JP5664374B2 - Digest video generation apparatus and program - Google Patents

Description

  The present invention relates to a parameter selection technique for analysis processing for extracting video parts by analyzing video data in digest video generation.

  In recent years, with the widespread use of hard disk recorders, a large amount of automatically recorded images such as TV images have been accumulated, and the trouble of finding the desired images has become a problem. In addition, there are similar problems with video of various other genres such as surveillance camera video, home video, and business video in a company.

  As a countermeasure to the above problem, there is a method of using a digest video in which a desired video portion (a part of a video having a certain unit) is extracted from the stored video and connected. The video extracted as the digest video is a so-called scene. For example, a “baseball scoring scene” is a video portion (scene) obtained by cutting out only a scene in which a score has been raised in a baseball game video.

  By viewing a digest video that connects scenes that are such highlights, the contents of the entire video can be grasped in a short time. Also, by playing back the digest video one after another, it is possible to reduce the trouble of finding the video you want to see.

  It takes time and effort to manually generate a digest video by extracting a scene (hereinafter referred to as a highlight scene) as a highlight desired by the user. Therefore, there is a need for a digest video generation device that automatically extracts scenes as highlights from video data and generates a digest video.

  Conventionally, as a processing technique for extracting a specific part from video data, content data is time-divided, a plurality of long scenes are extracted for each division time, and a specific thumbnail image is extracted from the extracted scene based on the feature points that become thumbnails. Devices for generating are known.

JP 2001-298711 A

  The highlight scene varies depending on the genre of the video and the content of the required attractions. For example, in the case of sports images such as baseball and soccer broadcast on television, a “scoring scene” is generally given as a highlight scene. On the other hand, highlight scenes in the case of action movies are “action scenes” such as battle scenes and car chase scenes. In addition, the highlight scene in the video taken by the surveillance camera is a scene in which the presence of a suspicious person is reflected.

  In the digest video generation process, when a highlight scene is extracted from the video, it is necessary to perform an analysis process corresponding to the video and the highlight. As an example, if the scoring scene of baseball is a highlight scene, the process of analyzing the cheer of the audience in the video, and if the car chase scene of the action movie is a highlight scene, the process of analyzing the sound during the car chase Thus, it is possible to extract a scene for an appropriate digest video.

  Therefore, when there are a plurality of analysis processing parameters executed in digest video generation, it is necessary to select an analysis processing parameter that can appropriately extract a highlight scene according to the genre of the target video data.

  An object of the present invention is to select an analysis processing parameter corresponding to a genre of target video data from a plurality of analysis processing parameters executed in digest video generation, and to appropriately select a highlight scene corresponding to the genre of video data. It is providing the digest image | video production | generation apparatus to extract to.

  Another object of the present invention is to provide a program for causing a computer to execute processing realized by the digest video generation apparatus.

The digest video generation apparatus disclosed in the present application is a digest video generation apparatus that generates a digest video using a section extracted by analyzing and extracting a video of a specific genre, and 1) specifies a section to be a digest video. accepted, the true interval designation unit for the acceptance interval and true interval, 2) receives the designation of the range for changing the value of the analysis parameter and the parameter should specify the section to be digest video, accepted analysis parameters and then extracting the section to determine the digest image by analyzing the image by the value, and digest video generation unit to extraction interval, 3) whenever the digest video generation unit extracts the extraction section determines that digest video from a video in a match degree calculating section for calculating a degree of matching between the extraction between out-ku said true interval 4) Parameter evaluation value calculation for calculating an evaluation value for each changed value of the analysis processing parameter by performing an analysis processing parameter extracted from the extraction section with a high degree of coincidence and an evaluation processing for which the value becomes a high evaluation value And 5) a parameter selection unit that selects the analysis processing parameter having the highest evaluation value and the value from the analysis processing parameters executed by the digest video generation unit.

  According to the digest video generation apparatus disclosed above, in the video data analysis processing for generating the digest video, the analysis processing parameters and the values to be the determination conditions for highlight scene extraction are selected based on the video genre. Therefore, an appropriate digest video according to the video genre can be generated.

It is a figure for demonstrating the relationship between an analysis process parameter and an extraction result. It is a figure for demonstrating the relationship between an analysis process parameter and an extraction result. It is a figure for demonstrating the example of an analysis processing parameter of a loudness. It is a figure which shows the structural example of the digest image | video production | generation apparatus in one Example. It is a figure which shows the example of the image | video and correct answer area in one Example. It is a figure which shows the example of the parameter information in one Example. It is a figure which shows the example of the correct answer area in one Example. It is a figure which shows the example of the parameter investigation table in one Example. It is a figure which shows the example of the digest image | video production | generation in one Example. It is a figure which shows the example of the digest image | video extraction result table in one Example. It is a figure which shows the example of a coincidence degree calculation in one Example. It is a figure which shows the example of the parameter matching degree table in one Example. It is a figure for demonstrating the irregularity of the conversion of the coincidence degree of the analysis process parameter in one Example. It is a figure which shows the example of the parameter evaluation table in one Example. It is a figure which shows the example of a processing flow of the digest image | video production | generation apparatus in one Example.

