CN103377479A - Event detecting method, device and system and video camera - Google Patents
Event detecting method, device and system and video camera Download PDFInfo
- Publication number
- CN103377479A CN103377479A CN 201210128919 CN201210128919A CN103377479A CN 103377479 A CN103377479 A CN 103377479A CN 201210128919 CN201210128919 CN 201210128919 CN 201210128919 A CN201210128919 A CN 201210128919A CN 103377479 A CN103377479 A CN 103377479A
- Authority
- CN
- China
- Prior art keywords
- zone
- event
- video
- scene
- zones
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an event detecting method, device and system and a video camera. The event detecting method includes the steps that an image sequence formed by an adjacent set of images on a time axis is obtained from the video of a monitoring scene; the moving feature of each zone among a plurality of zones in a video scene corresponding to the monitoring scene, is extracted; zone event detection is conducted on the zones according to at least one moving feature of the zones; overall event detection is conducted on the video scene according to zone event detection results of the zones, wherein the sizes of physical spaces, corresponding to the zones, in the monitoring scene are the same. According to the scheme, the event detecting method, device and system and the video camera have universality of application scenes and precision of event judging at the same time.
Description
Technical field
The present invention relates to video analysis and understanding field, more specifically, relate to a kind of methods, devices and systems and a kind of video camera that monitor video is carried out event detection.
Background technology
Event detection is an important technology in video content analysis and the understanding.The intellectual analysis of monitor video is used has strong demand to automatic event detection technology.Compare with the video of general content the unknown, monitor video have that scene is fixed and usually in advance as can be known, pay close attention to the characteristics such as event category is limited, definition clear-cut.
The event that the object with in the scene of paying close attention in the video surveillance applications is relevant can be divided into two classes substantially, and a class may be summarized to be and has the type event, the another kind of sports type event that then is summarised as.Exist the type event to comprise: the entering/occur and leave of the particular type object in the specific region/disappear, perhaps the particular type object enters/leaves the specific region along specific direction, and perhaps particular type object life period in the specific region satisfies specific condition etc.This class event often utilizes the methods such as target detection and tracking to solve in existing scheme, and has obtained preferably actual effect.The definition of sports type event is then mainly for the motor pattern of object in the scene (such as people, vehicle, animal etc.), such as the events such as falling, fight, run that occurs in the monitoring scene.
Sports type event detection in the monitor video general main comprise two core procedures, i.e. feature extraction and pattern classification.In order to obtain higher accuracy in pattern classification step, the feature that requires characteristic extraction step to extract has the good property distinguished, and namely the feature of similar events as has higher similarity, and the feature of different event has obvious difference.Yet because the position that the scene of video monitoring is different, video camera sets up the position difference, event occurs in scene is different, the feature that general characteristic extraction step is extracted is difficult to guarantee that the feature of similar events as has higher similarity.The result that this problem causes is exactly: carry out the training of event detection in some existing scenes, and the scheme of the model of application training gained is difficult to obtain gratifying classification accuracy rate in pattern classification step in new scene.Existing scheme often adopts trains respectively each new monitoring scene, and the disaggregated model that utilizes specialized training to obtain comes new monitoring scene is carried out having addressed the above problem since the event detection.But this itself has increased the initialized complexity of video monitoring system.And because training sample is limited, such scheme causes often not comprising in the training sample or only comprising the anomalous event of few generation of paying close attention to the most in particular event sample, the especially video monitoring of minute quantity.And the shortage of particular event sample directly causes the decline of pattern classification accuracy.Although one-class support vector machines (One-class SVM) sorter that only needs non-anomalous event sample to participate in training that adopts in the existing scheme has been evaded the problem that the particular event sample lacks, can't guarantee fundamentally that still classification accuracy rate reaches the level in the sufficient situation of sample.
Summary of the invention
Provided hereinafter about brief overview of the present invention, in order to basic comprehension about some aspect of the present invention is provided.Should be appreciated that this general introduction is not about exhaustive general introduction of the present invention.It is not that intention is determined key of the present invention or pith, neither be intended to limit scope of the present invention.Its purpose only is that the form of simplifying provides some concept, with this as the in greater detail preorder of discussing after a while.
The purpose of this invention is to provide a kind of event detecting method, device and system, and video camera, with in the above-mentioned technical matters that solves prior art at least one.
According to an aspect of the present invention, provide a kind of event detecting method.Described event detecting method comprises: obtain from the video of monitoring scene by one group of adjacent on time shaft image sequence that image forms; In described image sequence, extract each the regional motion feature in a plurality of zones in the video scene corresponding with described monitoring scene; At least according to each regional motion feature zone issue being carried out in described zone detects; And according to the zone issue testing result in described a plurality of zones, described video scene is carried out clobal detect, wherein, the physical space in the corresponding described monitoring scene in described a plurality of zones measure-alike.
Preferably, described event detecting method also comprises: select the one group local classifiers corresponding with the shooting angle of pitch in described zone from many groups local classifiers; And at least motion feature in described zone inputted selected local classifiers, described zone is carried out the event classification.
According to a further aspect in the invention, provide a kind of event detection device.Described event detection device comprises: image sequence obtains parts, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene; Characteristic extracting component is used for each the regional motion feature in a plurality of zones in the described image sequence extraction video scene corresponding with described monitoring scene; The zone issue detection part is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And the clobal detection part, be used for the zone issue testing result according to described a plurality of zones, described video scene is carried out clobal detect, wherein, the physical space in the corresponding described monitoring scene in described a plurality of zones measure-alike.
