WO2019008635A1

WO2019008635A1 - Video monitoring device adjusting method and video monitoring device

Info

Publication number: WO2019008635A1
Application number: PCT/JP2017/024350
Authority: WO
Inventors: 浜田高宏
Original assignee: 株式会社Ｋ－Ｗｉｌｌ
Priority date: 2017-07-03
Filing date: 2017-07-03
Publication date: 2019-01-10
Also published as: JP7033797B2; CN110870305A; JPWO2019008635A1; CN110870305B

Abstract

Provided is a method for adjusting a video monitoring device for appropriately detecting video and/or errors occurring in accordance with the content of digital video signals. Also provided is a video monitoring device for appropriately detecting video and/or errors through learning from digital video signals. The video monitoring device comprises a video/audio monitoring unit that incorporates a memory in which threshold values are stored, and a parameter optimization learning unit. The video monitoring device is configured to determine that an error has occurred when all of a plurality of parameters varying in accordance with video signals exceed respective threshold values specified in accordance with monitoring items. At least two of the plurality of parameters are associated with each other, and the video monitoring device selects a set of threshold value groups stored in the memory in accordance with the content of the video and/or audio to be monitored. Through an analysis of error warning results, the parameter optimization learning unit performs learning and updates the threshold values on the basis of the learning result.

Description

Method of adjusting image monitoring apparatus and image monitoring apparatus

The present invention relates to an adjustment method and an image monitoring apparatus of an image monitoring apparatus capable of mechanically detecting an error of an image and an audio included in a digital image signal.

In the age of analog television broadcasting in the past, the quality of the video and audio was that an observer actually viewed (monitored) the video and audio and performed error detection. On the other hand, with the transition to digital television broadcasting, on the other hand, processing of video signals has become easy, and environments for mechanically monitoring video and audio have been established, and some of them have already been monitored. On the other hand, there is also an evaluation that machine monitoring is still often inferior to human monitoring in the accuracy of error detection, and there is a fact that monitoring by a conventional monitor is also performed in parallel.

However, in addition to the terrestrial business that transmits digital video, video distribution services such as Internet content distribution services and Internet Protocol TeleVision (IPTV) business are expected to expand in the future, and will be distributed accordingly. Since the number of such content also increases dramatically, it is considered to be practically difficult for the observer to monitor all the content. Therefore, it can be said that improving the accuracy of error detection in machine monitoring to human level is a pressing issue in all video distribution services.

Patent Document 1 discloses an audio inspection method for detecting an audio error caused by a video error caused by noise caused by various causes in a digital video signal and a noise generated by a different cause in a digital audio signal. It is done.

International Publication 2015/059782

According to Patent Document 1, continuous digital audio signals are divided and sampled at 5 msec or less, high frequency components are extracted from the sampled signals, and a value based on the extracted high frequency components is compared with a threshold value to obtain voice. Errors that occur can be detected mechanically.

However, in the technology as disclosed in Patent Document 1, the error detection is performed by comparing the characteristic value (parameter) of the audio signal with the threshold, but there is a problem as to how to set the threshold. That is, if the threshold is set to be strict, error detection is frequently performed, but if many of the detected errors are negligible for the viewer, detection may be wasted. On the other hand, if the threshold value is set loosely, although the frequency of error detection decreases, there is a risk that the viewer may miss an error that can not be ignored. This is one of the reasons why machine surveillance is inferior to human surveillance. Since changing one parameter affects other parameters, especially when there are multiple parameters that are related to each other, two or more parameters must be changed at the same time, and appropriate threshold settings should be made. Is very difficult.

One of the objects of the present invention is to provide a method of adjusting an image monitoring apparatus for appropriately detecting an image and / or an error generated depending on content in a digital image signal.

Another object of the present invention is to provide an image monitoring apparatus for detecting an image and / or an error properly through learning in a digital image signal.

According to a first aspect of the present invention, there is provided a method of adjusting an image monitoring apparatus according to the first aspect of the present invention, wherein an image signal corresponding to an image to be monitored and / or audio is input to detect that an error occurs in a monitoring item. In the adjustment method,
The video monitoring apparatus is configured to determine that the error has occurred when all of a plurality of parameters that change according to the video signal exceed a threshold determined according to the monitoring item. , At least two of the plurality of parameters are mutually related,
The threshold values of the plurality of parameters are respectively determined to create a plurality of threshold value groups, which are stored in a memory,
It is characterized in that any set of the threshold group stored in the memory is selected according to contents of video and / or audio to be monitored.

