Summary of the invention
The present invention is directed to the defective of existing supervisory control system, a kind of multi-camera intelligent control method and device thereof are provided, it can carry out the omnidistance camera to automatically track of striding to the moving target in the guarded region of multiple-camera covering, and can fully utilize the moving target information of each camera acquisition, thereby to the control that links of the video camera of whole system.This guarded region can be the zone that one or several road is formed, or a sub-district.The said tracking of the present invention accuses that all the system camera pan-tilt rotates and lens focus (being called for short the control video camera), and target is in the picture of video camera picked-up with suitable size and position, carries out camera automatic focusing simultaneously, makes clear picture.Such system has improved the validity of monitoring image and recording image information greatly, has reduced the monitoring blind area.The present invention also carries out zoom amplification and tracking to detected moving target or area-of-interest, and the feature information of picked-up, record moving target has improved the validity to the moving target identification.Camera automatic focusing is generally carried out automatically by video camera, also can be undertaken by processor control, and as the subnetwork video camera, the present invention all refers to comprise automatic focus to the tracking of moving target.
Characteristics of the present invention are:
After in camera views, detecting moving target, the vision signal from each video camera is analyzed, detected the moving target in the vision signal earlier, analyze size, position and the kinematic parameter of moving target in image again; Again according to size, position and the kinematic parameter of moving target in image, control focus of camera and orientation, moving target is appeared in the video pictures of this video camera with suitable position and size, thereby moving target is followed the tracks of, promptly enter general tracing mode.Wherein the size of moving target in image is the pixel quantity in moving target shared zone in image, the position of moving target in image is the geometric center of motion target area or the position feature that other method is determined, the kinematic parameter of moving target comprises the direction of motion and movement velocity, is tried to achieve for the variation of time by the position of moving target.In tracing process, control focus of camera and orientation, the position and the size that moving target are had be suitable for following the tracks of on image, just make moving target be near zone, monitored picture center, and the size on the monitored picture of moving target remains within certain scope, simultaneous camera carries out automatic focus, makes moving target have higher definition.Control to focus of camera and orientation, be according to the current focal length of video camera and the parameters such as size, position and the direction of motion of orientation state parameter and moving target, the camera lens and the The Cloud Terrace of control video camera make this moving target have suitable position and size on image; Video camera when front focal length and direction parameter can not obtain, the method that can adopt repeatedly control is convergent-divergent focal length and adjust the orientation gradually, makes this moving target have suitable position and size on image.
When a moving target is in tracked state, information such as the position of moving target, the direction of motion in the extraction image.Again according to the position of moving target and the monitoring range of the direction of motion, current video camera and adjacent camera, the predicted motion target is with the adjacent camera monitor area that enters, control the orientation and the focal length of this adjacent camera, make the adjacent camera visual field catch this moving target, or wait for that moving target enters the visual field.
Characteristics of the present invention also are:
When in not detecting moving target and contiguous camera supervised scope in the camera video picture, not having moving target, automatically focus of camera and orientation are placed some predefined presetting bits, promptly automatically reset, and a plurality of automatically reset presetting bits are set as prediction scheme to video camera according to application requirements or ambient conditions, when not having moving target in the monitoring range of current video camera and contiguous video camera, the video camera presetting bit is adjusted automatically according to time, application need and detected ambient conditions, promptly adjusts to corresponding prediction scheme.Selection such as the presetting bit that automatically resets can be set according to people's flow path direction in commuter time.
After detecting moving target in the camera views,, promptly moving target is carried out track up by control focus of camera and orientation pursuit movement target.In the track up process, moving target is near the center of camera views all the time, and make it account for the whole image certain proportion, and control camera lens at regular intervals at interval, and change focal length, generally be to make focal length elongated, control camera orientation simultaneously, make camera lens alignment motion target, thereby make moving target be in a period of time in the picture, promptly enter the feature tracing mode in the mode of amplifying feature.Under the feature tracing mode, moving target is in tracking mode all the time.Compare with general tracing mode, moving target shared ratio in monitored picture is bigger under the feature tracking mode.Amplify feature and carry out, thereby obtain the more detailed information of interesting target, make interesting targets such as people, car more clear at interested moving target.If the feature tracing mode duration is shorter, or target speed is slower, not adjusting camera orientation also is unlikely to make moving target to break away from this camera coverage, this can judge according to the movement velocity of target with near the degree of image boundary, then can not adjust camera orientation when the feature tracing mode.After the feature tracing process continues for some time, the control video camera continues moving target is generally followed the tracks of with suitable focal length and orientation, just make the size of the general tracing mode of moving target before the size on the image returns to the feature tracking, thereby be convenient to tracking, also be convenient to monitor bigger zone moving target.This process can periodically be carried out, and breaks away from the monitoring range of current video camera up to moving target.Feature is followed the tracks of also and can be carried out a period of time when bigger variation takes place the moving target shape facility.
