, and determine the operator of each video segment, and download the video in blocks, thereby optimizing the download speed. 8. A kind of electric power information video search system according to claim 1, it is characterized in that, the cluster analysis method of classification module carries out data mining to the key frame in the video storehouse, carries out automatic clustering to key frame, adopts video semantics The way to combine the information with the visual features of the key frame, specifically: Step c1, perform pre-classification according to the video text, classify videos with similar text information into one category, and ensure that the main content of the video belongs to one category; Step c2, on the basis of pre-classification, clustering is carried out according to the color features of the key frames of the video library in each major category, and the key frames with similar color features are aggregated into a sub-category; In step c3, the videos after the cluster analysis and the corresponding text information are stored to form a classified video database, thereby providing a convenient data classification system for retrieval. 9. A kind of power information video search system according to claim 1, characterized in that the retrieval process is as follows: In step d1, the user provides an image, and the retrieval module extracts the features of the image, and then performs matching in the classification video library; Step d2, calculate the distance between the feature vector of the image to be retrieved and the cluster center vector of each class in the key frame feature library, and find out the three closest classes; Step d3, then calculate the distance between the feature vector of each image frame in the three categories and the feature vector of the image to be retrieved; Step d4, find out the 20 nearest image frames; In step d5, count the top 5 videos most associated with the 20 image frames, and return a total of 15 results.