According to a further aspect in the invention, provide a kind of event detection system.Described event detection system comprises: video camera, for the video that obtains monitoring scene; And event detection device, be used for carrying out event detection at described video.Described event detection device comprises: image sequence obtains parts, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene; Characteristic extracting component is used for each the regional motion feature in a plurality of zones in the described image sequence extraction video scene corresponding with described monitoring scene; The zone issue detection part is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And the clobal detection part, be used for the zone issue testing result according to described a plurality of zones, described video scene is carried out clobal detect, wherein, the physical space in the corresponding described monitoring scene in described a plurality of zones measure-alike.
According to a further aspect in the invention, provide a kind of video camera.Described video camera comprises: image acquisition component, for the video that obtains monitoring scene; And the event detection parts, be used for carrying out event detection at described video.Described event detection parts comprise: image sequence obtains parts, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene; Characteristic extracting component is used for each the regional motion feature in a plurality of zones in the described image sequence extraction video scene corresponding with described monitoring scene; The zone issue detection part is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And the clobal detection part, be used for the zone issue testing result according to described a plurality of zones, described video scene is carried out clobal detect, wherein, the physical space in the corresponding described monitoring scene in described a plurality of zones measure-alike.
In described event detecting method, device and system and video camera, the two layers of classified strategy of employing from the part to the overall situation, and the corresponding scene physical space of regional is measure-alike or basic identical, so that the location independent that the detection of Video Events is basic and monitoring scene and event occur in scene, the characteristics of image that can guarantee like this similar events as has higher similarity, thereby can accurately event be detected and classify.The solution of the present invention has versatility and the event Accuracy of Judgement of application scenarios concurrently.
Description of drawings
The present invention can by with reference to hereinafter by reference to the accompanying drawings given description be better understood, wherein in institute's drawings attached, used same or analogous Reference numeral to represent identical or similar parts.Described accompanying drawing comprises in this manual and forms the part of this instructions together with following detailed description, and is used for further illustrating the preferred embodiments of the present invention and explains principle and advantage of the present invention.In the accompanying drawings:
Fig. 1 illustrates the according to an embodiment of the invention indicative flowchart of event detecting method;
Fig. 2 illustrates the synoptic diagram of the shooting angle of pitch in zone;
Fig. 3 illustrates the according to an embodiment of the invention schematic block diagram of event detection device;
Fig. 4 illustrates the according to an embodiment of the invention schematic block diagram of event detection system;
Fig. 5 illustrates the according to an embodiment of the invention schematic block diagram of video camera; And
Fig. 6 illustrates the block diagram of the structure of the computing machine that can realize embodiments of the invention/example.
Embodiment
A kind of comparatively intuitively idea of the problems referred to above of reply prior art is exactly irrelevantization of scene, that is to say the optimization by scheme, to set up position, the event factors such as occurrence positions in scene irrelevant so that the feature of event detection is after treatment with scene, the video camera of video monitoring.If can just can adopt the training of in some existing scenes, carrying out event detection like this, and in new scene the scheme of the model of application training gained; And in known scene nature can gather great amount of samples satisfy train required.
What present a kind of scheme adopted is the design of resolving.This kind scheme at first attempts to detect all targets in the scene, then attempts to analyze behavior and the association of the behavior between the different target of each target.The essence of this design is with the visual information logical symbol, then by event detection is carried out in the classification of the irrelevant logical symbol of these scenes.Yet people's behavioural analysis is as example in the monitor video, and under current technical conditions, detection, attitude and the motion analysis of the people under environment, the visual angle condition are difficult to reach functional need arbitrarily, and be next extremely difficult especially in the situation that existence is blocked.
For the problems referred to above and the above analysis of prior art, the present invention proposes a kind of event detecting method, device and system, and video camera.In the solution of the present invention, adopt the strategy of local and overall two layers of classified, and the corresponding scene physical space of regional is measure-alike or basic identical, so that the basic location independent that in scene, occurs with monitoring scene and event of the detection of Video Events, make prior acquisition great amount of samples carry out sorter and train and become possibility.In addition, select the sorter corresponding with the shooting angle of pitch in this zone to carry out the event classification for each zone, so that the detection of Video Events is basic and the decorating position of video camera is irrelevant.The solution of the present invention has versatility and the event Accuracy of Judgement of application scenarios concurrently.
Embodiments of the invention are described below with reference to accompanying drawings.The element of describing in an accompanying drawing of the present invention or a kind of embodiment and feature can combine with element and the feature shown in one or more other accompanying drawing or the embodiment.Should be noted that for purpose clearly, omitted expression and the description of parts that have nothing to do with the present invention, known to persons of ordinary skill in the art and processing in accompanying drawing and the explanation.
Fig. 1 illustrates the according to an embodiment of the invention indicative flowchart of event detecting method.As shown in Figure 1, in step S110, from the video of monitoring scene, obtain image sequence.This image sequence is carried out event detection, as an event detection result to this video.Here, image sequence by in the described video on time shaft one group of adjacent image form.Can from video, mark off on a time period the image sequence with schedule time length, also can from video, mark off by contained picture number the image sequence of the adjacent image with predetermined number.The time span that image sequence is striden or included picture number can determine according to actual needs, and without limits.In addition, be appreciated that the requirement according to accuracy of detection and detection speed, each image in the image sequence can be one group of continuous in video image, also can be upper adjacent but discontinuous one group of image of time in the video.Preferably, on the time adjacent but in discontinuous one group of image the time interval between the adjacent image equate.
