Summary of the invention
In view of above content, be necessary to provide a kind of object supervisory control system, can be under busy, crowded, mixed and disorderly background, because of shooting angle, convergent-divergent cause under change of object profile or the light situation of change, simultaneously and effectively monitoring with judge the object that is removed in the guarded region or enter thing that and this enters thing to utilize characteristic point to describe vector identification.
Also be necessary to provide a kind of object method for supervising, can be under busy, crowded, mixed and disorderly background, because of shooting angle, convergent-divergent cause under change of object profile or the light situation of change, simultaneously and effectively monitoring with judge the object that is removed in the guarded region or enter thing that and this enters thing to utilize characteristic point to describe vector identification.
A kind of object supervisory control system, run in the image server, this system comprises: prospect object detecting unit, be used for utilizing the prospect object of the image that double-deck background model detecting watch-dog catches, and this bilayer background model comprises existing background model and temporary background model; Object and regional determination unit, be used for when the prospect object of this detecting still is judged as the prospect object after more than or equal to a setting-up time interval, being moved into the temporary described pixel of background model tense marker in the pixel of this prospect object is interested pixel, from temporary background model, search pixel identical in the zone that is close to described interested pixel as interested pixel with the pixel value of described interested pixel, obtain a pixel set b thus, when the area of set b during greater than a setting range, acquisition and the corresponding pixel of set b from existing background model, and obtain pixel set a thus; Described object and regional determination unit, also be used to utilize characteristic point algorithm pair set a and b enforcement computing respectively, find out the characteristic point in each pixel set and describe vector, to gather characteristic point among a then and implement the image cutting as seed and obtain block A, and will gather among the b with block A opposite position on characteristic point implement image cutting as seed and obtain block B; And the article identification unit, be used for judging in this guarded region to remove thing when the area of block B during greater than the area of block A, reach when the area of block B during less than the area of block A, judge in this guarded region to enter thing.
A kind of object method for supervising comprises the steps: to utilize the prospect object in the double-deck background model detecting watch-dog institute capturing video, and this bilayer background model comprises existing background model and temporary background model; If the prospect object of this detecting still is judged as the prospect object after more than or equal to a setting-up time interval, then being moved into temporary this respective pixel of background model tense marker in this prospect object respective pixel is interested pixel; From temporary background model, search the zone contiguous, find out the pixel identical, obtain a pixel set b thus as interested pixel with the pixel value of described interested pixel with described interested pixel; When the area of set b during, acquisition and the corresponding pixel of described set b from existing background model, and obtain pixel thus and gather a greater than a setting range; Utilize characteristic point algorithm pair set a and b enforcement computing respectively, find out the characteristic point in each pixel set and describe vector; The characteristic point of set among a implemented the image cutting as seed obtains block A, and will gather among the b with block A opposite position on characteristic point implement image cutting as seed and obtain block B; When the area of block B during, judge in this guarded region to remove thing greater than the area of block A; And, judge in this guarded region to enter thing when the area of block B during less than the area of block A.
Compared to prior art, described object supervisory control system and method, utilize colour element to set up background model, before judging with this, the background object, generally only adopt the supervisory control system and the method for GTG pixel, has better judgment, it can not only be busy, crowded, discern the object that is removed in the guarded region under the mixed and disorderly background or enter thing, also can be in shooting angle, when convergent-divergent causes change of object profile or light to change, simultaneously and effectively monitoring with judge the object that is removed in this guarded region and enter thing that this enters thing to utilize characteristic point to describe vector identification.
Embodiment
As shown in Figure 1, be the running environment figure of object supervisory control system of the present invention preferred embodiment.This object supervisory control system 10 is installed and is run in the image server 1.This image server 1 links to each other with a characteristic point data storehouse 3 with at least one watch-dog 2 by network.In the present embodiment, described watch-dog 2 can have the electronic equipment of monitoring function for web camera or other types.The characteristic point that has the multiple object (comprising the people) that training in advance crosses in the described characteristic point data storehouse 3 is described vector model.
As shown in Figure 2, be the function unit figure of object supervisory control system 10 preferred embodiments of the present invention.In this figure, except operation has object supervisory control system 10, also comprise memory device 20, processor 30 and display device 40 in the image server 1.