If the direction parameter of the focal length parameter of the camera lens of video camera (or Zoom parameter), camera pan-tilt is available, and can control, then may carry out quick control to video camera, thereby realize track up by parameter estimation by setup parameter; Otherwise by the mode of approaching gradually video camera is controlled, realized track up.
Characteristics of the present invention also are:
If in the camera video picture, detect a plurality of moving targets, as detect a plurality of pedestrians, then, one by one each moving target is carried out the amplification feature of a period of time, so that can obtain the picture rich in detail of each moving target according to the position and the size of moving target on image; Perhaps each detected moving target is carried out signature analysis, extract the moving target feature, when detecting a plurality of moving target, specify one of them moving target, select moving target to follow the tracks of with the clarification of objective that the feature or the preassigned feature of this moving target are followed the tracks of as needs, promptly current video camera or adjacent camera are only followed the tracks of the wherein feature moving target the most similar to specific characteristic.Can adopt one or more target signatures, such as target sizes, movement velocity, color characteristic, textural characteristics, shape facility, behavioural characteristic etc.
Characteristics of the present invention also are:
Result according to target detection analyzes moving target, finds interesting areas in the target, can obviously characterize the zone of target identities such as people's face or car plate etc.Suitably the orientation and the focal length of control video camera carry out the feature processing and amplifying to interesting areas, to improve the definition of area-of-interest, keep the tracking to the moving target area-of-interest simultaneously, promptly enter the area-of-interest tracing mode.If the area-of-interest tracing mode duration is shorter, or target speed is slower, not adjusting camera orientation also is unlikely to make this area-of-interest to move to beyond the camera field of view, this can judge near the degree of image boundary according to the movement velocity and the area-of-interest of target, then can not adjust camera orientation when the area-of-interest tracing mode.At this moment the tracking to moving target can also can be realized than complicated algorithm by other simply according to modes such as textural characteristics couplings.After obtaining clearly the region of interest area image, area-of-interest is carried out feature extraction and identification, as wherein car plate or people's face etc. are detected or discern, thereby realize the automation of video monitoring.Wherein feature extraction, car plate identification and recognition of face can be adopted the algorithm of various maturations.
Characteristics of the present invention also are:
Before the pursuit movement target, need to detect earlier the moving target in the vision signal, can adopt the algorithm that detects moving target in some ripe vision signals, as frame difference method, background subtraction method, mixed Gauss model method, target split plot design and optical flow method etc., can only use wherein a kind of method, also can several method be combined usefulness according to the vision signal characteristics, as having when stablizing background, adopt background subtraction method or mixed Gauss model method, when the vision signal change of background is very fast, can adopt frame difference method or target split plot design or optical flow method.
Characteristics of the present invention also are:
Motion target detection adopts the mixed Gauss model method in conjunction with the neighborhood characteristic, be that each is judged to background and the higher pixel of confidence level in the neighbor of current pixel, prior probability or its decision range when the background model that all strengthens current pixel is used to adjudicate, each is judged to prospect and the higher pixel of confidence level in the neighbor of current pixel, prior probability or its decision range when the background model that all reduces current pixel is used to adjudicate.
When adopting the mixed Gauss model method, be judged to the fiducial probability height of Gauss model as a setting, then be judged to background, otherwise be judged to prospect.Here the said confidence level that certain pixel is judged to background or prospect just adopts the Bayes rule to be judged to the posterior probability of background or foreground pixel, and the available probability sum that is judged to several Gauss models is as a setting calculated.But it is complicated because of calculating like this, adopt approximate method during concrete usually calculating, promptly use rule: declare the background of work or prospect Gauss model have minimum value or, replace rule for Mahalanobis distance less than certain threshold value with its average: whether the confidence level that is judged to background or prospect surpasses certain threshold value.During the mixed Gauss model algorithmic decision,, often suppose that each feature is separate, and use with the Mahalanobis distance of Gauss model average and whether adjudicate less than a predetermined threshold value Ti in order to simplify calculating.Wherein i represents i Gauss model.But conventional mixed Gauss model is each pixel independent judgement, do not utilize neighborhood relevance: when being background pixel around certain pixel, the possibility that this pixel belongs to background increases, and when being foreground pixel around it, the possibility that this pixel belongs to prospect increases.