In step S120, in image sequence, extract each the regional motion feature in a plurality of zones in the video scene corresponding with monitoring scene.Monitoring scene is presented on the following video scene that is called as in video pictures.A plurality of zones in the video scene are a plurality of zones in the video pictures.According to the requirement of accuracy of detection and detection speed, for each zone, can extract this regional motion feature in each image in image sequence, that is, each regional motion feature is to use two images to extract; Also can extract this regional motion feature take adjacent a plurality of images as unit in image sequence, that is, each regional motion feature is to use a plurality of images to extract; Perhaps, also can extract take whole image sequence as unit this regional motion feature, that is, each regional motion feature is to use all images in the whole image sequence to extract.
The different impacts that cause in position different for fear of monitoring scene and event occurs in monitoring scene, in event detecting method of the present invention, the physical space in the corresponding monitoring scene in described a plurality of zones measure-alike.Certainly, the size of the physical space in the corresponding monitoring scene in a plurality of zones also can be close or essentially identical.Can use suitable parameter as the size of the regional corresponding physical space in the video pictures.For example, can be with the area of the surface level at the place, ground in the regional corresponding monitoring scene size as this corresponding physical space in zone, also can with in the zone just fully the true altitude of visible upright object as the size of this corresponding physical space in zone, also can with in the zone just the actual diameter of complete visible spherical body as the size of this corresponding physical space in zone, perhaps, also can be simultaneously with two in above-mentioned three parameters or all as the size of this corresponding physical space in zone.
In a definite monitoring scene, the zone in video scene be distributed in the event detection system initialization time determine, after this do not change with what, the change of size, motor pattern of moving target in the video.
Here, motion feature can be any feature that motion is had description power.In addition, can adopt existing any suitable method to extract each regional motion feature.As example rather than restriction, below provide the feature extraction example that is characterized as example with motion vector histogram and light stream statistic.
In one example, extract the histogram of the part or all of motion vector in each zone as this regional motion feature.Particularly, adopt following steps to extract each regional motion vector histogram:
1) motion vector of the part or all of block of pixels in this zone in one or several image (motion vector) in the sequence of computed images;
2) if motion vector is the value under the rectangular coordinate system, then this intra-zone is divided or all the motion vector of block of pixels be transformed in the polar coordinate system;
3) in polar coordinate system the motion vector of block of pixels is mapped on one of a plurality of angular ranges by angle, each angular range is corresponding to a direction; And
4) add up to the amplitude that is mapped to the motion vector on each direction, form this regional motion vector histogram.
In another example, extract the statistic feature of the part or all of light stream in each zone as this regional motion feature.Particularly, adopt following steps to extract each regional light stream statistic feature:
A) light stream of the part or all of pixel in this zone in one or several image (optical flow) in the sequence of computed images;
B) according to the light stream in this zone in one or several image, energy feature (but the definition list of references N.Ding of energy feature of calculating section or whole pixels, Y.Chen, et al, Energy-based surveillance systems for ATM machines, Proceedings of the8th World Congress on Intelligent Control and Automation, 2010);
C) scope according to energy feature marks off several energy levels; And
D) to the part or all of pixel in this zone, add up respectively the pixel count of each energy level, form the statistic feature of this regional light stream.
For the motion feature that makes extraction more accurate to the description of the motion in the zone, according to another embodiment of the present invention, extract the motion feature of a plurality of subregions in each zone, and the motion feature that makes up described a plurality of subregions is as this regional motion feature.For example, can the motion feature of a plurality of subregions be combined into by forms such as series winding, weighted sums the motion feature in zone corresponding to described a plurality of subregion.With zone similarity, subregion also is ready-portioned in advance.
From image sequence for the motion feature of each extracted region may be multidimensional.According to one embodiment of present invention, carry out before zone issue detects in the provincial characteristics of utilizing described below, the motion feature that each is regional is mapped as respectively the lower motion feature of dimension, the difficulty that detects to reduce zone issue.In zone issue detecting step subsequently, according to the motion feature behind the dimensionality reduction zone issue is carried out in each zone and detect.As example rather than restriction, can be mapped as respectively the lower motion feature of dimension with clustering methodology or the principal component analysis (PCA) motion feature that each is regional.
Continuation, in step S130, is carried out zone issue according to each regional motion feature to the zone at least and is detected after having extracted regional motion feature with reference to figure 1.That is to say that the motion feature that can utilize the same area in the regional or a plurality of images in the image in described image sequence to extract detects this zone contingent event in the time period that described image sequence is striden.Whether the result of territorial classification certain or some events has occured in the zone, or the probability etc. of certain or some events occurs in the zone.Can use suitable existing event detecting method that event detection is carried out in the zone.
Should be appreciated that not only can carry out zone issue to the zone according to each regional motion feature detects, and can also according to each regional motion feature other features with each zone, carry out zone issue to the zone and detect.For example, utilize the features such as number in the zone that people's technology such as detection, face and head detection extract, people's posture and the position in image, size, be to the significant information of event detection, can together input local classifiers with motion feature, be used for improving classification accuracy rate.
In one embodiment of the invention, further contemplate the impact that will avoid be used to the decorating position of the video camera that obtains monitor video, carry out zone issue by following steps and detect:
I) from many groups local classifiers, select the one group local classifiers corresponding with the shooting angle of pitch in zone; And
Ii) motion feature in zone is inputted selected local classifiers, this zone is carried out the event classification.