  First, the analysis process executed by the digest video generation device considered by the present inventor will be described.

  The process of extracting a highlight scene from a video in order to generate a digest video is realized by combining various analysis processes such as “extraction of sound volume” and “extraction of image change” in the video. These analysis processes have parameters, and each parameter has one numerical value. As an example of the analysis processing parameter, there is a time interval for extracting the loudness in the “sound loudness extraction” processing. Therefore, the analysis process can be changed by changing the analysis process parameter, and the extraction result of the analysis process can be changed by changing the numerical value of the analysis process parameter.

  As described above, the highlight scene differs for each video genre. The video genre is the genre of the video content. Examples of video genres include broadcast program video, movies, home video video, surveillance camera video, and commercial video. Each genre has a lower genre. For example, in the case of broadcast program video, there are subordinate genres such as news, sports, and dramas, and sports include subgenres such as baseball, soccer, and figure skating.

  Therefore, since the highlight scene is different for each video genre, it is necessary to prepare a digest video generation apparatus in which analysis processing parameters corresponding to the video genre to be processed are set.

  On the other hand, if the highlight scene to be extracted is the same, it may be possible to cope by changing the value of the analysis processing parameter. For example, it is assumed that a score scene is extracted as a highlight scene when a digest video of a sports video is generated. In the scoring scene, since the audience generally cheers, if the cheer can be extracted, the scoring scene can be extracted. As a feature of the cheers to be extracted, if the sport is baseball, the audience will cheer immediately when a timely hit or the like occurs, so “the loudness of the sound suddenly increases”. Therefore, by analyzing the audio of the video, it is possible to extract a video portion that satisfies the condition that “the volume of the sound is large and the speed at which the sound is loud” is high as a highlight scene. In actual digest video generation, many other analysis processes are combined. Also, in the case of soccer video, since the audience generally cheers in the scoring scene, if the cheer can be extracted in the same way, the scoring scene can be extracted. Therefore, the same analysis processing parameters can be used when a scoring scene is extracted as a highlight scene from a baseball or soccer game video.

  However, since the method of raising the cheering of the audience at the time of scoring differs between baseball and soccer, the values of the analysis processing parameters of “sound volume” and “sound increase speed” are not the same. For example, in the case of a baseball video, the scoring scene is a home run or a timely hit, and the audience responds and cheers at the moment it is hit. Is done. On the other hand, in the case of a soccer video, there is a case where a chance is generated from a momentary offense and defense similar to baseball, and the audience reacts, but there is also a case where the ball is dribbling from the center and shot. When shooting from dribbling, it takes a certain amount of time to bring the ball to the vicinity of the goal, so the cheer of the audience tends to gradually increase as the goal approaches, making it difficult to satisfy the condition of sudden increase. Therefore, among the above two parameters, “sound volume” is more important.

  In this way, the problem in building a process for extracting highlight scenes from video is what kind of analysis processing parameters are set and what values are set as described above. Is a point. That is, the processing result of the extraction process, that is, the highlight scene to be extracted varies greatly depending on the analysis processing parameter and its value.

  Therefore, it is necessary to select an analysis processing parameter for generating a digest video and a value of the analysis processing parameter based on the genre of the video to be processed.

  Hereinafter, an embodiment of a digest video generation apparatus disclosed as one aspect of the present invention will be described.

(1) Selection of analysis processing parameters The digest video generation apparatus executes analysis processing parameters selected from a plurality of analysis processing parameter candidates, and analyzes and extracts one or more highlights extracted from the processing target video. A digest video is generated using the scene.

  The digest video generation apparatus has a plurality of executable analysis processing parameter candidates in advance. The digest video generation device receives a video of a specific genre of interest, and further, from the video, a section (a start time and an end time) that should be extracted as a scene (highlight scene) that is a highlight that the user desires. A designation indicating a set) is received, and the accepted section is set as a correct section.

  Next, the digest video generation apparatus extracts a plurality of analysis processing parameter candidates one by one, changes the value (parameter value), analyzes the video, and determines the interval between the section extracted as a result of the analysis and the correct section. The degree of coincidence is calculated. After that, the digest video generator selects the analysis processing parameter candidate and its value when the degree of coincidence between the extracted section and the correct section is the highest, and sets it as the analysis processing parameter in digest video generation for the processed genre. To do.