In order to make full use of the correlation between the neighborhood territory pixel in the monitoring image, the present invention adopts the method that combines with statistical property based on neighborhood territory pixel to carry out target detection, be in particular the mixed Gauss model method of employing in conjunction with the neighborhood characteristic, be that each is judged to background and the higher pixel of confidence level in the neighbor of current pixel, prior probability or its decision range when the background model that all strengthens current pixel is used to adjudicate, each is judged to prospect and the higher pixel of confidence level in the neighbor of current pixel, prior probability or its decision range when the background model that all reduces current pixel is used to adjudicate.When its body is realized, neighborhood territory pixel is that the fiducial probability of background is when higher, or conventional mixed Gauss model be will be judged to background with more neighborhood territory pixel the time, and the Mahalanobis that temporarily reduces those models as a setting promptly uses Ti-ε as decision threshold apart from decision threshold; Neighborhood territory pixel is the fiducial probability of prospect when higher, or conventional mixed Gauss model be will be judged to prospect with more neighborhood territory pixel the time, and the temporary transient Mahalanobis that increases those models as a setting promptly uses Ti+ ε as decision threshold apart from decision threshold.Wherein i represents i Gauss model, if n Gauss model is as a setting arranged, and i=1 then, 2 ... n, ε are an arithmetic number.Therefore, in conjunction with the decision threshold of the mixed Gauss model method of neighborhood characteristic is that to increase be to subtract, depend in the neighborhood with the quantity of high confidence pixel as a setting with the contrast of high confidence as the quantity of the pixel of prospect, or depend on that simply conventional mixed Gauss model is judged to neighborhood territory pixel the quantity and the contrast that is judged to the quantity of foreground pixel of the pixel of background.Here the increase of decision threshold or reduce only is used for the judgement of current frame pixel, is not used in the judgement of subsequent frame pixel.When the conventional mixed Gauss model and the judgement of the mixed Gauss model method that combines the neighborhood characteristic were inconsistent, the reliability of judgement was lower, can not upgrade the Gauss model of this pixel.
Another object of the present invention provides a kind of multiple-camera intelligence controlling device, and it comprises: Video Detection module, video camera control module.This multiple-camera intelligence controlling device combines with multiple cameras, video encoder, communication module, video recording equipment and other expansion equipment, can constitute intelligent video monitoring system.
Video camera has The Cloud Terrace and controlled camera lens, or integrated The Cloud Terrace and controlled camera lens, separates the video camera of The Cloud Terrace as integration spherical video camera, band, has the controlled characteristics in focal length and orientation.
The Video Detection module realizes moving object detection, and extracts size, position and the kinematic parameter of moving target on image, and wherein kinematic parameter comprises the direction of motion and movement velocity.The Video Detection module sends to the video camera control module with size, position and the kinematic parameter of the moving target that extracted.
The video camera control module receive from the Video Detection module of each video camera correspondence about information such as moving target size, position and kinematic parameters, movement tendency is carried out analyses and prediction, and described movement tendency is meant adjacent camera and the monitoring range thereof that moving target may arrive behind the monitoring range that next constantly possible position or moving target break away from current video camera.The video camera control module is sent control information based on predicting the outcome to video camera, control the orientation and the focal length of corresponding video camera or its contiguous video camera, moving target is appeared in the camera video picture with suitable size, moving target is followed the tracks of, or make its contiguous video camera wait for that moving target enters the visual field, moving target is striden the relay tracking of video camera.
In the tracing process to moving target, the video camera control module can realize the amplification to the moving target short time, promptly realizes the feature tracking.In tracing process to moving target, the Video Detection module also detects area-of-interest, the information such as size, position and kinematic parameter of area-of-interest are sent to the video camera control module, and the video camera control module realizes the short time feature of area-of-interest in the moving target is followed the tracks of.When area-of-interest being carried out the feature tracking, also can carry out image recognitions such as recognition of face and car plate identification.