The shooting angle of pitch in zone refers to center and the line of video camera photocentre and the angle between this surface level of the surface level at the place, ground in the regional corresponding monitoring scene in the video scene.Understand for convenient, Fig. 2 illustrates the synoptic diagram of the shooting angle of pitch in zone.In Fig. 2,4 ground region R1, R2, R3 and R4 in the monitoring scene correspond respectively to the respective regions in the video scene.Their central point is respectively P1, P2, P3 and P4.C is video camera.Angle A1 illustrates video camera to the shooting angle of ground region R1, i.e. the shooting angle of pitch in the zone in the video scene corresponding with ground region R1.Angle A2 illustrates video camera to the shooting angle of ground region R2, i.e. the shooting angle of pitch (shooting angle of R3 and R4 is not shown) in the zone in the video scene corresponding with ground region R2.
Similarly, in this embodiment, when with other features in this zone the zone issue detection is carried out in the zone according to each regional motion feature, the motion feature in zone is inputted selected local classifiers with other features, this zone is carried out the event classification.
Described many group local classifiers can be to utilize the good sorter of great amount of samples image sequence training in advance.Be not described in detail the training process of sorter at this, in order to avoid unnecessarily fuzzy the present invention.
In described many group local classifiers, every group of local classifiers is used for that the shooting angle of pitch is in a zone in the continuous angle of pitch interval and carries out the event classification.Every group of local classifiers comprises at least one local classifiers.Each local classifiers in every group of local classifiers is at least a event type, such as fighting, fall or run etc.
Select the sorter corresponding with the shooting angle of pitch in this zone to carry out the event classification for each zone, strengthened the separability between the different event, so that the detection of Video Events is basic and the decorating position of video camera is irrelevant, further improved the accuracy rate of event detection.
Continuation, according to the zone issue testing result in a plurality of zones, is carried out clobal to monitoring scene and is detected in step S140 with reference to figure 1.The result that clobal detects, exactly in the video scene within the time period that described image sequence is striden the testing result of contingent event.
When clobal detects, the zone issue testing result of All Ranges all can be input in the global classification device, video scene is carried out an event classification.The spatial correlation of the event of considering in adjacent area also can be input to the zone issue testing result of the part adjacent area in a plurality of zones in the global classification device, to carry out the event classification.That is to say, be input in the global classification device after the zone issue testing result can being made up by regional neighbouring relations.Particularly, can be with the part adjacent area compositing area group in a plurality of zones, and the zone issue testing results of one or more zones group are input to respectively in the global classification device, to carry out the event classification.If the zone issue testing result of a plurality of zones group is input to respectively in the global classification device, the multiple events that then according to the zone issue testing result of a plurality of zones group video scene is carried out is classified resulting a plurality of event classification results all as the clobal testing result.
This local event of carrying out first detects the two layers of classified strategy that carries out again the clobal detection, the less characteristics of the event correlation between the zone of apart from each other in the short period section have been utilized in the scene, carry out again comprehensively event detection being there is no negative effect after regional processed respectively, further, the corresponding physical space of regional is measure-alike in the video scene, thereby so that the detection of Video Events can be substantially and the location independent that occurs in scene of monitoring scene and event.
According to another embodiment of the present invention, described event detecting method can also comprise the time domain post-processing step, to utilize the time domain relevance of event that the clobal testing result is optimized after clobal detects.Particularly, can according in the video on time shaft the clobal testing result of one or more other image sequences adjacent with the specific image sequence proofread and correct the clobal testing result of described specific image sequence.For example, when the clobal testing result of the clobal testing result of specific image sequence and former and later two or a plurality of image sequences that are adjacent at time shaft is all inconsistent, get the clobal testing result of adjacent image sequence as the clobal testing result of this specific image sequence, remove thus some classification errors.Again for example, video can be divided into uniform section in time domain, in each video-frequency band, utilize the clobal testing result of the image sequence that this section comprise to vote, determine the event detection result of this video-frequency band, thereby eliminate the impact of the classification error of individual image sequence.In addition, can also add up the state transition probability of clobal testing result on time domain, the association of the clobal testing result of image sequence/video-frequency band before and after utilizing on hidden Markov model (Hidden Markov Model) the technology mining time domain, come the clobal testing result of filtering image sequence/video-frequency band on time domain, reach the effect that improves accuracy of detection.
As indicated above, a plurality of zones in the video scene can be ready-portioned in advance.In one embodiment of the invention, before obtaining image sequence, video scene is divided into a plurality of zones, and determines the shooting angle of pitch in each zone.This takes full advantage of the known characteristics of monitoring scene.
Be to be understood that, in the event detecting method according to the embodiment of the invention, be used for the employed sample of parts (for example territorial classification device) that zone issue detects in training, known with the size (for simplicity, hereinafter referred to as sample physics bulk) of physical space in the regional corresponding monitoring scene in the video scene.When stating event detecting method on the implementation, the size of the physical space in the regional corresponding monitoring scene (application scenarios) in the video scene is identical with sample physics bulk.According to above-mentioned situation, can determine the size in the zone in the known video scene of corresponding physical space size and take the angle of pitch with existing any suitable method.