  Since the range of the extraction result is changed by changing the analysis processing parameter and its value, the digest video generation apparatus can select the optimal analysis processing parameter value for each video genre.

(2) Adjustment of analysis processing parameter value In some cases, new analysis processing is added to improve the processing accuracy of digest video generation. For example, since an advertisement video (CM) exists in a baseball video broadcast on television, the CM may be extracted as a highlight scene. In order to cope with such erroneous extraction, a function for analyzing and excluding CMs in video is added as analysis processing in digest video generation. In such a case, not only the number of new parameters increases, but the optimum value of the existing parameter may change, and in many cases, the value of the analysis processing parameter needs to be readjusted. Actually, it is not necessary to change the values of all analysis processing parameters, but it is necessary to know which analysis processing parameter values are effective adjustments. As the number of analysis processing parameters executed for digest video generation increases, it takes a lot of time to specify an analysis processing parameter whose value is to be adjusted.

  Furthermore, some analysis processing parameters do not improve the extraction results even if their values are changed. In other words, if the extracted highlight scene does not approach the correct answer section even after the analysis process with the value changed, it will be understood by humans how the extracted highlight scene changes from the one before the change. It may be difficult.

  1 and 2 are diagrams for explaining the relationship between analysis processing parameters and extraction results.

  As shown in FIG. 1, when the value is changed to 100, 150, and 300, there is an analysis processing parameter that is not related to each value and the extraction result (extracted section). Such analysis processing parameters are called irregular analysis processing parameters. On the other hand, as shown in FIG. 2, there are analysis processing parameters that are considered to be related to each value and the extraction result when the values are similarly changed. Such analysis processing parameters are called non-irregular analysis processing parameters.

  It is not efficient to change the value of an irregular analysis processing parameter when adjusting the analysis processing parameter. For example, in the sound volume analysis process, when subtracting the time interval for extracting the sound volume or dividing the image into partial areas, the parameters for the size of the partial area are irregular. Conceivable.

  Conversely, regarding the cheer length analysis process, the length parameter is considered not irregular. When the audience cheers in baseball, the cheering generally lasts longer for scoring scenes, but for loud play scenes such as large outfield flies, the size suddenly increases after cheering Becomes smaller. For example, as shown in Fig. 2, if the value of the parameter that determines the length of cheers is increased (longer), it will be difficult to extract the part other than the part corresponding to the scoring scene. Even parts that do will not be extracted. Also, if the value of the parameter that determines the length of cheers is reduced (shortened), parts other than the scoring scene can be easily extracted. Therefore, the extraction result can be changed efficiently by setting the value of the image analysis processing parameter that is not irregular as the adjustment target.

  The disclosed digest video generation apparatus can select an analysis processing parameter that is not irregular as described above and can change the parameter value. Therefore, the adjustment of the image analysis processing parameters can be made efficient.

(3) Specific examples of analysis processing parameters Next, the analysis processing parameters for digest video generation executed by the digest video generation device will be described more specifically.

  As an example, a score scene is extracted as a digest video of a baseball video. In the scoring scene, the audience generally cheers, so if the cheer can be extracted, the scoring scene can be extracted.

  One of the features of cheers is that when a timely hit occurs, the audience loudly cheers and the “sound volume” increases. Furthermore, in the case of a scoring scene, cheers last longer, but conversely, in the case of ugly play such as a large outfield fly, the size suddenly decreases after cheering.

  Therefore, as a method for extracting a scene used for a digest video from a baseball video, first, a process of extracting “sound volume” is executed.

  FIG. 3 is a diagram for explaining an example of a sound volume analysis processing parameter.

  FIG. 3 shows an average of audio included in the video at predetermined time intervals. The loudness of the sound can be extracted by averaging the sound data at a fixed time interval (for example, 0.1 second interval).

  In order to extract the point in time when the cheer suddenly increased, the sound volume is averaged within a certain time before and after a certain point.

  When cheers are raised at a certain point in time, the average loudness in the interval before that point becomes small, and the average loudness in the interval after that point becomes large. As shown in FIG. 3, the length of the fixed time range for averaging is set to different values before and after.

  In particular, it is conceivable that the length of the subsequent section greatly affects the extraction of the scoring scene. In the scoring of the scoring scene, the cheering continues for a long time, and the length of the interval that takes the average of the latter is designated as the cheering length by the analysis processing parameter. The difference between the previous average and the subsequent average is extracted by setting the length of the section taking the previous average to about “2 seconds” and the length of the section taking the subsequent average (cheering length) to about “10 seconds”. Then, the time when the difference is large is set as one of the evaluation values of the section to be extracted.