As a result of earnest research, the inventor of the present invention has determined that the appropriate value of the threshold group is the content of the video and / or audio content to be monitored ("movie", "drama", "variety", "sports", "documentary" , "Entertainment", "animation", etc.) have been found to be unevenly distributed. The present invention has been derived based on such findings. That is, if the threshold values of the plurality of parameters are respectively determined and a plurality of sets of threshold values are created and stored in the memory, they are stored in the memory according to the content of video and / or audio to be monitored. By selecting any set of the threshold group, a threshold group more appropriate for monitoring can be set as compared to the case where a single threshold group is used, so erroneous detection of an error in the image monitoring apparatus is effectively performed. It can be prevented.

Furthermore, the video monitoring device performs learning in parallel with the video and / or audio monitoring operation, updates at least one threshold of the threshold group based on the learning result, and stores the updated threshold group in a memory It is preferable to store in

Furthermore, the video surveillance device issues an alarm when an error is detected.
In the learning, a video signal corresponding to the video and / or audio to be monitored is input, and the alarm is issued at predetermined time intervals, the first event of the alarm for the error to be issued Count the number of events and the number of second events for which the alarm has been issued for errors that should not be issued,
Furthermore, a video signal for inspection whose content and place of occurrence of the error are known is input to the video monitoring apparatus, and the number of third events for which the alarm has not been issued for an error that should be issued. Count,
An evaluation value is determined by weighting and calculating the number of obtained first events, the number of second events, and the number of third events, and the threshold group is determined based on the determined evaluation values. It is preferable to update the

In the learning, it is preferable to count the number of the first events, the number of the second events, and the number of the third events while individually changing the thresholds of the threshold group.

A video monitoring apparatus according to a second aspect of the present invention is a video monitoring apparatus that receives a video signal corresponding to video and / or audio to be monitored and detects that an error has occurred in a monitoring item.
A threshold storage unit that stores a plurality of parameters that change according to the video signal, wherein at least two of the plurality of parameters are associated with each other;
A determination unit that determines that the error has occurred and issues an alarm when all of the plurality of parameters exceed the threshold determined according to the monitoring item;
A learning unit that performs learning in parallel with the video and / or audio monitoring operation;
An updating unit that updates the threshold based on a learning result in the learning unit;
A video storage unit for storing video signals before and after the occurrence of the error when the determination unit detects the occurrence of the error;
The learning unit may learn by analyzing the notification result of the error.

According to the present invention, the result of the error notification is analyzed and classified into, for example, "correct alarm", "unwanted alarm", "pass through", etc., whereby the learning unit performs learning to make the threshold more appropriate. The error detection accuracy is improved because the value can be updated to

Furthermore, when the learning unit inputs a video signal corresponding to the video and / or audio to be monitored, the alarm is issued for an error that should be issued at a constant time interval. Counts the number of first events and the number of second events for which an alarm should have been issued for errors that should not have triggered an alarm When the signal is input, the number of third events for which the alarm has not been issued for the error that should be issued is counted, and the number of the first event obtained and the number of the second events It is preferable to calculate an evaluation value by calculating by weighting the number of cases and the number of cases of the third event, and ranking the threshold values stored in the threshold storage unit based on the obtained evaluation values.

Furthermore, it is preferable that the learning unit count the number of first events, the number of second events, and the number of third events while individually changing the threshold value of the monitoring item.

Furthermore, the learning unit may be configured to cause the alarm to be issued for an error that should be issued for the m-th video signal of the M video signals during the time t1 to t2. Let T (m) be the number of events, and F (m) be the number of second events for which an alarm should have been issued for errors that should not have triggered an alarm. Assuming that the number of third events for which the alarm has not been issued is H (m), it is preferable to obtain an evaluation value A (m) according to the following evaluation function.

However,
Wt (m) is the weight of the first event Wf (m) is the weight of the second event Wh (m) is the weight of the third event

Furthermore, it is preferable that the video storage unit cuts out and stores the input video signal with a predetermined length at a predetermined timing.

Furthermore, it is preferable to have a display unit that displays the learning result of the learning unit and reproduces video and / or audio based on the video signal before and after the occurrence of the error.