Wherein, memory device 20 is used to store the computerization program code of described object supervisory control system 10, and storage is by the captured chromatic image of watch-dog 2.In other embodiments, this memory device 20 can be the external memory of image server 1.
Processor 30 is carried out the computerization program code of described object supervisory control system 10, promptly the image that watch-dog 2 is caught carries out the detecting of prospect object, object interested and zone in the described image is judged, and judge remove thing or enter thing in the guarded region after identification this which kind of object enters thing be, and send alarm.
Display device 40 is used to show the chromatic image that described watch-dog 2 is captured, and processor 30 pairing each picture when carrying out object supervisory control systems 10, cuts picture, schematic diagram as shown in Figure 8 as the image of background area and prospect object.
Described object supervisory control system 10 comprises: prospect object detecting unit 100, object and regional determination unit 102 and article identification unit 104, the function of this object supervisory control system 10 can specifically describe by Fig. 3 to Fig. 8.
Described prospect object detecting unit 100 comprises model building module shown in Figure 3 1000, pixel separation module 1002, memory module 1004, temporary background model monitoring module 1006 and background model update module 1008.This prospect object detecting unit 100 is used for utilizing the prospect object of the image that double-deck background model detecting watch-dog 2 caught, and concrete grammar will be described in detail in Fig. 5.Wherein, described double-deck background model comprises existing background model and temporary background model, and this existing background model is meant the background model that a width of cloth image is generated before the current image of detecting.
Described object and regional determination unit 102, be used for when described prospect object when still being judged as the prospect object after at interval more than or equal to setting-up time, if form the pixel of this prospect object be moved into temporary background model then automatically the described pixel of mark be interested pixel.Object and regional determination unit 102 search in the zone that is close to described interested pixel whether have the pixel identical with these interested pixels from temporary background model, and this pixel that searches is considered as interested pixel equally, obtain a pixel set b thus.In the present embodiment, described identical pixel is meant the pixel that pixel value is identical with the pixel value of described interested pixel.
When the area of set b during greater than a setting range, during greater than 50 pixels * 50 pixels, described object and regional determination unit 102 also are used for from existing background model acquisition and the corresponding pixel of set b as set b, and obtain pixel thus and gather a.Wherein, described setting range can be determined voluntarily by the user, for example, when the user only wants the bigger object of volume detected, this setting range can be arranged to a bigger value, so that the follow-up object of comparatively being concerned about that filters out is monitored from image.
Described object and regional determination unit 102 also are used to utilize characteristic point algorithm pair set a and b enforcement computing respectively, find out the characteristic point in each pixel set and describe vector.In the present embodiment, (scale-invariant feature transform, SIFT) algorithm, SURF algorithm or other can be used for detecting and describing the algorithm of image locality feature to described characteristic point algorithm for yardstick invariant features conversion.Wherein, the characteristic point of utilizing the SIFT algorithm to be extracted is based on the point of interest of some local appearance on the object, with the size and the irrelevant to rotation of image.As the black dot among Fig. 8 (b2) is the characteristic point of finding out in set b, and the black dot among Fig. 8 (a2) is the characteristic point of finding out in set a.
Subsequently, object and regional determination unit 102 utilize seed region to increase algorithm will gather characteristic point among a and implement image cutting as seed and obtain block A, shown in Fig. 8 (a3), and will gather among the b with block A opposite position on characteristic point implement the image cutting as seed and obtain block B, shown in Fig. 8 (b3).
Article identification unit 104 be used to judge the area of block B be greater than or less than the area of block A, when the area of block B during less than the area of block A, enter thing in article identification unit 104 these guarded regions of judgement, and, remove thing in article identification unit 104 these guarded regions of judgement when the area of block B during greater than the area of block A.
This article identification unit 104 also is used for the thing of being judged that enters is carried out size, color and entry time filtration, and utilize general machine learning algorithm such as neural network (Neural Networks), SVMs (Support Vector Machine) etc., the characteristic point of each object of storage in the characteristic point that enters thing after filtering and description vector and the characteristic point data storehouse 3 is described vector model to be compared, enter thing to discern this, and judge whether the described thing that removes at the appointed time is removed in the section.
Wherein, described filtration is meant that specifically a plurality of objects that interested pixel is formed filter, make the requirement that size, the color of object of final decision and time of entering guarded region all meet the user, the time that the automobile size need be arranged, ignore the color of city taxi and enter guarded region as the object after filtering needed in the uncovered time period.