  Further, in order to extract the speed at which the cheer increases, the slope of the loudness of the sound before and after that point specified by the above processing is calculated. Taking the magnitude of this slope as one of the evaluation values, the final evaluation value is calculated by adding two evaluation values with weights.

  In addition, other analysis processing parameters can be used in digest video generation. In the case of a broadcast video, since an advertising video (CM) is inserted, the CM section may be erroneously extracted as a highlight scene. In order to deal with such erroneous extraction, it is possible to detect a CM section in advance and not to extract a highlight scene from the detected CM section.

  When detecting a CM section in digest video generation, the length of the CM section is a multiple of 15 seconds, and about 0.5 seconds at the start and end of the CM section (total of 1 second between two CM sections) ) Features of the CM section such as a silent section is included. In the case of an actual broadcast video, the CM section does not necessarily include a silent section of about 0.5 seconds due to the effects of program editing and noise. Further, since a silent section is included in a news program or the like inserted in a baseball video, erroneous detection increases.

  Therefore, as analysis processing parameters for accurately detecting the CM section, there are analysis processing parameters such as a silent threshold when extracting the silent section from the loudness and a minimum value (seconds) of the silent section in the CM section. use. Furthermore, since CM sections often appear continuously, analysis processing parameters such as the maximum length and minimum length of CM sections that appear as continuous are used.

(4) Configuration Example of Digest Video Generation Device Next, a configuration example in one embodiment of the disclosed digest video generation device will be described.

  FIG. 4 is a diagram illustrating a configuration example of a digest video generation apparatus according to an embodiment.

  The digest video generation apparatus 1 that realizes the above-described processing includes a correct answer section designation unit 11, a parameter range setting unit 12, a parameter change unit 13, a digest video generation unit 14, a matching degree calculation unit 15, and a parameter evaluation value calculation unit 16. , And a parameter selection unit 17.

  The correct answer section designation unit 11 accepts designation of a section to be a digest video in the video 2 and holds a correct answer section 3 indicating the accepted section.

  FIG. 5 is a diagram illustrating an example of the video 2 and the correct answer section 3.

  Video 2 is video data including audio that is a target of digest video generation.

  The correct answer section specifying unit 11 receives the video 2 and specifies the start time and end items of the section indicating the correct answer section in the video 2 specified by the user, and sets it as the correct answer section 3.

  The correct answer section 3 is information indicating a correct section (highlight scene) that is manually specified for a certain video 2 and is extracted as the digest video 5.

  FIG. 6 is a diagram illustrating an example of the correct answer section 3.

  The correct answer section table 31 as an example of the correct answer section 3 has items of video ID, section ID, start time, and end time.

  The section ID is a unique identification number in a section of a certain video 2. The video ID is a number that uniquely identifies the target video 2. The correct answer section table 31 in FIG. 6 shows an example having an item of video ID, but when there is only one video 2, the data item of video ID is unnecessary. When there are a plurality of videos 2, a video table for managing the file names of the corresponding videos is separately provided and referred to to describe the video file names. The start time and the end time indicate the start and end positions of the section indicated as the correct section by the time from the head of the video 2.

  In the correct answer section table 31, the start time and end time of each section designated as the correct answer section 3 are described.

  The parameter range setting unit 12 receives an analysis processing parameter for specifying a section to be a digest video and parameter information 4 indicating a range in which the value of the parameter is changed, and generates a digest video based on the received parameter information 4 For each analysis processing parameter used by the unit 14, a value to be changed within a specified range is determined and set in the parameter investigation table 121.

  The parameter information 4 is information in which an initial value for each analysis processing parameter in digest video generation and a value change range (maximum value and minimum value) are set.

  FIG. 7 is a diagram illustrating an example of the parameter information 4.

  The parameter table 41 as an example of the parameter information 4 has items of parameter ID, parameter name, initial value, minimum value, and maximum value. The parameter ID is a unique identification number for each analysis processing parameter. The initial value is a value used in the first analysis process of the digest video generation unit 14, and the minimum value and the maximum value are values indicating a range in which the value is changed in the second and subsequent processes.

  In the parameter table 41, data is set in these data items for each analysis processing parameter. For example, “Cheer Length” with parameter ID = 0 takes “10.0” as the initial value, and the value is changed within the range from the minimum value “5.0” to the maximum value “20.0”. It is used for.

  FIG. 8 is a diagram illustrating an example of the parameter investigation table 121.