Furthermore, it is preferable that the updating unit prioritizes and lists new threshold candidates that replace the current threshold.

In the present specification, "voice" includes sounds in addition to human and animal voices. "Video signal" includes the case of either video signal or audio signal. "Content" is the content of the video. Specifically, there are "movie", "drama", "variety", "sports", "documentary", "entertainment", "animation" etc. as the major classification of content. On the other hand, sub-classification of contents, if taking "sports" as an example, includes "baseball", "soccer", "volleyball", "judo", "sumo", "land athletics" and the like. “Learning” associates the monitoring result of watching the video and / or audio flowing in real time with the monitoring result of the same video and / or audio by the video monitoring device, and uses it as a threshold Including feedback. "Interrelated" means that adjusting one parameter affects the other parameter. The "above the threshold" includes both cases where the threshold is exceeded and below the threshold.

According to the present invention, it is possible to provide a method of adjusting an image monitoring apparatus for appropriately detecting an image and / or an error generated according to content in a digital image signal. Also, according to the present invention, it is possible to provide an image monitoring apparatus which detects an image and / or an error properly through learning in a digital image signal.

FIG. 2 is a block diagram showing an image monitoring apparatus 100. FIG. 6 is a diagram showing an example of inspection items in the video / audio monitoring unit 103 displayed on the display unit 107. FIG. 6 is a diagram showing an example of alarm information displayed on a display unit 107. It is a figure which shows the example of the parameter for every monitoring item displayed on the display part 107. FIG. It is a figure which shows the example which put together the correct item alarm which generate | occur | produced when error detection was performed using the same threshold value, an unnecessary alarm, and a slip through for every monitoring item and content. It is a figure which shows the list of an example ((a) before an update, (b) after an update) of several groups of the threshold-value at the time of performing such update several times in the inspection item freeze regarding a certain content.

An image monitoring apparatus according to the present embodiment and an adjustment method thereof will be described with reference to the drawings. FIG. 1 is a block diagram showing an image monitoring apparatus 100. As shown in FIG. In FIG. 1, a video monitoring apparatus 100 inputs a main input unit 101 for inputting a video signal (hereinafter referred to as a video signal to be inspected) corresponding to video and / or audio to be monitored, and a video signal for inspection. Sub-input unit 102, video / audio monitoring unit (determination unit) 103 incorporating memory for storing threshold, video / audio clip storage unit (video storage unit) 104, alarm output unit 105, internal memory (threshold A parameter optimization learning unit (learning unit and updating unit) 106 including a storage unit 106a, a display unit 107 such as a display on which the information displayed by the supervisor MN can be viewed, and the supervisor MN can input information And a supervisor input unit 108 such as a keyboard.

The monitoring operation of the video monitoring device 100 will be described. FIG. 2 is a diagram showing an example of inspection items in the video / audio monitoring unit 103 displayed on the display unit 107. As shown in FIG. The video signals to be monitored are video signals of all formats such as an SDI signal, a file, an IP format, and HDMI (registered trademark). The supervisor MN changes the inside of the box corresponding to each monitoring item via the supervisor input unit 108 while looking at the monitoring item displayed on the display unit 107, and performs “inspection” or “off” for each monitoring item. You can choose one of these. The inspection item for which the supervisor has selected “inspection” is to be monitored by the video monitoring apparatus 100, but the inspection item for which “off” is selected is not monitored.

The monitoring items related to the image include "freeze", "black out", "block freeze", "block black out", "block noise", "red blink", "brightness blink", "scene change", "image reverse" There are “line noise”, “cut point abnormality”, and “time code discontinuous”. On the other hand, monitoring items relating to audio include "mute", "cut-off mute", "audio pop noise", "sound skipping", "voice noise", "loudness" and "true peak". The inspection items are not limited to these.

As a premise of the monitoring operation, it is assumed that in the video / voice monitoring unit 103, a plurality of thresholds (a threshold group, the details will be described later) are set for each monitoring item. Referring to FIG. 1, when the video monitoring apparatus 100 receives an inspection target video signal from the main input unit 101, an image frame, pixel data, and audio sampling data are extracted from the input inspection target video signal. The parameter is transmitted to the voice monitoring unit 103, where parameters corresponding to various sensitivity adjustments and durations that can be applied to the monitoring algorithm are calculated.