As described in Figure 4, be the operation process chart of object method for supervising of the present invention preferred embodiment.
Step S400, prospect object detecting unit 100 utilize the prospect object in the image that double-deck background model detecting watch-dog 2 caught, and specifically describe as described in Figure 5.This bilayer background model comprises existing background model and temporary background model.
If the above-mentioned prospect object that detects still is judged as the prospect object after more than or equal to a setting-up time interval, in step S402, object and regional determination unit 102 can be labeled as interested pixel with described pixel when the pixel of this prospect object is moved into temporary background model, the zone contiguous with described interested pixel searched in object and regional determination unit 102 from temporary background model, find out the pixel identical with the pixel value of described interested pixel, and it is judged to be interested pixel, obtain a pixel set b (as the pixel set of composition diagram 8 (b1)) thus, when the area of described pixel set b during greater than a setting range, object and regional determination unit 102 be acquisition and the corresponding pixel of described pixel set b from existing background model, and obtains pixel set a (as the pixel set of five-pointed star in the composition diagram 8 (a1)) thus.
Step S404, object and regional determination unit 102 utilize the characteristic point algorithm respectively pixel set a and b to be implemented computing, find out the characteristic point (as the black dot among Fig. 8 (a2), (b2)) in each pixel set and describe vector, utilize seed region to increase algorithm then and will gather characteristic point among a and implement the image cutting as seed and obtain block A (as the black part among Fig. 8 (a3)), and will gather among the b with block A opposite position on characteristic point cut as seed enforcement image and obtain block B (as the black part among Fig. 8 (b3)).
Step S406, article identification unit 104 judges that the area of this block B is greater than the area of block A or less than the area of block A.If the area of block B is the area greater than block A, then flow process enters step S408, if the area of block B is the area less than block A, then flow process enters step S414.In the present embodiment,, show that both not entered thing in the guarded region does not remove thing yet if the area of block B is the area that equals block A.
Step S408, article identification unit 104 judges in this guarded region have object to be removed, and promptly removes thing in this guarded region.
Step S410, article identification unit 104 judges that this removes thing and whether at the appointed time is removed in the section, if judged result is at the appointed time to be removed in the section for this removes thing, then flow process enters step S412.If judged result is not at the appointed time to be removed in the section for this removes thing, then process ends.
Step S412, article identification unit 104 sends this guarded region of alarm Security Officer threat, and flow process finishes then.
Step S414, article identification unit 104 judges in this guarded region have object to enter, and promptly enters thing in this guarded region.
Step S416,104 pairs of these size, color and the entry times that enter thing in article identification unit are filtered the thing that enters after this filtration of back identification, and flow process enters step S412 then.Particularly, described size, color and the entry time that enters thing (a plurality of) analyzed whether in the claimed range that the user is provided with in article identification unit 104, and the satisfactory thing that enters discerned, as the characteristic point that enters thing after machine learning algorithm such as neural network (neural networks) or SVMs (support vector machine) will filter utilizing and describe in vector and the characteristic point data storehouse 3 characteristic point of each object of storage and describe vector model and compare, which kind of object enters thing be to discern this.
As shown in Figure 5, be the particular flow sheet of prospect object detecting among Fig. 4 step S400.This flow process is that example describes with the prospect object detecting of certain two width of cloth image in the N width of cloth chromatic image only, and the prospect object detecting in other images is all carried out according to this method for detecting.
Step S500 sets an empty background model by model building module 1000, receives first width of cloth image in the N width of cloth chromatic image, that is to say that this sky background model is used to store first width of cloth image.In the present embodiment, the detecting of the prospect of the image after the 2nd width of cloth~N width of cloth and the N width of cloth need not to re-establish the sky background model again.
Step S502, successively with the width of cloth image in this N width of cloth image as current image, be to have background model now to detect the background model that a width of cloth image is generated before this image.