  The parameter survey table 121 has items of parameter ID, parameter survey ID, and parameter value. The parameter ID is a unique identification number of the analysis processing parameter. The parameter value is a value determined by the parameter range setting unit 12, and is set in association with a parameter investigation ID that uniquely identifies the value.

  The parameter changing unit 13 changes the value of the analysis processing parameter of the digest video generating unit 14 using the value set in the parameter investigation table 121.

  Based on the parameter investigation table 121, the digest video generation unit 14 analyzes the video 2 based on the analysis processing parameters and the changed values, and extracts a section that is determined to be a digest video.

  FIG. 9 is a diagram illustrating an example of digest video generation.

  As an example, the digest video generation unit 14 extracts the section (highlight scene) by analyzing the audio with the value of the designated analysis processing parameter for the video 2 including the image and the audio, and the digest video extraction result table In 141, the extraction result is described.

  FIG. 10 is a diagram illustrating an example of the digest video extraction result table 141.

  The digest video extraction result table 141 has items of video ID, parameter survey ID, section ID, start time, and end time.

  The video ID is an identification number of the extracted video 2. The parameter investigation ID is an identification number of each value of the analysis processing parameter. The section ID is a number that uniquely identifies the section extracted by the analysis process. The start time and end time indicate the start and end positions of the extracted section by the time from the head of the video 2.

  In the example of FIG. 10, in the video 2 with the video ID “0”, the section (section ID is “0”) extracted when the parameter survey ID is “0 (value = 0.1)” is the start time. Represents an interval from “11.0” to the end time “14.5”.

  The degree of coincidence calculation unit 15 calculates the degree of coincidence between the extracted section and the correct answer section 3 every time the digest video generation section 14 analyzes the video 2 and extracts a section determined to be the correct section 3.

  FIG. 11 is a diagram showing an example of coincidence calculation.

  As shown in FIG. 11, for example, the coincidence degree calculation unit 15 compares the correct answer section 3 with the section that is the extraction result of the digest video generation section 14, and obtains the ratio between the matching section and the mismatch section. The degree of coincidence with the analysis processing parameter from which the highlight scene is extracted and its value is calculated.

  FIG. 12 is a diagram illustrating an example of the parameter matching degree table 151.

  The parameter matching degree table 151 has items of video ID, parameter ID, parameter investigation ID, and matching degree. The video ID is the identification number of video 2, the parameter ID is the identification number of the processed analysis processing parameter, the parameter survey ID is the identification number of the value of the processed analysis processing parameter, and the matching degree is the calculated matching degree. .

  The parameter evaluation value calculation unit 16 calculates an evaluation value for each value of the analysis processing parameter executed by the digest video generation unit 14 using an expression that increases as the degree of matching increases.

  Alternatively, the parameter evaluation value calculation unit 16 examines the irregularity of the change in the degree of coincidence calculated corresponding to the changed value for each analysis processing parameter executed by the digest video generation unit 14. An evaluation value is calculated using an expression that results in a high evaluation value for those not present.

  FIG. 13 is a diagram for explaining irregularities in the conversion of the degree of coincidence of analysis processing parameters.

  FIG. 13A is a graph showing changes in the degree of coincidence when the analysis parameters are irregular, and FIG. 13B is a graph showing changes in the degree of coincidence when the analysis parameters are not irregular. Shown in the form.

  As shown in FIG. 13A, when the value of an irregular parameter changes (for example, increases), the degree of coincidence repeats up and down regardless of the change in value. On the other hand, as shown in FIG. 13B, when the parameter is not irregular, when the value changes (for example, increases), the degree of coincidence with each parameter value corresponds to the change in the parameter value. Indicates an increasing trend, a decreasing trend, or an increasing / decreasing trend with one peak.

  As shown in FIGS. 13A and 13B, the irregularity of the degree of coincidence with respect to the value of the analysis processing parameter increases the distance of the graph line when the degree of coincidence is graphed.

The parameter evaluation value calculation unit 16 graphs the change in the degree of coincidence m (p _v ) when the value p _v of the analysis processing parameter p is changed, and obtains the length of the graph path. As an example, the parameter evaluation value calculation unit 16 calculates an evaluation value for each analysis processing parameter under the following conditions.

・ When the change in the degree of coincidence is graphed, the path length is short.
・ The maximum value of matching is large.

  The parameter evaluation value calculation unit 16 calculates the evaluation value r using, for example, the following evaluation formula.

  The left item on the right side of the above equation (1) is the maximum value of the matching degree. The central item on the right side is the difference between the maximum value and the minimum value of the degree of coincidence, and is an item (normalization term) for adjusting the size when calculating the path of coincidence. The right-hand item on the right side is the length of the road. The smaller this value, the larger the evaluation value. Since the central item of equation (1) is necessarily smaller than the right item, the evaluation value r takes a value between 0 and 1.