According to the monitoring algorithm, the video / voice monitoring unit 103 compares a plurality of obtained parameters with a threshold value corresponding to the monitoring item, and when all the parameters exceed the threshold value, an error corresponding to the monitoring item is detected. It judges and outputs the alarm signal matched with the said monitoring item to the alarm output part 105. FIG. Since the alarm output unit 105 inputs alarm information including the error occurrence time, the content of the detected error, and the severity of the error to the display unit 107 according to the input alarm signal, the display unit 107 receives the alarm information. It can be displayed and viewed by the observer MN. Also, the video / audio monitoring unit 103 cuts out video signals before and after the parameter exceeds the threshold and stores the video signals in the video / audio clip storage unit 104.

FIG. 3 is a diagram showing an example of alarm information displayed on the display unit 107. As shown in FIG. In FIG. 3, the lower part of the screen shows the detected error and content in chronological order, and the upper part of the screen shows "category" indicating that the error is video or audio, and the level of error is normal or severe. The "class" indicating, the "inspection item" indicating the type of error, and the "number of occurrences of error" are displayed together. When the supervisor MN selects any error via the supervisor input unit 108, the video / audio clip storage unit 104 reads out the corresponding video signal before and after the error, and the display unit 107 It is input. Since the video and / or the sound before and after the error are output from the display unit 107 by this, the supervisor MN can actually visually recognize or listen to the content of the error.

By the way, when the alarm is issued for the error that should be issued by the observer MN comparing the content of the detection error displayed on the display unit 107 with the actual video and / or audio, the correct answer (correct answer If an alarm is issued for an error that should not issue an alarm (an unnecessary alarm) or if an alarm is not issued for an error that should issue an alarm Can be recognized respectively. Essentially, it is ideal that all alarms issued when the image monitoring apparatus 100 detects an error are all correct alarms, but in reality, unnecessary alarms and passing through occur. This results from the fact that the error detection standard by machine monitoring of the video / voice monitoring unit 103 does not exactly match the error detection standard by monitoring of the supervisor MN. Therefore, unless the error detection by the machine monitoring approaches the error detection by the human monitoring, the monitoring by the image monitoring apparatus 100 alone becomes difficult.

Therefore, in order to bring the error detection standard by machine monitoring of the video / voice monitoring unit 103 closer to the error detection standard by monitoring of the supervisor MN, the threshold value of the parameter set for each monitoring item is changed. This is called threshold tuning. Here, when there are m monitoring items, assuming that the number of parameters is I (m), the number of threshold values to be tuned is 1 · I (1) + 2 · I (2) +. It can be seen that I (m) increases as the number of monitoring items increases.

FIG. 4 is a diagram showing an example of parameters for each monitoring item displayed on the display unit 107. As shown in FIG. The supervisor MN can now input an arbitrary numerical value for each parameter in the box corresponding to the parameter via the supervisor input unit 108 while looking at the parameter displayed on the display unit 107. There is.

For example, in response to the inspection item "freeze", 4 parameters "sensitivity threshold (activity)", "sensitivity threshold (noise)", "time threshold (start)", and "time threshold (end)" There is one. "Graph scale" represents the scale of the graph for display, and is not a parameter here. That is, in order to detect a video freeze as an error, thresholds (threshold groups) of four parameters must be set appropriately. However, "sensitivity threshold (activity)" and "sensitivity threshold (noise)" are upper and lower limit values of the variance for each small block included in one video frame as a parameter, and they are mutually related. It can be said that they fit. Such parameters are disclosed in detail in WO 2015-059782. Also, “time threshold (start)” and “time threshold (end)” indicate the length of a period during which it is determined that the video is phrased, and they are mutually related. Therefore, when one of the thresholds is adjusted, it is impossible to appropriately detect the freeze unless the other threshold is also changed.

Here, if an appropriate threshold value is not input for all inspection items, unnecessary alarms or omissions at the time of error detection are caused. However, as shown in FIG. 4, when the number of inspection items is large, all inspection items It is difficult to determine an appropriate threshold for Therefore, it is desirable to determine a default threshold (initial value) in advance. Such a default threshold can be stored and used, for example, in the built-in memory of the video / audio monitoring unit 103. However, according to the study results of the inventor, when the same threshold value is used for all the content, the unnecessary alarm and the passing through are reduced for one content, but the unnecessary alarm and the passing occurs for another content. I understand.