As example rather than restriction, can the usage monitoring scene in the known bar of the known spheroid of diameter and length or the combination of line determine the size in the zone in the video scene and take the angle of pitch, thereby in video scene, mark off the zone according to the size in zone.As another kind of example rather than restriction, when the physics scene has when not having barrier on smooth ground and the ground, can determine zone in the video scene by the mode of laying on the ground grid pattern.
In order fully to observe moving target (such as people, vehicle, animal etc.) thus the motion in video scene detects event more accurately, the size that each is regional is so that as seen moving target can be complete in the zone when being in the corresponding physical space in described zone.
Be appreciated that more near the zone in the corresponding video scene of the physical space of video camera just greatlyr, vice versa.That is to say that although the regional corresponding physical space in the video scene is measure-alike or substantially the same, the size in zone is not quite similar in the video.Therefore, in according to one embodiment of present invention, carrying out according to the size in a plurality of zones, the amplitude of each regional motion feature being carried out respectively normalization before zone issue detects, with the movable information of reflecting regional more accurately.
In addition, event may occur in zone and the zone part of joining, therefore, can have between the zone of dividing overlapping, in order to can detect more accurately in the edges of regions event.In other words, two or more zones in the zone of dividing can overlap each other.
In addition, because the shooting angle of video camera is different, a plurality of zones that mark off may partly or wholly cover video scene.For the zone that does not need in the video scene to monitor, can manually specify its zone that needn't be divided out to cover, to save unnecessary processing.
In addition, when the zoning, can be a plurality of subregions with regional Further Division also, so that the motion feature that follow-up characteristic extraction step is extracted is more accurate to the description of the motion in the zone.
Fig. 3 illustrates the according to an embodiment of the invention schematic block diagram of event detection device.As shown in Figure 3, event detection device 300 comprises that image sequence obtains parts 310, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene; Characteristic extracting component 320 is used for each the regional motion feature in a plurality of zones in the image sequence extraction video scene corresponding with monitoring scene; Zone issue detection part 330 is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And clobal detection part 340, be used for the zone issue testing result according to described a plurality of zones, video scene is carried out clobal detect, wherein, the physical space in the corresponding monitoring scene in described a plurality of zones measure-alike.
Preferably, zone issue detection part 330 can comprise: sorter alternative pack (not shown) is used to each zone to select the one group local classifiers corresponding with the shooting angle of pitch in described zone from many groups local classifiers; And described local classifiers, at least motion feature that receives respective regions, this zone is carried out the event classification.
Preferably, clobal detection part 340 is configured to receive the zone issue testing result of the Zone Full in described a plurality of zone, or receive the one or more regional zone issue testing result of organizing that the part adjacent area in described a plurality of zone forms, video scene is carried out the event classification.
Preferably, clobal detection part 340 can be realized by the above global classification device of describing in the embodiment of event detecting method.
Preferably, event detection device 300 also comprises the optimization component (not shown), is used for proofreading and correct according to the clobal testing result of video one or more other image sequences adjacent with the specific image sequence on time shaft the clobal testing result of described specific image sequence.
Preferably, event detection device 300 comprises that also the zone divides the parts (not shown), is used for obtaining before parts 310 obtain image sequence at image sequence, video scene is divided into described a plurality of zone, and determines the shooting angle of pitch in each zone.
Preferably, characteristic extracting component 320 also is used for using clustering methodology or the principal component analysis (PCA) motion feature that each is regional to be mapped as respectively the lower motion feature of dimension.
Preferably, characteristic extracting component 320 also is used for the size according to described a plurality of zones, and the amplitude of each regional motion feature of extracting from image sequence is carried out respectively normalization.
Preferably, characteristic extracting component 320 also is used for the motion feature of a plurality of subregions in each zone of extraction; And make up the motion feature of described a plurality of subregions as the motion feature in described zone.
Fig. 4 illustrates the according to an embodiment of the invention schematic block diagram of event detection system.As shown in Figure 4, event detection system comprises video camera 410 and event detection device 420.Video camera 410 is used for obtaining the video of monitoring scene.Event detection device 420 is used for carrying out event detection at video.Event detection device 420 comprises: image sequence obtains parts 421, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene; Characteristic extracting component 422 is used for each the regional motion feature in a plurality of zones in the image sequence extraction video scene corresponding with monitoring scene; Zone issue detection part 423 is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And clobal detection part 424, be used for the zone issue testing result according to described a plurality of zones, video scene is carried out clobal detect.Wherein, the physical space in the corresponding monitoring scene in described a plurality of zone is measure-alike.
Here, video camera 410 can be existing various camera head.Event detection device 420 can be event detection device 300 according to the abovementioned embodiments of the present invention.
According to one embodiment of present invention, event detection device 420 can arrange discretely with video camera 410.For example, event detection device 420 is arranged in the Surveillance center, and the video that video camera 410 collects offers event detection device 420 by wired or wireless mode.
According to another embodiment of the invention, event detection device 420 also can arrange integratedly with video camera 410.For example, event detection device 420 can or be attached to video camera 410 with embeddings such as various forms such as IC chip, DSP, softwares.
Although a video camera 410 in the event detection system 400 only is shown among Fig. 4, it will be understood by those skilled in the art that to comprise more video cameras 410 in the event detection system 400.For example, for monitoring scene is carried out more fully event detection, can comprise a plurality of video cameras 410 in system, the video information of the part of each video camera 410 acquisition monitoring scene is perhaps with the video information of different angles acquisition monitoring scene.