  The parameter evaluation value calculation unit 16 stores the evaluation value calculated for each analysis processing parameter in the parameter evaluation table 161.

  FIG. 14 is a diagram illustrating an example of the parameter evaluation table 161.

  The parameter evaluation table 161 has data items of parameter ID and evaluation value. The parameter ID is a unique identification number of the analysis processing parameter, and the evaluation value is an evaluation value calculated correspondingly.

  The parameter selection unit 17 selects an analysis processing parameter having a high evaluation value and its value from the result of the parameter evaluation value calculation unit 16.

(5) Process Example of Digest Video Generation Device Next, the flow of processing of the digest video generation device 1 will be described.

  FIG. 15 is a diagram illustrating a processing flow example of the digest video generation device 1.

  Step S1: The correct answer section specifying unit 11 specifies the correct answer section 3 for the video 2.

  The correct answer section specifying unit 11 prepares a correct answer section table 31 in advance, detects the start time and end time of each section indicating the correct answer section (highlight scene) specified by the user in the received video 2, and corrects the correct answer. Set in the interval table 31.

  Step S2: The parameter range setting unit 12 specifies a list of analysis processing parameter values to be changed.

  The parameter range setting unit 12 accepts a parameter table 41 which is an example of the parameter information 4, and determines and determines a plurality of values to be extracted as a range to be changed for each analysis processing parameter based on the parameter table 41. A value is set in the parameter investigation table 121.

  The parameter range setting unit 12 performs one of the following processes as a value determination process, and determines a value using the following equation. In the following equations (2) to (4), the i-th value of the parameter is p (i), the initial value of the parameter is init, the maximum value of the parameter is max, the minimum value of the parameter is min, and the number of parameters is N.

  The parameter range setting unit 12 obtains p (i) by Equation (2) when determining the value by equally dividing the maximum value and the minimum value.

Formula (2): p (i) = (max-min) · i / N + min
Further, the parameter range setting unit 12 obtains p (i) by the expression (3) when determining the values by equally dividing the maximum value and the minimum value from the initial value.

Formula (3):
if i> = N / 2 p (i) = (max-init) ・ (iN / 2) N + init
else p (i) = (max-min) i / N + min
When the parameter range setting unit 12 determines a value based on the relationship between the maximum value, the minimum value, and the initial value, for example, the minimum value is 1/10 of the initial value, and the maximum value is 10 times the initial value. If it is, the value is determined by dividing the logarithmic axis equally. In this case, p (i) can be obtained by the following equation (4).

Formula (4):
if i> = N / 2 p (i) = exp (log max-log init) ・ (iN / 2) N + log init
else p (i) = exp (log max-log min) i / N + log min
Step S3: The parameter range setting unit 12 sets the analysis processing parameters executed by the digest video generation unit 13 to initial values based on the parameter table 41.

  Step S4: The digest video generation unit 14 extracts a section from the video 2 with the current value of the analysis processing parameter. For example, the digest video generation unit 14 extracts the section (highlight scene) from the baseball video 2 by analyzing the volume of the sound, and extracts the extracted section, the analysis processing parameter executed, and the extraction result. The value and the like are stored in the digest video extraction result table 141.

  The digest video generation unit 14 extracts the highlight scene from the video 2 by performing analysis processing with all the analysis processing parameters being initial values as the first processing. In the second and subsequent processing, the digest video generation unit 14 extracts a highlight scene that has been similarly analyzed using the value (changed value) changed by the parameter changing unit 13 based on the parameter survey table 121.

  Step S5: The coincidence degree calculation unit 15 calculates the degree of coincidence between the correct answer range 3 and the extraction result of the digest video generation unit 14.

  The coincidence degree calculation unit 15 obtains a coincidence degree for each parameter survey ID based on the digest video extraction result table 141 and the correct answer section table 31. The degree of coincidence is the ratio (coincidence) that matches when overlapping sections.

  For example, the coincidence degree calculation unit 15 obtains a matching section and a non-matching section based on the start time and end time of each section of two tables (the digest video extraction result table 141 and the correct answer section table 31). The degree of coincidence is obtained from the length of each section.

  The coincidence degree calculation unit 15 acquires one section in order from the top of the two tables, determines the overlap between the two acquired sections, and repeats the following processing until the data of the two tables is completed.

  If there is no overlap between the two extracted sections, the coincidence calculation unit 15 adds the length of the earlier section (starting time before) to the length of the unmatched section. Then, the coincidence degree calculation unit 15 acquires a section next to the previous section in terms of time. If there is no next section, the coincidence degree calculation unit 15 adds the length of the other section to the length of the unmatched section, and ends the calculation.