Characteristics of the present invention also are:
Its Video Detection module is also analyzed the characteristic information of moving target, and characteristic information comprises target sizes, color characteristic, textural characteristics, shape facility, behavioural characteristic etc., and characteristic parameter is carried out record.When detecting a plurality of moving target, the video camera control module is according to the characteristic specified parameter, or the characteristic parameter of the moving target of appointment, selects one and follow the tracks of near the moving target of specific characteristic in the moving target of video camera picked-up.When specific characteristic parameter not, adopt the characteristic parameter of acquiescence to select, as select big and movement velocity faster moving target follow the tracks of.The video camera control module is to the video camera sending controling instruction, controls current video camera or contiguous video camera, makes its tracking and intended target or specific characteristic have the moving target of similar features.
Characteristics of the present invention also are:
Video Detection module and video camera control module constitute an autonomous device, analyze the vision signal from video camera.This vision signal is from the analog signal of video camera or digital video bit stream signal.Wherein the Video Detection module can be analyzed the vision signal from or multiple cameras, and analysis result is delivered to the video camera control module.The video camera control module can receive the analysis result to or multiple cameras Video Detection from the Video Detection module, determine tracking target and tracking mode according to this analysis result, send the video camera control signal, orientation and focal length to video camera are controlled, realization is followed the tracks of the general tracking and the feature of moving target, or realizes the feature of area-of-interest is followed the tracks of.
Characteristics of the present invention also are:
Video Detection module and video camera control module can be built in the video camera, or are built in the video encoder.In the time of in being built in video camera, the vision signal of this video camera is detected, can only give this video camera, or give this video camera and its contiguous video camera sending controling instruction simultaneously.In the time of in being built in video encoder, can each the road vision signal that be encoded be detected, and to corresponding video camera sending controling instruction, also can be simultaneously to other contiguous video camera sending controling instruction.
Embodiment
Figure 1 shows that an embodiment process of multiple-camera control of the present invention:
Before beginning, this process needs system is carried out initialization, place some predefined presetting bits (aiming at the inlet of road) as initial focal length and orientation with video camera, define the neighbouring relations of each video camera, define some zones related as each video camera with adjacent camera, this zone can define with focal length of camera, direction parameter scope, and the title of definition adjacent camera or numbering and corresponding presetting bit thereof, also can utilize GIS-Geographic Information System (GIS) to carry out the definition of adjacent camera associated region.
Image from video camera is imported frame by frame, after each two field picture input, detect moving target wherein, can adopt the whole bag of tricks to carry out object detection method, as frame difference method, background subtraction method, mixed Gauss model method, target split plot design or optical flow method etc. and improvement or combined method commonly used.
If do not detect moving target in present image, then process is returned, and waits for two field picture input;
If in present image, detect moving target, then analyze size, position and the kinematic parameter of this moving target, by control focal length of camera and orientation pursuit movement target, camera automatic focusing.As adopt moving target on image area (pixel count) or the area of area-encasing rectangle as the size of moving target, the focal length of control camera lens, the area on image is a suitable ratio (needing to consider the control required time) in video camera control back to make moving target, as near 1/8.If video camera has adopted high resolution sensor, then can also realize virtual zoom by the mode of digital zooming.Utilize position and the direction of motion, movement velocity of moving target on present image, can estimate moving target position on image (needing to consider the control required time) after video camera control, the orientation of control camera pan-tilt makes moving target be in picture central authorities as far as possible when control is finished.
Simultaneously according to predefined adjacent camera associated region, certain focal length and bearing range have been entered during as the pursuit movement target, then the corresponding contiguous video camera of control is corresponding predefined focal length and orientation, make the adjacent camera visual field catch this moving target, or wait for that moving target enters the visual field.
Said process returns after handling, and waits for the input of next frame image.
Above-mentioned in tracing process, according to size, position and the kinematic parameter of detected moving target in image, by controlling the tracking of focus of camera and orientation realization to moving target, but the control to video camera is every number frame (as 5 frames) control once, moving target changes when little on image, can control, therefore, to motion target detection also is to detect once every the number frame, and other picture frame is used to safeguard background image, mixed Gauss model etc.