Fig. 5 illustrates the according to an embodiment of the invention schematic block diagram of video camera.As shown in Figure 5, video camera 500 comprises image acquisition component 510 and event detection parts 520.Image acquisition component 510 is used for obtaining the video image of monitoring scene.Event detection parts 520 are used for carrying out event detection at video.Event detection parts 520 comprise: image sequence obtains parts 521, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene; Characteristic extracting component 522 is used for each the regional motion feature in a plurality of zones in the image sequence extraction video scene corresponding with monitoring scene; Zone issue detection part 523 is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And clobal detection part 524, be used for the zone issue testing result according to described a plurality of zones, video scene is carried out clobal detect.Wherein, the physical space in the corresponding monitoring scene in described a plurality of zone is measure-alike.
Here, event detection parts 520 can be the event detection devices 300 according to the embodiment of the invention.Image acquisition component 510 can be existing various image acquisition component.Miscellaneous part in the video camera 500 can be the common components of video camera of the prior art, and is here for brevity, not shown in Figure 5.
In event detecting method, device and system and video camera according to the embodiment of the invention, the two layers of classified strategy of employing from the part to the overall situation, and the corresponding scene physical space of regional is measure-alike or basic identical, so that the basic location independent that in scene, occurs with monitoring scene and event of the detection of Video Events, make prior acquisition great amount of samples carry out sorter and train and become possibility.In addition, select the sorter corresponding with the shooting angle of pitch in this zone to carry out the event classification for each zone, strengthened the separability between the different event, so that the detection of Video Events is basic and the decorating position of video camera is irrelevant.The employing the solution of the present invention solves the sports type event detection problem in the video monitoring, only need to be when system initialization carry out the parameter mark to monitoring scene, then just can utilize the existing sorter that trains to carry out the accurate detection of anomalous event as fighting, fall, run etc.The solution of the present invention has versatility and the event Accuracy of Judgement of application scenarios concurrently.
Should be appreciated that according to each building block, unit in above-mentioned each device of the embodiment of the invention and can be configured by the mode of software, firmware, hardware or its combination.Dispose spendable concrete means or mode and be well known to those skilled in the art, do not repeat them here.In situation about realizing by software or firmware, to the computing machine with specialized hardware structure the program that consists of this software is installed from storage medium or network, this computing machine can be carried out various functions etc. when various program is installed.
Fig. 6 illustrates the block diagram of the structure of the computing machine that can realize embodiments of the invention/example.In Fig. 6, CPU (central processing unit) (CPU) 601 carries out various processing according to the program of storage in the ROM (read-only memory) (ROM) 602 or from the program that storage area 608 is loaded into random access memory (RAM) 603.In RAM 603, also store as required data required when CPU 601 carries out various processing etc.CPU 601, ROM 602 and RAM 603 are connected to each other via bus 604.Input/output interface 605 also is connected to bus 604.
Following parts are connected to input/output interface 605: importation 606 (comprising keyboard, mouse etc.), output 607 (comprise display, such as cathode-ray tube (CRT) (CRT), liquid crystal display (LCD) etc., with loudspeaker etc.), storage area 608 (comprising hard disk etc.), communications portion 609 (comprising that network interface unit is such as LAN card, modulator-demodular unit etc.).Communications portion 609 is processed such as the Internet executive communication via network.As required, driver 610 also can be connected to input/output interface 605.Detachable media 611 can be installed on the driver 610 as required such as disk, CD, magneto-optic disk, semiconductor memory etc., so that the computer program of therefrom reading is installed in the storage area 608 as required.
Realizing by software in the situation of above-mentioned series of processes, such as detachable media 611 program that consists of software is being installed such as the Internet or storage medium from network.
It will be understood by those of skill in the art that this storage medium is not limited to shown in Figure 6 wherein has program stored therein, distributes separately to provide the detachable media 611 of program to the user with equipment.The example of detachable media 611 comprises disk (comprising floppy disk (registered trademark)), CD (comprising compact disc read-only memory (CD-ROM) and digital universal disc (DVD)), magneto-optic disk (comprising mini-disk (MD) (registered trademark)) and semiconductor memory.Perhaps, storage medium can be hard disk that comprises in ROM 602, the storage area 608 etc., computer program stored wherein, and be distributed to the user with the equipment that comprises them.
The present invention also proposes a kind of program product that stores the instruction code that machine readable gets.When described instruction code is read and carried out by machine, can carry out above-mentioned event detecting method according to the embodiment of the invention.
Correspondingly, be used for carrying the above-mentioned storage medium that stores the program product of the instruction code that machine readable gets and be also included within the present invention.Described storage medium includes but not limited to floppy disk, CD, magneto-optic disk, storage card, memory stick etc.
In the above in the description to embodiments of the invention, can in one or more other embodiment, use in identical or similar mode for the feature that a kind of embodiment is described and/or illustrated, combined with the feature in other embodiment, or the feature in alternative other embodiment.
Should emphasize that term " comprises/comprise " existence that refers to feature, key element, step or assembly when this paper uses, but not get rid of the existence of one or more further feature, key element, step or assembly or additional.
In addition, the time sequencing of describing during method of the present invention is not limited to is to specifications carried out, also can according to other time sequencing ground, carry out concurrently or independently.The execution sequence of the method for therefore, describing in this instructions is not construed as limiting technical scope of the present invention.