FIG. 5 is a diagram showing an example in which the correct answer alarm, the unnecessary alarm, and the slip-through that occur when the error detection is performed using the same threshold are summarized for each monitoring item and content. From the example of FIG. 5, it can be seen that the occurrence frequency of the correct alarm, the unnecessary alarm, and the slip-in differs for each content. Based on the examination result, the inventor has found that it is sufficient to determine the default threshold according to the content of the content. Such default thresholds can be determined from simulations and accumulated experience.

On the other hand, if the default threshold is determined according to the content of the content, there is a possibility of reducing unnecessary alarms and passing through to some extent, but it is not always optimal. Even if the contents of the content are the same, the frequency of unnecessary alarms and passing through may change depending on the reception state and the like. Therefore, in parallel with the monitoring operation, by causing the video monitoring apparatus 100 to learn about the currently input video signal, it is determined whether the default threshold is appropriate or not, and it is preferable to further update it if it is not appropriate. It can be said. This makes it possible to increase the number of correct alarms and reduce the frequency of unnecessary alarms and passing through. The updated threshold may be newly set as the default threshold of the content.

(Learning mode)
The learning function of the video surveillance device 100 will be described below. In the following learning example, it is possible to determine which is better for the case where there are threshold candidates to be changed with respect to the default threshold. In the video monitoring apparatus 100 of FIG. 1, while the video and / or audio monitoring operation is actually performed, the supervisor MN sets a predetermined learning period from the supervisor input unit 108. Then, the video surveillance device 100 can perform learning during this learning period.

Specifically, the monitoring operation is performed using the default threshold value for one monitoring item in M video signals of the same type of content. First, an error detected by the image monitoring apparatus 100 is displayed on the display unit 107 as shown in FIG. 3 during a learning period (time t1 to t2, for example, a time unit such as hours, days, weeks, or months). . Since the errors displayed as shown in FIG. 3 may include inappropriate ones, it is necessary to check whether the errors are appropriate for learning.

Here, the supervisor MN reads the video signals before and after the error stored in the video / audio clip storage unit 104 by designating one of the errors displayed in FIG. Video and / or audio can be viewed on the display unit 107. As a result, the supervisor MN can count the number of correct alarms generated during the time t1 to t2 and the number of unnecessary alarms. Since the video signal before and after each error stored in the video / audio clip storage unit 104 has a length of about 3 seconds, it does not take a long time to view, and this check is completed in a short time. There is little burden.

However, in the above-mentioned check, the number of passing can not be counted. Therefore, the video signal for inspection whose content and occurrence point (time) of the error prepared beforehand are known is input to the video / audio monitoring unit 103 via the sub input unit 102, and the same threshold as described above The video / audio monitoring unit 103 detects an error using As a result, when alarm information similar to that shown in FIG. 3 is obtained, the supervisor MN finds the number of correct alarms and the number of slips by collating the contents of the known error with the occurrence part. be able to. Here, the number of passing is extracted and used. Since the video signal for inspection has a length corresponding to a viewing time of several tens of minutes for each content, it does not take time for inspection. As described above, the supervisor MN learns from the supervisor input unit 108 the parameter optimization learning from the number of correct alarms, the number of unnecessary alarms, and the number of slips calculated for each of the M video signals. Input to the part 106.

Here, let T (m) be the number of correct alarms (first event) in the m-th video signal, F (m) be the number of unnecessary alarms (second event), and let the number of slips (third event) be Assuming that H (m), the parameter optimization learning unit 106 obtains an evaluation value A (m) according to the following evaluation function.

However,
Wt (m) is the correct alarm weighting Wf (m) is the unnecessary alarm weighting Wh (m) is the passing weight

Generally, in error detection, it is preferable that the number of correct alarms be large, and it be preferable that the number of unnecessary alarms be small. Furthermore, it is preferable to minimize passing through. Therefore, it is preferable to set Wt (m)> 0> Wf (m)> Wh (m) as the weighting stored in the memory 106a. For example, Wt (m) = + 3 and Wf (m) = − 1. Let Wh (m) = -5. However, the weighting may be fixed for the same type of content. According to this evaluation function, the higher the obtained evaluation value A (m), the more accurate the error detection, and the lower the value, the more inaccurate the error detection.