  When there is an overlap between two sections and the start time is deviated, the coincidence calculation unit 15 adds the length of the non-matching section by the amount of “deviation”, and the overlapping section is matched. Add to the length of. Then, the coincidence degree calculation unit 15 acquires the next section of the section with the earlier end time. The start time of the section with the later end time is set as the overlap end section. When there is no next section, the coincidence degree calculation unit 15 adds the length of the other section to the length of the unmatched section, and ends the calculation.

  As another method, the coincidence calculation unit 15 can obtain the coincidence by a predetermined calculation formula.

  The calculation formula for the degree of coincidence m is expressed by the following formula (5), where e is the total length of the extraction results, c is the total length of the correct answer sections, and u is the total length of the matching sections. be able to.

Formula (5): m = 2u / (e + c)
The total length c of the correct answer sections can be obtained by adding all the time differences between the start time and end time of the same video ID in each section of the correct answer section table 31. Similarly, the total length u of the matching sections can be obtained by adding all the time differences between the start time and the end time of the section of the same parameter ID of the same video ID in the digest video extraction result table 141. In the above formula (5), the denominator is the sum of all the sections, and the length of the section where the numerators match (× 2). Therefore, when all the sections are completely coincident, 1 is 0 when they are not overlapped at all, and when neither is, the value is larger than 0 and smaller than 1. It can be seen that the larger the value is, the more consistent the two sections are.

  The coincidence calculation unit 15 stores the calculated coincidence in the parameter coincidence table 151.

  Step S6: The parameter changing unit 13 checks whether or not all values of the analysis processing parameters have been changed.

  If all the values of the analysis process parameters set in the parameter investigation table 121 have not been changed (N in step S6), the process in step S7 is performed. If the change has been completed (Y in step S6). , Step S8 is performed.

  Step S7: The parameter changing unit 13 changes the analysis processing parameter.

  The parameter changing unit 13 obtains the parameter ID and the next value of the analysis processing parameter specified by the parameter investigation ID based on the parameter investigation table 121, and changes the analysis processing parameter value of the digest video generation unit 14. .

  Step S8: The parameter evaluation value calculation unit 16 calculates the evaluation value of the analysis processing parameter.

  The parameter evaluation value calculation unit 16 calculates the evaluation value of the analysis processing parameter based on the result of the parameter matching degree table 151.

  The parameter evaluation value calculation unit 16 calculates an evaluation value based on the parameter matching degree table 151 by using an evaluation formula that makes the evaluation value of the irregular analysis processing parameter lower.

  For example, the number of parameter survey IDs per parameter is N, the i-th coincidence in the parameter coincidence table 151 is m (i), and the maximum and minimum values of coincidence when i is 0 or more and less than N Is described as follows.

Maximum value = max (m (i))
Minimum value = min (m (i))
In that case, the evaluation value r can be expressed by the following equation (6).

  In the case of the parameter matching degree table 151 shown in FIG. 12, in the analysis processing parameter with the parameter ID “0”, in the above formula (6), the left item (maximum value of matching degree) is “0.3”, the central item The (normalization term) is “0.15”, the right-hand item (distance length) is “0.35”, and as a result, the evaluation value r is “about 0.13”.

  The parameter evaluation value calculation unit 16 stores the evaluation value calculated for each parameter in the parameter evaluation table 161.

  Step S9: It is checked whether the parameter evaluation value calculation unit 16 has evaluated all the analysis processing parameters. If the parameter evaluation value calculation unit 16 has not evaluated all the analysis processing parameters (N in Step S9), the parameter evaluation value calculation unit 16 performs the processing in Step S10 and evaluates all the analysis processing parameters (Y in Step S9). , Step S11 is performed.

  Step S10: The parameter changing unit 13 returns the changed analysis process parameter value to the initial value, and proceeds to the process of step S7.

  Step S11: The parameter selection unit 17 selects an analysis processing parameter with a high evaluation value.

  The parameter selection unit 17 sorts the parameter IDs based on the evaluation values based on the parameter evaluation table 161, and selects a predetermined number of analysis processing parameters in order from the analysis processing parameter having the largest evaluation value. The parameter selection unit 17 holds the selected analysis processing parameter and its value in association with the genre of the video 2.

  When the parameter adjustment unit (not shown in FIG. 4) included in the digest video generation device 1 adjusts the value of the analysis processing parameter, the selected analysis processing parameter and its value are referred to and preferentially presented as an adjustment target. Then, the parameter value can be changed. This enables efficient adjustment.