Although more than describe by reference to the accompanying drawings embodiments of the invention in detail, should be understood that embodiment described above just is used for explanation the present invention, and be not construed as limiting the invention.For a person skilled in the art, can make various modifications and changes and not deviate from the spirit and scope of the invention above-mentioned embodiment.Therefore, scope of the present invention is only limited by appended claim and equivalents thereof.
Claims (20)
1. event detecting method comprises:
From the video of monitoring scene, obtain by one group of adjacent on time shaft image sequence that image forms;
In described image sequence, extract each the regional motion feature in a plurality of zones in the video scene corresponding with described monitoring scene;
At least according to each regional motion feature zone issue being carried out in described zone detects; And
According to the zone issue testing result in described a plurality of zones, described video scene is carried out clobal detect,
Wherein, the physical space in the corresponding described monitoring scene in described a plurality of zone is measure-alike.
2. according to claim 1 event detecting method, wherein, described zone issue detects and comprises:
From many groups local classifiers, select the one group local classifiers corresponding with the shooting angle of pitch in described zone; And
At least the motion feature in described zone is inputted selected local classifiers, described zone is carried out the event classification.
3. according to claim 2 event detecting method, wherein, every group of local classifiers in described many group local classifiers is used for that the shooting angle of pitch is in a zone in the continuous angle of pitch interval and carries out the event classification, and every group of local classifiers comprises at least one local classifiers, and each local classifiers in every group of local classifiers is at least a event type.
4. each described event detecting method according to claim 1-3, wherein, described clobal detects and comprises:
Zone issue testing result input global classification device with the Zone Full in described a plurality of zones, or with the part adjacent area compositing area group in described a plurality of zones, and the zone issue testing results of one or more zone groups are inputted respectively the global classification device, described video scene is carried out the event classification.
5. according to claim 1 event detecting method also comprises:
According to the clobal testing result that the clobal testing result of one or more other image sequences adjacent with described image sequence is proofreaied and correct described image sequence on time shaft in the described video.
6. each described event detecting method according to claim 1-3, wherein, before obtaining image sequence, described method also comprises:
Described video scene is divided into described a plurality of zone, and determines the shooting angle of pitch in each zone.
7. according to claim 6 event detecting method wherein, is determined the size that each is regional and is taken the angle of pitch with the known bar of diameter is known in the described monitoring scene spheroid and length or the combination of line.
8. according to claim 1 event detecting method, wherein, the size that each is regional is so that as seen the moving target of event detection can be complete in described zone when being in the corresponding physical space in described zone.
9. according to claim 1 event detecting method, wherein, two or more zones in described a plurality of zones overlap each other.
10. each described event detecting method according to claim 1-3, wherein, before carrying out the zone issue detection, described method also comprises:
According to the size in described a plurality of zones, the amplitude of each regional motion feature is carried out respectively normalization.
11. each described event detecting method according to claim 1-3 extracts each regional motion feature and comprises:
Extract the motion feature of a plurality of subregions in each zone; And
Make up the motion feature of described a plurality of subregions as the motion feature in described zone.
12. each described event detecting method according to claim 1-3, wherein, each regional motion feature comprises the histogram of the part or all of motion vector in the described zone and/or the statistic feature of the part or all of light stream in the described zone.
13. an event detection device comprises:
Image sequence obtains parts, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene;
Characteristic extracting component is used for each the regional motion feature in a plurality of zones in the described image sequence extraction video scene corresponding with described monitoring scene;
The zone issue detection part is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And
The clobal detection part is used for the zone issue testing result according to described a plurality of zones, and described video scene is carried out clobal detect,
Wherein, the physical space in the corresponding described monitoring scene in described a plurality of zone is measure-alike.
14. event detection device according to claim 13, wherein, described zone issue detection part comprises:
The sorter alternative pack is used to each zone to select the one group local classifiers corresponding with the shooting angle of pitch in described zone from many groups local classifiers; And
Described local classifiers is at least motion feature that receives respective regions, this zone is carried out the event classification.
15. according to claim 13 or 14 event detection device, also comprise:
Parts are divided in the zone, are used for the video scene corresponding with described monitoring scene is divided into a plurality of zones, and determine the shooting angle of pitch in each zone.
16. event detection device according to claim 15, wherein, described zone is divided parts and is configured to determine size and the shooting angle of pitch that each is regional with the known bar of diameter is known in the described monitoring scene spheroid and length or the combination of line.
17. event detection device according to claim 14, wherein, every group of local classifiers in described many group local classifiers is used for that the shooting angle of pitch is in a zone in the continuous angle of pitch interval and carries out the event classification, and every group of local classifiers comprises at least one local classifiers, and each local classifiers in every group of local classifiers is at least a event type.
18. according to claim 13 or 14 event detection device, wherein, described clobal detection part is configured to receive the zone issue testing result of the Zone Full in described a plurality of zone, or receive the one or more regional zone issue testing result of organizing that the part adjacent area in described a plurality of zone forms, described video scene is carried out the event classification.
19. an event detection system comprises:
Video camera is for the video that obtains monitoring scene; And
Event detection device is used for carrying out event detection at described video,
Wherein, described event detection device comprises:
Image sequence obtains parts, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene;
Characteristic extracting component is used for each the regional motion feature in a plurality of zones in the described image sequence extraction video scene corresponding with described monitoring scene;
The zone issue detection part is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And
The clobal detection part is used for the zone issue testing result according to described a plurality of zones, and described video scene is carried out clobal detect,
Wherein, the physical space in the corresponding described monitoring scene in described a plurality of zone is measure-alike.