Next, the supervisor MN replaces the default threshold value with the threshold value desired to be changed via the supervisor input unit 108, and executes the same monitoring operation as above using the video signal of the same content. The evaluation value is determined. If the evaluation value of the monitoring operation using the default threshold is equal to or more than the evaluation value of the monitoring operation using the threshold desired to be changed, the parameter optimization learning unit 106 continues using the default threshold in the content. As a thing, do not change the stored default threshold. On the other hand, if the evaluation value of the monitoring operation using the default threshold is lower than the evaluation value of the monitoring operation using the threshold for which change is desired, the parameter optimization learning unit 106 determines the default threshold for the content. It is determined that the change is necessary, and the change is replaced with a desired threshold, that is, learning is performed. Thereby, the accuracy of error detection can be further enhanced.

FIG. 6 is a diagram showing a list of examples of a plurality of threshold values when such an update is performed a plurality of times in an inspection item regarding a certain content: freeze, ((a) before update, (b) after update) . A plurality of sets of threshold value groups shown in FIG. 6 are listed up and stored in the built-in memory 106a, displayed on the display unit 107 as needed, and can be confirmed by the supervisor MN.

According to the pre-update list of FIG. 6A, the default threshold group is currently listed at the top with priority 1 but its evaluation value A (m) is “85”. On the other hand, the evaluation value A (m) of the threshold group newly evaluated as the first candidate is "92", and the evaluation value A (m) of the threshold group newly evaluated as the second candidate is "81" Suppose that there was.

In this case, since the evaluation value A (m)-"92" of the first candidate threshold group is higher than the evaluation value A (m) "85" of the default threshold group so far, the parameter optimization learning unit 106 However, as shown in FIG. 6B, by setting the priority to 1 by updating, it is replaced as a new default threshold group and transmitted to the video / audio monitoring unit 103 for monitoring. By this, the default threshold group so far will be carried back to priority 2.

On the other hand, since the evaluation value A (m) “81” of the second candidate threshold group is lower than the evaluation value A (m) “85” of the replaced default threshold group, the parameter optimization learning unit 106 However, by setting the priority to 3 by updating, it will be listed in lower order. The list is stored and used in the memory 106 a of the parameter optimization learning unit 106. Thereby, the threshold value group of high evaluation value A (m) can always be made into the default threshold value group regarding the said content. However, if the content is different, even if the threshold value group uses the same value, the priority may be different. Note that such updating of the threshold value group may be performed automatically by the parameter optimization learning unit 106, or may be performed after waiting for the supervisor MN's permission. Alternatively, regardless of the value of the evaluation value A (m), the threshold designated by the supervisor MN via the supervisor input unit 108 can be set as the default threshold.

It should be noted that the first candidate, the second candidate,... Are determined, but each time the supervisor MN inputs an arbitrary value from the supervisor input unit 108 to the parameter optimization learning unit 106, the correct answer is obtained each time The evaluation value may be calculated after obtaining the number of alarms, the number of unnecessary alarms, and the number of by-passes. Alternatively, the parameter optimization learning unit 106 individually changes the threshold by + 5% or -5% with respect to the default threshold group, and in each case, the number of correct alarms, the number of unnecessary alarms, and the bypass The evaluation value may be calculated after finding the number of. Evaluation values can be similarly obtained for other monitoring items.

The video signal for inspection may be input between the input of the monitoring target video signal, or may be performed in parallel with the input of the monitoring target video signal to perform inspection in the background. The same threshold can then be used to perform error detection. Furthermore, the video / audio storage clip unit 104 cuts out the monitoring target video signal input from the main input unit 101 over a predetermined length at a predetermined timing, and associates it with the monitoring result performed by the supervisor MN. Alternatively, it may be stored as a video signal for inspection in a memory (not shown). The stored test video signal is input from the sub input unit 102 at a necessary timing, and is used for threshold updating as described above.