Be illustrated in figure 2 as based on the multiple-camera interlock of multiple-camera control and follow the tracks of schematic diagram.Great circle among the figure is represented the monitoring range of video camera, and A, B, C, D are 4 video cameras, and blockage M represents moving target, and as the pedestrian, dotted line is represented the movement locus of this moving target.The visual field was an inlet of this road when video camera A was flat, and moving target M enters from the porch, in case enter in the monitoring range of video camera A, then is detected.Orientation and the focal length of control video camera A, M follows the tracks of to moving target, video camera A will counterclockwise rotate among the figure, in case the α angular region shown in the azimuth inlet point line of video camera is then controlled related adjacent camera B, according to default related Control Parameter, forward video camera B to orientation as shown in phantom in FIG., control focal length simultaneously, wait for entering of target M, target also may directly enter the visual field of video camera B.Control video camera B then and follow the tracks of this moving target M.Similarly, if moving target M enters the zone approaching with video camera C, as the β angular region at inlet point line place among the figure, then control video camera C to as the orientation among the figure, wait for that moving target M comes into view, after moving target enters camera coverage, continue to follow the tracks of by video camera C.
In above-mentioned tracing process, also can amplify feature to moving target as required and handle moving target.
Figure 3 shows that an embodiment to the motion target tracking process, in tracing process moving target has been carried out the amplification feature, specific algorithm is as follows:
1. whether whether preestablish needs moving target to amplify feature, needs area-of-interest (ROI) to amplify parameters such as feature, secondary feature time interval T1 (being non-feature tracking time), motion target close duration T 2, area-of-interest (ROI) feature duration T 3.After detecting motion, follow the tracks of beginning.
2. by the control video camera moving target is carried out non-feature mode and follow the tracks of duration T 1.
3. check that whether needing that moving target is carried out feature follows the tracks of; If, next step; Otherwise, changeed for the 8th step.
4. the control video camera makes moving target at picture larger proportion be arranged, and control video camera pursuit movement target, moving target is carried out feature follow the tracks of duration T 2.
5. check that whether needing that area-of-interest on the moving target is carried out feature follows the tracks of; If, next step; Otherwise, changeed for the 8th step.
6. the detection area-of-interest is people's face generally to human body target, and vehicle target is generally car plate.
7. control video camera makes that area-of-interest has larger proportion on the moving target on picture, and the area-of-interest on the control video camera pursuit movement target, duration T 3.
8. check to follow the tracks of whether finish, leave as the artificial interruption of quilt, moving target etc.If then finish; Otherwise change 2, continue the motion target tracking of following one-period.
In tracing process, can change the setting that whether needs feature to follow the tracks of, whether need the area-of-interest feature to follow the tracks of, thus control feature number of times, or increase the feature process when needed.
In tracing process, make moving target or its area-of-interest on picture, have certain proportion or a certain size, be to realize, moving target or the size of its area-of-interest in the video pictures of this video camera are within [S-Δ S, S+ Δ S] by control to the camera lens focal length.Here S and Δ S all are predefined values.For making image keep clear, simultaneous camera automatic focus.
In tracing process, move when very fast at moving target, consider the time-delay of camera orientation machinery control, the control of azimuth of video camera is according to position, size and the direction of motion of moving target on image, movement tendency to moving target is predicted, camera pan-tilt is controlled again, made moving target near the middle section that is in image (visual field) after the control, promptly the control of azimuth of video camera need utilize the movement tendency prediction of moving target to adjust.When movement velocity is slow, according to position and the size of moving target on image, camera pan-tilt is controlled simply, the middle section of the position of moving target on present image as control back image.To the control of camera coverage, generally be to control realization by The Cloud Terrace to video camera, also be so to the incorporate sphere of The Cloud Terrace camera lens, for high-resolution camera or panoramic camera, then can adopt the mode of virtual roaming to control.
In above-mentioned tracing process, if detect a plurality of moving targets, and need carry out feature tracking and area-of-interest (ROI) tracking to a plurality of moving targets, then one by one moving target is carried out feature tracking, area-of-interest tracking, after each target close tracing process end, return to general tracing mode.In general tracing mode, with wherein the maximum or the fastest target of moving make other detected moving target also come into view as main tracking target simultaneously; Or with the centre of form of the boundary rectangle of these moving targets as center, the visual field, detected moving target is all come into view.
In above-mentioned tracing process, if detect a plurality of moving targets, and only need follow the tracks of intended target or the moving target with specific characteristic, then detected each moving target is carried out the extraction of characteristic information, the master sample feature that the feature of intended target or specific characteristic are followed the tracks of as need, the master sample feature can have a plurality of, adopt method and each master sample feature of pattern recognition to mate the feature of each moving target, as minimum distance method, the moving target that has similar features with one of them master sample feature is followed the tracks of.When multiple-camera was followed the tracks of, the video camera of next relay tracking was adopted the selection tracking target that uses the same method.