(6) Hardware Configuration of Digest Video Generation Device The digest video generation device 1 is realized by a computer system including hardware having a CPU and a memory and a software program, or dedicated hardware. That is, the digest video generation device 1 includes a computing device (CPU), a temporary storage device (DRAM, flash memory, etc.) and a permanent storage device (HDD, flash memory, etc.), and a computer that inputs and outputs data to the outside. Can be implemented.

  The digest video generation apparatus 1 can also be implemented by a program that can be executed by this computer. In this case, a program describing the processing contents of the functions that the digest video generation apparatus 1 should have is provided. When the computer executes the provided program, the processing function of the digest video generation apparatus 1 described above is realized on the computer. The computer can also read a program directly from a portable recording medium and execute processing according to the program. Furthermore, the program can be recorded on a computer-readable recording medium.

(7) Effect of Digest Video Generation Device 1 As described above, the digest video generation device 1 evaluates each value of the analysis processing parameter as the degree of coincidence between the extracted highlight scene and the correct answer section 3 increases. The evaluation value is calculated by the formula that becomes the value. Therefore, according to the digest video generation device 1, it is possible to select an analysis processing parameter obtained by extracting a section closest to the correct answer section 3 and its value from the analysis processing parameters included in the parameter information 4. A digest video can be generated using the selected analysis processing parameter and value, and a digest video 5 including a scene that matches the correct answer section 3 can be output.

  In addition, the digest video generation apparatus 1 calculates an evaluation value for each analysis processing parameter with an expression that becomes a lower evaluation value as the irregularity of the degree of coincidence with respect to the changed value is irregular. Therefore, according to the digest video generation device 1, an analysis processing parameter whose extraction section changes corresponding to a change in value can be selected from the analysis processing parameters included in the parameter information 4. It is possible to adjust the preferential value of the selected analysis processing parameter, thereby realizing efficient adjustment.

DESCRIPTION OF SYMBOLS 1 Digest video production | generation apparatus 11 Correct segment specification part 12 Parameter range setting part 13 Parameter change part 14 Digest video generation part 15 Matching degree calculation part 16 Parameter evaluation value calculation part 17 Parameter selection part 2 Video 3 Correct answer area 4 Parameter information 5 Digest video

Claims (3)

A digest video generation device that generates a digest video using a section extracted by analyzing and extracting a video of a specific genre,
A correct section specifying unit that receives a specification of a section to be a digest video and sets the received section as a correct section;
An analysis processing parameter to specify a section to be a digest video and a specification of a range in which the value of the parameter is changed are accepted, and an analysis processing parameter and its value are analyzed to extract a section to be determined as a digest video . A digest video generation unit as an extraction section ;
Every time the digest video generation unit extracts the extraction section determines that digest video from the video, and the matching degree calculation unit that calculates a degree of coincidence between the Extraction Zone between said true interval,
A parameter evaluation value calculation unit that performs an analysis process parameter that extracts the extraction section having a high degree of coincidence and an evaluation process that results in a high evaluation value, and calculates an evaluation value for each changed value of the analysis process parameter; ,
A digest video generation apparatus comprising: an analysis processing parameter having the highest evaluation value and a parameter selection unit for selecting the value from the analysis processing parameters executed by the digest video generation unit. The parameter evaluation value calculation unit performs an evaluation process for an analysis processing parameter that has a lower evaluation value as the change in the degree of coincidence with the extraction result when the value is changed is irregular. Instead of the evaluation value for the changed value, an evaluation value based on the irregularity change of the matching degree is calculated for each analysis processing parameter,
The digest video generation device according to claim 1, wherein the parameter selection unit selects an analysis processing parameter having the highest evaluation value from the analysis processing parameters. A digest video generation program for causing a computer to execute a process of generating a digest video using a section extracted by analyzing and extracting a video of a specific genre,
Accepting the designation of the section to be the digest video, and the correct section designating process in which the accepted section is the correct section;
An analysis processing parameter to specify a section to be a digest video and a specification of a range in which the value of the parameter is changed are accepted, and an analysis processing parameter and its value are analyzed to extract a section to be determined as a digest video . Digest video generation processing as an extraction section ;
Each time of extracting the extraction section determines that digest video from the video in the digest video generation process, and the matching degree calculation process of calculating the degree of coincidence between the true interval and between the Extraction Ward,
A parameter evaluation value calculation process for calculating an evaluation value for each changed value of the analysis processing parameter by performing an analysis process parameter for extracting the extraction section having a high degree of coincidence and an evaluation process for which the value becomes a high evaluation value; ,
From the analysis processing parameters executed in the digest video generation processing, an analysis processing parameter having the highest evaluation value and a parameter selection processing for selecting the value are selected.
A digest video generation program that is executed by the computer.

Images