Step S504, pixel separation module 1002 in will this current image each pixel and the pixel in the existing background model compare, to determine the difference and the luminance difference of the pixel value between respective pixel.In the present embodiment, second width of cloth image is to be existing background model to deposit first width of cloth image in the sky background model in; After this second width of cloth image processing is intact, take out the 3rd width of cloth image again and handle, the 3rd width of cloth image is be to have background model now with the background model that is generated by detecting first width of cloth, second width of cloth image, by that analogy, up to all image processing are finished.For example, as shown in Figure 6, N width of cloth image is to have background model now to detect the obtained background model A0 of the 1st~the N-1 width of cloth image, and N+1 width of cloth image is existing background model with background model A.
Step S506, pixel separation module 1002 judges whether the difference and the luminance difference of above-mentioned definite pixel value all are less than or equal to predefined threshold value.
When if the difference of the pixel value between the respective pixel in described pixel and the existing background model and luminance difference all are less than or equal to predefined threshold value, in step S508, pixel separation module 1002 judges that this pixel is a background pixel, memory module 1004 adds this pixel in the existing background model, thereby generated new background model, enter step S518 then, wherein, the object present embodiment of being made up of all background pixels is referred to as the background object.For example, suppose that guarded region does not have external object (as people or car) and involves in, only light has slight variation, and can not cause the existing background model of pixel in the current image to have changing too greatly by the light of this variation the time, pixel separation module 1002 still can continue the pixel in the current image is judged to be background pixel, and memory module 1004 adds the new background model of generation in the existing background model with this pixel.
Otherwise, if the difference of the pixel value between the respective pixel in described pixel and the existing background model and luminance difference are all greater than described predefined threshold value, in step S510, pixel separation module 1002 judges that this pixel is a foreground pixel, and the object present embodiment of being made up of all foreground pixels is referred to as the prospect object.As shown in Figure 6 and Figure 7, if the background model of being made up of above-mentioned the 1st~the N-1 width of cloth chromatic image is A0, this background model A0 is made up of the tree that stops in the guarded region, road, in N width of cloth image, if there is vehicle to enter guarded region, then the pixel of the detecting process decidable composition vehicle of process step S506 is the prospect object.
Step S512, memory module 1004 is kept in the pixel and the existing background model of the prospect object among the step S510, obtains described temporary background model B.
Step S514, whether pixel value and brightness value that temporary background model monitoring module 1006 is monitored the pixel among the described temporary background model B in real time change within a preset time interval.If the pixel value and the brightness value of the pixel in this Preset Time interval among the described temporary background model B change, temporary background model after supposing to change is B`, then temporary background model monitoring module 1006 repeated execution of steps S514 judge whether this temporary background model B` changes in the default time interval.Otherwise if the pixel value and the brightness value of the pixel among the described temporary background model B (or temporary background model B`) do not change in this Preset Time interval, then flow process enters step S516.
Step S516, background model update module 1008 is upgraded described existing background model with described temporary background model B or B`, thereby generated new background model, for example, as shown in Figure 7, background model update module 1008 is upgraded described existing background model with temporary background model B and is obtained new background model (as background model A).At the image after the described N width of cloth, as the N+1 width of cloth image among Fig. 6, pixel separation module 1002 detect prospect object and this prospect object by temporary behind temporary background model B`, if monitoring described temporary background model B`, temporary background model monitoring module 1006 in described Preset Time interval, do not change, then background model update module 1008 can be upgraded described background model A with this temporary background model B` and obtains background model A`, by that analogy, background model can constantly obtain upgrading, the method of this background real-time update can avoid image to rock, light changes, the interference of Periodic Object, detect the prospect object in the image more accurately, to reach to purposes such as guarded region effective monitorings.In addition, utilize this method also the object that stays for some time can be considered as background automatically in guarded region.
Step S518, the chromatic image that pixel separation module 1002 receives by checking judges whether that image is not detected in addition, that is to say that pixel separation module 1002 judges whether that the prospect object of chromatic image separates with the pixel of background object correspondence in addition.If judged result is for denying then direct process ends.If it is current image with the image of not detecting that judged result, is then returned step S504 for being, be to have background model now with the background model of detecting before this image that image was generated, execution in step S504 is to step S516 successively.
It should be noted that at last, above embodiment is only unrestricted in order to technical scheme of the present invention to be described, although the present invention is had been described in detail with reference to preferred embodiment, those of ordinary skill in the art is to be understood that, can make amendment or be equal to replacement technical scheme of the present invention, and not break away from the spirit and scope of technical solution of the present invention.