Fig. 4 is a kind of device example structure figure that realizes multi-camera intelligent control method of the present invention.
The core of this device is a digital information processing system, SOC (system on a chip) (SoC) chip, random memory (RAM), the non-volatile memory devices such as (as Nor Flash) that comprise the needed digital signal processor of general digital information processing system (DSP) or have digital signal processing capability.Comprise Video Detection module and video camera control module in the application software of digital information processing system operation, the Video Detection module is used for detecting the moving target of video, analyzes size, position and the kinematic parameter of moving target; The video camera control module is controlled this video camera and contiguous camera orientation and focal length according to size, position and the kinematic parameter of moving target.Also move other control and communication software on the digital information processing system, as the Ethernet protocol stack.Image processor or multimedia processor as DM64x, DM643x and the DM644x of TI company, all are desirable digital signal processors.
This device has one or more video input interfaces, is used to receive one or more vision signal.When the composite video signal that receives from the common simulation video camera, need a slice video decoding chip (Video Decoder), video decoding chip adopts ITU656 or ITU601 video data format with vision signal supplied with digital signal treatment system usually; When the vision signal that directly receives from CMOS or ccd image sensor, directly with the vision signal supplied with digital signal treatment system after the digitlization; In the vision signal that receives other form, need corresponding video acquisition interface.
This device also has ethernet interface, is used to receive the control command from video monitoring system, also can send alarm signal to Surveillance center after detecting moving target or special exercise target.
This device also has the control communication interface, as RS232, RS485, is used for to the video camera transmitting control commands.This interface also can be used for and devices communicatings such as digital video recorder (DVR), digital video frequency server (DVS) and video switching matrix, and the video camera control command is sent to these equipment, realizes video camera control by them.
Control communication interface and also can accept the control command of self-monitoring system, send alarm signal, at this moment can save ethernet interface to Surveillance center.
Also can send to miscellaneous equipment in DVR, DVS or the supervisory control system to the control command of video camera, realize video cameras control, then can save the control communication interface by these equipment by Ethernet.
Vision signal also can be imported in the mode of video flowing by Ethernet, at this moment can not have video input interface.
If had the digital signal processor of enough disposal abilities in DVS, DVR, the web camera, then, can realize multiple-camera Based Intelligent Control of the present invention as long as Video Detection and video camera Control Software module on the video signal processing system in accompanying drawing 4 devices are embedded in the digital signal processor software of DVR, DVS, web camera.
Ordinary PC also can be used as the hardware platform of realizing multi-camera intelligent control method, as long as increase the video acquisition interface, as video frequency collection card, and increases Video Detection and video camera control application software gets final product.
Supplied with digital signal is handled treatment system if vision signal adopts the Ethernet mode, all can pass through Ethernet supplied with digital signal treatment system as the vision signal of vision signal behind the compressed encoding or uncompressed, then can save video input interface among Fig. 4.The equipment that generally has Ethernet interface and have digital signal processing capability all can become the device of realizing multi-camera intelligent control method as embedded system or PC.
Fig. 5 realizes the embodiment of video monitoring system of the device of multi-camera intelligent control method for adopting the present invention.
The video camera that adopts in this system has or integrated The Cloud Terrace and controlled camera lens, and the video camera as integration spherical video camera, band separation The Cloud Terrace has the controlled characteristics in focal length and orientation.
The video camera intelligent controller is exactly a device of realizing multi-camera intelligent control method in the system.
Video encoder is meant the equipment of realizing video coding, and as DVR, DVS or web camera, its is to encoding video signal, output compressed code flow of input, and accepts the control of video monitoring system (monitor terminal or monitoring server).Video data behind the compressed encoding sends to monitor terminal, also can send to concentrated video recording equipment (as the IP-SAN disk array).
The video monitoring terminal is communicated by letter with video encoder, video camera intelligent controller by Ethernet with video monitoring server.
Vision signal from video camera is divided into two-way, as adopting video distributor.One road signal enters the video camera intelligent controller; Another road signal enters video encoder.