20. a video camera comprises:
Image acquisition component is for the video that obtains monitoring scene; And
The event detection parts are used for carrying out event detection at described video,
Wherein, described event detection parts comprise:
Image sequence obtains parts, is used for obtaining by one group of adjacent on time shaft image sequence that image forms from the video of monitoring scene;
Characteristic extracting component is used for each the regional motion feature in a plurality of zones in the described image sequence extraction video scene corresponding with described monitoring scene;
The zone issue detection part is used for according to each regional motion feature zone issue being carried out in described zone at least and detects; And
The clobal detection part is used for the zone issue testing result according to described a plurality of zones, and described video scene is carried out clobal detect,
Wherein, the physical space in the corresponding described monitoring scene in described a plurality of zone is measure-alike.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201210128919 CN103377479A (en) | 2012-04-27 | 2012-04-27 | Event detecting method, device and system and video camera |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201210128919 CN103377479A (en) | 2012-04-27 | 2012-04-27 | Event detecting method, device and system and video camera |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103377479A true CN103377479A (en) | 2013-10-30 |
Family
ID=49462546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201210128919 Pending CN103377479A (en) | 2012-04-27 | 2012-04-27 | Event detecting method, device and system and video camera |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103377479A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014173274A1 (en) * | 2013-04-23 | 2014-10-30 | 湖南工学院 | Reliability judgment method for monitoring transferring actions by operator in digitization main control room of nuclear power plant |
| CN106066990A (en) * | 2015-04-23 | 2016-11-02 | 阿迪达斯股份公司 | For the method and apparatus that the frame in the motion video of people is associated with event |
| CN108171222A (en) * | 2018-02-11 | 2018-06-15 | 清华大学 | A kind of real-time video sorting technique and device based on multithread neural network |
| CN110998594A (en) * | 2017-08-07 | 2020-04-10 | 三菱电机株式会社 | Method and system for detecting motion |
| CN112022066A (en) * | 2020-09-25 | 2020-12-04 | 电子科技大学 | Digestive tract hookworm discovery method and system based on deep learning |
-
2012
- 2012-04-27 CN CN 201210128919 patent/CN103377479A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014173274A1 (en) * | 2013-04-23 | 2014-10-30 | 湖南工学院 | Reliability judgment method for monitoring transferring actions by operator in digitization main control room of nuclear power plant |
| CN106066990A (en) * | 2015-04-23 | 2016-11-02 | 阿迪达斯股份公司 | For the method and apparatus that the frame in the motion video of people is associated with event |
| CN106066990B (en) * | 2015-04-23 | 2019-06-14 | 阿迪达斯股份公司 | For the frame method and apparatus associated with event in the motion video by people |
| CN110998594A (en) * | 2017-08-07 | 2020-04-10 | 三菱电机株式会社 | Method and system for detecting motion |
| CN110998594B (en) * | 2017-08-07 | 2024-04-09 | 三菱电机株式会社 | Method and system for detecting motion |
| CN108171222A (en) * | 2018-02-11 | 2018-06-15 | 清华大学 | A kind of real-time video sorting technique and device based on multithread neural network |
| CN108171222B (en) * | 2018-02-11 | 2020-08-25 | 清华大学 | Real-time video classification method and device based on multi-stream neural network |
| CN112022066A (en) * | 2020-09-25 | 2020-12-04 | 电子科技大学 | Digestive tract hookworm discovery method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103116987B (en) | Traffic flow statistic and violation detection method based on surveillance video processing | |
| CN109977782B (en) | Cross-store operation behavior detection method based on target position information reasoning | |
| Porikli et al. | Video surveillance: past, present, and now the future [DSP Forum] | |
| CN102638675B (en) | Method and system for target tracking by using multi-view videos | |
| CN101410855A (en) | Method for automatically attributing one or more object behaviors | |
| CN103377479A (en) | Event detecting method, device and system and video camera | |
| CN110706247B (en) | Target tracking method, device and system | |
| Piciarelli et al. | Surveillance-oriented event detection in video streams | |
| CN112185098B (en) | Shared bicycle monitoring method and system based on city monitoring video | |
| CN111274886B (en) | Deep learning-based pedestrian red light running illegal behavior analysis method and system | |
| CN102385592A (en) | Image concept detection method and device | |
| CN115660262B (en) | Engineering intelligent quality inspection method, system and medium based on database application | |
| CN103390151A (en) | Face detection method and device | |
| CN103577795A (en) | Detection equipment and method, detector generation equipment and method and monitoring system | |
| CN109993032A (en) | A kind of shared bicycle target identification method, device and camera | |
| CN105469054A (en) | Model construction method of normal behaviors and detection method of abnormal behaviors | |
| CN114998819A (en) | Passenger flow statistical method, device, equipment and medium for multi-dimensional detection and tracking | |
| CN105160285A (en) | Method and system for recognizing human body tumble automatically based on stereoscopic vision | |
| Wang et al. | Traffic camera anomaly detection | |
| CN113537170A (en) | Intelligent traffic road condition monitoring method and computer readable storage medium | |
| CN117115412A (en) | Small target detection method based on weighted score label distribution | |
| CN115620098B (en) | Evaluation method and system of cross-camera pedestrian tracking algorithm and electronic equipment | |
| CN111723664A (en) | Pedestrian counting method and system for open type area | |
| CN108960013A (en) | A kind of pedestrian recognition methods and device again | |
| CN106778765A (en) | A kind of method and device of Car license recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131030 |