According to the present invention, it is possible to provide a method of adjusting a video monitoring apparatus for appropriately detecting a video and / or an error generated according to content in a digital video signal, and in the digital video signal, the video and / or the error are properly learned It is possible to provide an image monitoring device that detects

100 video monitoring apparatus 101 main input unit 102 secondary input unit 103 video / audio monitoring unit 104 video / audio clip storage unit 105 alarm output unit 106 learning unit for parameter optimization built-in memory 107 a display unit 108 monitor unit 108 monitor input unit MN monitor

Claims

A method of adjusting an image monitoring apparatus, which receives an image signal corresponding to an image and / or an audio to be monitored and detects that an error has occurred in a monitoring item,
The video monitoring apparatus is configured to determine that the error has occurred when all of a plurality of parameters that change according to the video signal exceed a threshold determined according to the monitoring item. , At least two of the plurality of parameters are mutually related,
The threshold values of the plurality of parameters are respectively determined to create a plurality of threshold value groups, which are stored in a memory,
A method of adjusting a video surveillance apparatus, comprising selecting any set of the threshold group stored in the memory according to contents of video and / or audio to be monitored.
The video monitoring device performs learning in parallel with the video and / or audio monitoring operation, updates at least one threshold of the threshold group based on the learning result, and stores the updated threshold group in a memory The adjustment method of the image | video monitoring apparatus of Claim 1 characterized by the above-mentioned.
The video surveillance device issues an alarm when it detects an error.
In the learning, a video signal corresponding to the video and / or audio to be monitored is input, and the alarm is issued at predetermined time intervals, the first event of the alarm for the error to be issued Count the number of events and the number of second events for which the alarm has been issued for errors that should not be issued,
Furthermore, a video signal for inspection whose content and place of occurrence of the error are known is input to the video monitoring apparatus, and the number of third events for which the alarm has not been issued for an error that should be issued. Count,
An evaluation value is determined by weighting and calculating the number of obtained first events, the number of second events, and the number of third events, and the threshold group is determined based on the determined evaluation values. The method of adjusting an image monitoring apparatus according to claim 2, wherein
4. The method according to claim 3, wherein in the learning, the number of the first event, the number of the second event, and the number of the third event are counted while changing the thresholds of the threshold group individually. The adjustment method of the image | video monitoring apparatus of description.
In an image monitoring apparatus that receives an image signal corresponding to an image and / or audio to be monitored and detects that an error has occurred in a monitoring item,
A threshold storage unit that stores a plurality of parameters that change according to the video signal, wherein at least two of the plurality of parameters are associated with each other;
A determination unit that determines that the error has occurred and issues an alarm when all of the plurality of parameters exceed the threshold determined according to the monitoring item;
A learning unit that performs learning in parallel with the video and / or audio monitoring operation;
An updating unit that updates the threshold based on a learning result in the learning unit;
A video storage unit for storing video signals before and after the occurrence of the error when the determination unit detects the occurrence of the error;
The image monitoring apparatus characterized in that the learning unit learns by analyzing the notification result of the error.
When the learning unit inputs a video signal corresponding to video and / or audio to be monitored, the alarm is issued for an error that should issue an alarm at predetermined time intervals. Count the number of events and the number of second events for which an alarm should have been issued for errors that should not have triggered an alarm. Furthermore, the video signal for inspection whose content and location of the error are known The number of third events for which an alarm should not be issued when the input is made is counted, and the number of the first event obtained and the number of the second event Calculating an evaluation value by weighting and calculating the number of cases of the third event, and ranking the threshold values stored in the threshold storage unit based on the obtained evaluation values. Video director described in 5 Apparatus.
The learning unit is characterized by counting the number of first events, the number of second events, and the number of third events while individually changing the threshold value of the monitoring item. The video surveillance device described in.
The learning unit is configured to generate an alarm for the m-th video signal of the M video signals during a period from time t1 to time t2. Let T (m) be the number, and let F (m) be the number of second events for which an alarm should have been issued for errors that should not have triggered the alarm. The image monitoring apparatus according to claim 6 or 7, wherein an evaluation value A (m) is determined according to the following evaluation function, where H (m) is the number of third events for which no event has been reported. .

However,
Wt (m) is the weight of the first event Wf (m) is the weight of the second event Wh (m) is the weight of the third event
The video monitoring apparatus according to any one of claims 6 to 8, wherein the updating unit lists up the threshold for each of the monitoring items in accordance with the priority.
The video monitoring apparatus according to any one of claims 5 to 9, wherein the video storage unit cuts out and stores the input video signal at a predetermined timing at a predetermined length.
The display device according to any one of claims 5 to 10, further comprising a display unit for displaying the learning result of the learning unit and reproducing video and / or audio based on the video signal before and after the occurrence of the error. Video surveillance device.