Video Detection module in the video camera intelligent controller realizes moving object detection, and extracts size, position and the kinematic parameter of moving target on image, and wherein kinematic parameter comprises the direction of motion and movement velocity.Moving target accounts for to such an extent that ratio or pixel number are represented in the size on the image on image with moving target; Moving target is represented with the geometric center (center of gravity or boundary rectangle, external polygonal center etc.) of moving target on image in the position on the image.The kinematic parameter of moving target comprises movement velocity, the direction of motion, and wherein movement velocity is represented with the translational speed of the geometric center of moving target, the direction of motion is represented with the moving direction of the geometric center of moving target.
Video camera control module in the video camera intelligent controller receives the information such as the size about moving target, position and kinematic parameter from the Video Detection module of each video camera correspondence, and movement tendency is carried out analyses and prediction.Movement tendency prediction is according to the position and the kinematic parameter of current moving target, estimates moving target and will appear at position in the monitoring image constantly at the next one.
The pixel number that video camera control module in the video camera intelligent controller accounts for the ratio of image or moving target according to moving target in the monitoring image is regulated focus of camera and is changed, and makes the ratio that tracked moving target accounts in monitoring image or the pixel number of moving target reach restricted portion; According to moving target in monitoring image the position and the movement tendency of moving target predict the orientation of regulating camera lens, make tracked moving target in motion process, be in all the time in the localized area of monitoring image.
In the tracing process to moving target, the video camera control module can realize the amplification to the moving target short time, promptly realizes the feature tracking.In tracing process to moving target, the Video Detection module can also adopt detection algorithms such as people's face or car plate to detect area-of-interest, the information such as size, position and kinematic parameter of area-of-interest are sent to the video camera control module, and the video camera control module realizes the short time feature of area-of-interest in the moving target is amplified tracking.When area-of-interest being carried out the feature tracking, also can adopt corresponding algorithm to carry out image recognitions such as recognition of face and car plate identification.
Its Video Detection module is also analyzed the characteristic information of moving target, and characteristic information comprises target sizes, color characteristic, textural characteristics, shape facility, behavioural characteristic etc., and the characteristic parameter of moving target is carried out record.When detecting a plurality of moving target, the video camera control module is selected one according to the characteristic specified parameter and is followed the tracks of near the moving target of specific characteristic in the moving target of video camera picked-up.Adopt the characteristic parameter of acquiescence to select when specific characteristic parameter not, the spy of acquiescence sets according to actual conditions in parameter, as select the big and movement velocity of area faster moving target follow the tracks of.The video camera control module compares the characteristic information of the moving target of appearance in the current camera supervised scope with the characteristic information that is close to the moving target that occurred or occurred in the video camera, analysis and judgement goes out identical moving target, and, make current video camera or adjacent camera follow the tracks of wherein the moving target the most similar in the moving target record to specific characteristic to the video camera sending controling instruction.
In system shown in Figure 5, the video camera intelligent controller is a separate equipment, can connect one or more video camera simultaneously, processing is from the video data of one or more video camera, and control one or more focus of camera and orientation and change, the result data that process Video Detection module and video camera control module are handled sends to server.
Video Detection module and video camera control module in the video camera intelligent controller also can be built in the video camera, or are built in the video encoder (as DVR, DVS).In the time of in being built in video camera, the vision signal of this video camera is detected, can only give this video camera, or give this video camera and its contiguous video camera sending controling instruction simultaneously.In the time of in being built in video encoder, can analyze each road vision signal of importing this encoding device, and to corresponding video camera sending controling instruction, also can be simultaneously to other contiguous video camera sending controling instruction.
Fig. 6 realizes another embodiment of video monitoring system of the device of multi-camera intelligent control method for adopting the present invention.Wherein the video camera intelligent controller can be one or more equipment, and different with Fig. 5 is, the video camera intelligent controller receives one or more video stream signal by network, and built-in video decode module obtains vision signal behind the decompress(ion) by video decode.The function of Video Detection module and video camera control module is identical with Fig. 5 in the video camera intelligent controller.The video camera control module transmits control signal for video encoder or monitoring server by Ethernet, thereby realizes video camera control.The video camera intelligent controller can be by the equipment realization that has Ethernet interface and have digital signal processing capability, the equipment based on flush bonding processor (as equipment) that has network interface as employing based on DM644x, DM643x and DM64X series processors, or the computer with network interface, wherein dispose software modules such as video decode, Video Detection, video camera control.